The top considerations that come into play when I advise someone whether to earn-to-give or work directly on x-risk look like this:
1) Does this person have a large comparative advantage at the direct problem domain? Top-rank math talent can probably do better at MIRI than at a hedge fund, since there are many mathematical talents competing to go into hedge funds and no guarantee of a good job, and the talent we need for inventing new basic math does not translate directly into writing the best QT machine learning programs the fastest.
2) Is this person going to be able to stay motivated if they go off on their own to earn-to-give, without staying plugged into the community? Alternatively, if the person’s possible advantage is at a task that requires a lot of self-direction, will they be able to stay on track without requiring constant labor to keep them on track, since that kind of independent job is much harder to stick at then a 9-to-5 office job with supervision and feedback and cash bonuses?
Every full-time employee at a nonprofit requires at least 10 unusually generous donors or 1 exceptionally generous donor to pay their salary. For any particular person wondering how they should help this implies a strong prior bias toward earning-to-give. There are others competing to have the best advantage for the nonprofit’s exact task, and also there are thousands of job opportunities out there that are competing to be the maximally-earning use of your exact talents—best-fits to direct-task-labor vs. earning-to-give should logically be rare, and they are.
The next-largest issue is motivation, and here again there are two sides to the story. The law student who goes in wanting to be an environmentalist (sigh) and comes out of law school accepting the internship with the highest-paying firm is a common anecdote, though now that I come to write it down, I don’t particularly know of any gathered data. Earning to give can impose improbability in the form of likelihood that the person will actually give. Conversely, a lot of the most important work at the most efficient altruistic organizations is work that requires self-direction, which is also demanding of motivation.
I should pause here to remark that if you constrain yourself to ‘straightforward’ altruistic efforts in which the work done is clearly understandable and repeatable and everyone agrees on how wonderful it is, you will of course be constraining yourself very far away from the most efficient altruism—just like a grant committee that only wants to fund scientific research with a 100% chance of paying off in publications and prestige, or a VC that only wanted to fund companies that were certain to be defensible-appearing decisions, or someone who constrained their investments to assets that had almost no risk of going down. You will end up doing things that are nearly certain never to appear to future historians as a decisive factor in the history of Earth-originating intelligent life; this requires tolerance for not just risk but scary ambiguity. But if you want to work on things that might actually be decisive, you will end up in mostly uncharted territory doing highly self-directed work, and many people cannot do this. Just as many other people cannot sustain altruism without being surrounded by other altruists, but this can possibly be purchased elsewhere via living on the West or East Coast and hanging around with others who are earning-to-give or working directly.
These are the top considerations when someone asks me whether they should work directly or earn to support others working directly—the low prior, whether the exact fit of talent is great enough to overcome that prior, and whether the person can sustain motivation / self-direct.
My main comment on this is that if self-direction is as important as it appears to be, it would seem to me that ‘become self directed’ really should be everyone’s first priority if they can think of any way to do that.
My second comment is that it seems to me that if one is self-directed and seeks appropriate mentorship, the expected value of pursuing a conventional career is very low compared to that of pursuing an entrepreneurial career. Conversely, mentorship or advice that doesn’t account for the critical factor of how self-directed someone is, as well as a few other critical factors such at the disposition to explore options, respond to empirical feedback from the market, etc, is likely to be worse than useless.
My second comment is that it seems to me that if one is self-directed and seeks appropriate mentorship, the expected value of pursuing a conventional career is very low compared to that of pursuing an entrepreneurial career.
Can you expand on this? How does one seek appropriate mentorship?
Every full-time employee at a nonprofit requires at least 10 unusually generous donors or 1 exceptionally generous donor to pay their salary.
Isn’t $36K/yr the modal MIRI salary? That doesn’t feel like it should be too hard on a $100K/yr software developer salary considering that charitable donations are tax-deductible up to 50% of your income (supposedly). If one donated $36K/yr out of their software developer salary, at $64K/yr they’d still be earning much more than a typical nonprofit employee (heck, much more than a typical college graduate), and if they were to pretend they were working at a nonprofit and subsist on $36K/yr themselves, they could probably subsidize 2 people at the modal MIRI salary (after several years’ worth of promotions/raises).
$36k/yr salary works out to be about $50k-$70k gross expense (including benefits, insurance, taxes etc) for a regular employer, not sure how much it is for a non-profit like MIRI.
10% is the charitable giving limit. There is another thing to be asked about, and that is the impact of the job. If I were to be a tax lawyer, I would be directly harming the ability of the US government to spend on social welfare programs. If I worked on Wall Street anywhere but Vanguard I would be bilking people out of their life savings, and at Vanguard I wouldn’t be making $100 K a year. Someone working as a tobacco farmer to raise money for cancer research has some misplaced priorities.
I understand why you said this, but most people interested in this are interested in the transition from 10% to >10% (say, 20), not in 10% to 50%. I presume you would estimate a higher number for whom this is healthy?
I were to be a tax lawyer, I would be directly harming the ability of the US government to spend on social welfare programs.
You could always go work for the IRS. It employs a lot of tax lawyers.
But there’s a bigger issue: you think that the work of a (privately employed) tax lawyer intrinsically harms the ability of the US government to spend? That belief has LOTS of issues. I’ll start with two: One, why do you think the capability of the US government to spend is an unalloyed good thing? And two, do you happen to know the volume (say, in feet of shelf space) of the current tax laws, regulations, and rulings? I’d recommend you find out and then think about whether any moderately complicated business can comply with them without the help of a tax lawyer.
If I worked on Wall Street anywhere but Vanguard I would be bilking people out of their life savings
Sigh. First, Vanguard is not part of Wall Street. Second… you really should not believe everything the popular media keeps feeding you.
By “Wall Street” I’m including the Buy Side as well as the Sell Side. The big buyside firms like Fidelity and Charles Schwab sell products that most people shouldn’t buy. Insurance probably has a better case to buy some actively managed products, or some exotic derivatives, but I don’t know why it can’t do it itself.
To the extent that finance reallocates risk it can provide a positive utility benefit. However, the very productive businesses have questionable utility. Promoting active trading, picking hot funds etc, all eat into the returns clients can expect. Justify the existence of Charles Schwab’s S&P 500 index fund, with expense ratio twice that of Vanguards. The most profitable divisions of investment banks tend to be the ones with the least competition, and hence most questionable social benefit.
I’m aware Dodge and Cox is in SF, and Vanguard in Valley Forge, Blackrock in Princeton, etc. However, they are all on “the Street”.
The IRS doesn’t pay well: for government pay one might as well work for NASA and accomplish something fun.
Tax lawyers can not decrease taxes taken by US government in the long run, because US government gets to make the law adjusting for the existence of tax lawyering. This is why I have absolutely no qualms about employing a tax lawyer in the US.
Not in the U.S. (note these are in pre-tax earnings, so they translate into less in foregone consumption than they do in donations made).
There are limits to how much you can deduct, but they’re very high.
For most people, the limits on charitable contributions don’t apply. Only if you contribute more than 20% of your adjusted gross income to charity is it necessary to be concerned about donation limits. If the contribution is made to a public charity, the deduction is limited to 50% of your contribution base. For example, if you have an adjusted gross income of $100,000, your deduction limit for that year is $50,000.
Regarding this:
If I worked on Wall Street anywhere but Vanguard I would be bilking people out of their life savings, and at Vanguard I wouldn’t be making $100 K a year. Someone working as a tobacco farmer to raise money for cancer research has some misplaced priorities.
Goldman has 32,000 employees. An upper bound for the harm caused by the marginal employee is thus the total harm caused divided by 32,000. For the harm to outweigh the good, Goldman would therefore have to be killing at least 3.2 million young people each year, or doing something else that is similarly harmful. That would mean that Goldman Sachs would need to be responsible for around 5% of all deaths in the world. Bear in mind that Goldman Sachs only makes up 22% of American investment banking, and 3% of the American financial industry—if the rest of finance is similarly bad, then it would imply that finance is doing something as bad as causing all the deaths in the world..
Let’s consider the American financial industry in general. Upcoming Giving What We Can research estimates that it would take $200 billion a year to move everyone in the world above the $1.25 poverty line. That figure will only be $74 billion in 2030. The employees of the financial sector could do this if they transferred (e.g. via GiveDirectly) 30-75% of their salaries to those in extreme global poverty (depending on what date you want to achieve the goal by). In other words, if everyone in finance were Earning to Give, it would be possible to end extreme global poverty within the next twenty years. Harm would only dominate if the financial sector is doing something roughly as bad as single handedly causing all global poverty.
THIS. Although I`m unsure about the particulars you mention here, being an European, people and effective altruists need to realize that your job is INSIDE the world you live in. Estimating how much good you’re producing is not just about how much money/time you’re giving to effective charities, but also how much your way of life is helping/damaging the world.
I’m not convinced. The amount of saved lives, QALYs, or whatever you are counting that the US government welfare program gets per dollar is (or seems to be to me) quite a bit less than the amount that, say, the AMF could get with that money. I don’t know how many dollars per QALY US government welfare manages to get, but I wouldn’t be surprised if it were on the order of $1000-$10000 per QALY. And that’s not even counting the fact that even if the US goverment had that bit more money from you not being a tax lawyer, that money would not all go to welfare and other such efficient (relative to what else the government spends money on) projects. I would imagine a fair portion would go to, say, bombing Syria, or hiring an extra parking-meter enforcer, or such inefficent stuff, that get an even worse $/QALY result.
And that is still not to mention the fact that some of that money would go to, say, funding the NSA to spy on your phone calls and read your email, or to the TSA to harass, strip-search, and detain you, which are net negatives.
And even that is not counting that MIRI may end up having a QALY/$ result far, far higher than anything the AMF or whoever could ever hope of possibly getting.
I’m not saying you’re flat-out wrong, and it is something to take into consideration when figuring out the altruistic impact of your job, but taking into account these objections, it seems highly unlikely that the marginal dollar from the government goes far enough to weigh very heavily in ones analysis.
On the topic of how much it takes to save a QALY in the US:
“Most, but not all, decision makers in the United States will conclude that interventions that cost less than $50,000 to $60,000 per QALY gained are reasonably efficient. An example is screening for hypertension, which costs $27,519 per life-year gained in 40-year-old men.3, 8 For interventions that cost $60,000 to approximately $175,000 per QALY, certain decision makers may find the interventions sufficiently efficient; most others will not agree.”
The first paragraph of this gives more on the cost of QALYs in the US. So, kidney dialysis is an intervention that is paid for by the government in the US, and it comes in at more than $100,000 per QALY saved.
Since marginal funding generally goes to pay for interventions which are no more effective than those already being paid for, I wouldn’t expect the cost of a marginal QALY to be below (say) $50,000.
I’m not sure if you were answering my comment or wubbles’s one. What I was saying was that you need to take into account the negative impact your job and way of life have on the world.
I agree that the US government probably is terrible at using tax money to better the world.
If I worked on Wall Street … I would be bilking people out of their life savings
Do you actually think that a finance professional who donated a significant portion of their income to effective charity would be doing more harm than good? Even given that you can save the life of a child in the developing world for on the order of $2000?
I don’t think the problem is finance professionals in general—it’s finance professionals in particularly corrupt parts of the industry.
Figuring out in advance that a job is doing particularly corrupt work seems to be something that people are very bad at—I don’t know whether it’s mostly that it would be hard for a neutral observer, or that people don’t want the problems of dealing with the consequences to their own lives if they find that their job is destructive.
Hmm, I was thinking the assumption (which I don’t necessarily entirely agree with) was that finance professionals were simply earning money without providing any benefit to society, and so a net negative. It sounds like your comment assumes that some of them are actually actively doing harm (though perhaps unintentionally), beyond just taking their own paycheck’s worth out of the productive economy. Is that your understanding?
The mortgage crisis was a result of banks being able to sell mortgages to other banks. This meant that the bank making the loan could make money just by the mortgage being initiated—the first bank no longer had a strong interest in the loan being repaid.
There were some other pieces to the situation that I don’t have clear in my mind at the moment, but I think there were incentives for the mortgage to actually not be repaid and the house to be taken by a bank.
One piece that I am clear on is that there were people who decided it wasn’t worth it for banks to keep accurate track of who owned which mortgage, or what had been paid, or what had been agreed to.
This is stealing people’s houses. It’s a degree of damage which it’s hard to imagine being covered by charity.
The other side of the story is that not every bank behaved like that—not all of finance is fraudulent.
Its at least somewhat true, if perhaps not well stated- packaged mortgages and derivatives based on packaged mortgages (mortgages sold as investment vehicles to other banks and funds) played a very large role in the crisis.
Without “selling mortgages to other banks” the popping of the housing bubble wouldn’t have turned into the liquidity crunch that started in 2008.
So were mortgages by themselves. Without the widespread availability of mortgages “the popping of the housing bubble wouldn’t have turned into the liquidity crunch” too. Or, for that matter, without the fact that the “standard” mortgage is a 30-year fixed—not, say, a 1⁄1 ARM.
But anyway, the reason for the contagion from mortgages to liquidity wasn’t the ability to sell mortgages. It was the mispricing of mortgage derivatives, specifically the widespread belief that certain tranches of collateralized mortgage obligations (CMOs) were effectively risk-free.
If you want to dig deeper, the real cause was the global asset bubble helped by the too-loose monetary policy in the mid-2000s.
Financial economics are complicated. Snap judgements from popular press rarely have much relationship to reality.
So were mortgages by themselves. Without the widespread availability of mortgages “the popping of the housing bubble wouldn’t have turned into the liquidity crunch” too.
Well, sure. And also without houses themselves…
My point is that the statement wasn’t false on its face. Repackaging and reselling was A proximate cause of the liquidity crunch, its not the only cause but its a part of what happened.
This is stealing people’s houses. It’s a degree of damage which it’s hard to imagine being covered by charity.
Is it really? I suppose this depends on how many houses any individual is responsible for and how much money they capture per house. I guess that second part is the real issue—any individual who would be giving to charity probably only captures a fraction of what they earn for their firm.
But if you could capture the whole value of a predatory mortgage and convert it into developing world lives saved, it’s not hard to imagine the numbers adding up. (One American family goes bankrupt and 20 Malawian children who otherwise would have don’t die in childhood? On the face of it that looks like a pretty positive net outcome.)
If you can do outsized damage significantly beyond what you can capture as income though, then I suppose it gets a bit tougher to justify.
(One American family goes bankrupt and 20 Malawian children who otherwise would have don’t die in childhood? On the face of it that looks like a pretty positive net outcome.)
If we’re talking about donations on the scale of the activities that went into the mortgage crisis, I think you’d start to suffer seriously diminishing returns.
Even if you didn’t, there are other problems you’d run into, such as the limited ability of the Malawian (or other impoverished African) society and economy to accommodate such a sudden spike in children surviving to adulthood. The lives that you save from extermination at the hands of malaria or other preventable causes are probably mostly going to be relatively lousy or short due to other causes, pending much further investment.
As I understood it, the hypothetical was a single individual deciding to work in finance and donate a large portion of their income to efficient charity. In that case I don’t think the diminishing returns are so much of an issue.
I would worry more about negative flow-through effects of a decline in trust and basic decency in society. I think those are much more clear than flow-through effects of positive giving. I’m not sure if this outweighs the 20-to-1 ratio.
If I were to be a tax lawyer, I would be directly harming the ability of the US government to spend on social welfare programs.
Government social welfare spending is notoriously inefficient. So if your client is at all generous with his money you’re coming out ahead. Heck even if he doesn’t give to charity but does use the money to invest in productive enterprises, you’re probably coming out ahead. And that’s before taking into account how you spend your money.
The cost (to an organization) of an employee is more than just his salary, often considerably more. There is health insurance and other benefits, payroll taxes, infrastructure support (e.g. a computer, a desk to put it on, a room to put the desk in), etc.
I know the comment was probably fine, but overall it seemed like it could be read as unnecessarily snarky and hence lower the tone—PMing you the correction would have been a better move.
All in all I think that the standard for discussion here on LessWrong could be increased a lot if people stopped giving “wiseass” replies to things, were more forgiving of minor errors (while still pointing them out), and so on.
Every full-time employee at a nonprofit requires at least 10 unusually generous donors or 1 exceptionally generous donor to pay their salary.
If you define “generous” by “amount of capital” then this is tautologically true. But by this standard, extraordinarily wealthy people are capable of being exceptionally exceptionally exceptionally generous. I’d recur to my remark about the Giving Pledge. I believe that the projects of highest humanitarian value will generally get funded.
I should pause here to remark [...] but this can possibly be purchased elsewhere via living on the West or East Coast and hanging around with others who are earning-to-give or working directly.
In principle this could fall under the “unusual values” consideration that I raise above. But I don’t think that the sociological phenomenon that you seem to be implying to exist prevails in practice. I think that there a lot of funders who are not risk-averse, and indeed, many who are actively attracted to high risk projects.
Well, if James Simons wanted to retire from Renaissance and work on FAI full-time, it would not be entirely obvious to me that this was a bad move, but only if Simons had enough in the bank to also pay as much other top-flight math talent as could reasonably be used, and was already so paying, such that there was no marginal return to his further earning power relative to existing funds.
I think that James Simons is an example of someone with an unusually strong comparative advantage at making money. But this wouldn’t necessarily have been clear a priori: if you put yourself in Simons’ shoes in 1980 the expected earnings of going into finance would be much lower than his actual earnings turned out to be. So it’s not clear that he would have done better to “earn to give” than doing something of direct humanitarian value (though maybe it was clear from the outset that his comparative advantage was in finance.)
I’ll also highight another point implicit in my post: even if one assumes that there’s not enough funding in the nonprofit world for the projects of highest value, there may be such funding available in other contexts (for-profit, academic and government). This makes the argument for earning to give weaker.
I recognize that I haven’t addressed the specific subject of Friendly AI research, and will do so in future posts.
I think that there’s substantial overlap between my values and MIRI staff’s values, and that the difference regarding the relative value of “earning to give” is epistemic rather than normative. But obviously there’s a great deal more that needs to be said about the epistemic side, with reference to the concrete example of Friendly AI.
I can imagine someone thinking that FHI was a better use of money than MIRI, or CFAR, or CSER, or the Foresight Institute, or brain-scanning neuroscience, or rapid-response vaccines, or any number of startups, but considering AMF as being in the running at all seems to require either a value difference or really really different epistemics about what affects the fate of future galaxies.
Realistic amounts of difference in epistemics + the “humans best stick to the mainline probability” heuristic seem enough (where by “realistic” I mean “of the degree actually found in the world”). I.e., I honestly believe that there are many people out there who would care the hell about the fate of future galaxies if they alieved that they had any non-vanishing chance of significantly influencing that fate (and to choose the intervention that influences it in the desired direction).
If you’re one of 10^11 sentients to be born on Ancient Earth with a golden opportunity to influence a roughly 10^80-sized future, what exactly is a ‘vanishing chance’… eh, let’s all save it until later.
I meant that the alieved probability is small in absolute terms, not that it is small compared to the payoff. That’s why I mentioned the “stick to the mainline probability” heuristic. I really do believe that there are many people who, if they alieved that they (or a group effort they could join) could change the probability of a 10^80-sized future by 10%, would really care; but who do not alieve that the probability is large enough to even register, as a probability; and whose brains will not attempt to multiply a not-even-registering probability with a humongous payoff. (By “alieving a probability” I simply mean processing the scenario the way one’s brain processes things it assigns that amount of credence, not a conscious statement about percentages.)
This is meant as a statement about people’s actual reasoning processes, not about what would be reasonable (though I did think that you didn’t feel that multiplying a very small success probability with a very large payoff was a good reason to donate to MIRI; in any case seems to me that the more important unreasonableness is requesting mountains of evidence before alieving a non-vanishing probability for weird-sounding things).
[ETA: I find it hard to put a number on the not-even-registering probability the sort of person I have in mind might actually alieve, but I think a fair comparison is, say, the “LHC will create black holes” thing—I think people will tend to process both in a similar way, and this does not mean that they would shrug it off if somebody counterfactually actually did drop a mountain of evidence about either possibility on their head.]
though I did think that you didn’t feel that multiplying a very small success probability with a very large payoff was a good reason to donate to MIRI
Because on a planet like this one, there ought to be some medium-probable way for you and a cohort of like-minded people to do something about x-risk, and if a particular path seems low probability, you should look for one that’s at least medium-probability instead.
So do you believe in the Simulation Hypothesis or the Doomsday Argument, then? All attempts to cash out that refusal-to-believe end in one or the other, inevitably.
From where I stand, it’s more like arcane meta-arguments about probability are motivating a refusal-to-doubt the assumptions of a prized scenario.
Yes, I am apriori skeptical of anything which says I am that special. I know there are weird counterarguments (SIA) and I never got to the bottom of that debate. But meta issues aside, why should the “10^80 scenario” be the rational default estimation of Earth’s significance in the universe?
The 10^80 scenario assumes that it’s physically possible to conquer the universe and that nothing would try to stop such a conquest, both enormous assumptions… astronomically naive and optimistic, about the cosmic prospects that await an Earth which doesn’t destroy itself.
Okay, so that’s the Doomsday Argument then: Since being able to conquer the universe implies we’re 10^70 special, we must not be able to conquer the universe.
Calling the converse of this an arcane meta-argument about probability hardly seems fair. You can make a case for Doomsday but it’s not non-arcane.
Perhaps this is hairsplitting but the principle I am employing is not arcane: it is that I should doubt theories which imply astronomically improbable things. The only unusual step is to realize that theories with vast future populations have such an implication.
I am unable to state what the SIA counterargument is.
In the theory that there are astronomically large numbers of people, it is a certainty that some of them came first. The probability that YOU are one of those people equal to the probability that YOU are any one of those other people. However, it does define a certain small narrow equivalence class that you happen to be a member of.
It’s a bit like the difference between theorizing that: A) given that you bought a ticket, you’ll win the lottery, and B) given that the lottery folks gave you a large sum, that you had the winning ticket.
That’s not the “SIA counterargument”, which is what I want to hear (in a compact form, that makes it sound straightforward). You’re just saying “accept the evidence that something ultra-improbable happened to you, because it had to happen to someone”.
But it’s still a simple idea once you grasp it. I was hoping you could state the counterargument with comparable simplicity. What is the counterargument at the level of principles, which neutralizes this one?
I largely agree with your skepticism. I would go even farther and say that even the 10^80 scenario happens, what we do now can only influence it by random chance, because the uncertainty in the calculations of the consequences of our actions in the near term on the far future overwhelms the calculations themselves. That said, we should still do what we think is best in the near term (defined by our estimates of the uncertainty being reasonably small), just not invoke the 10^80 leverage argument. This can probably be formalized, by assuming that the prediction error grows exponentially with some relevant parameter, like time or the number of choices investigated, and calculating the exponent from historical data.
Doomsday for me, I think. Especially when you consider that it doesn’t mean doomsday is literally imminent, just “imminent” relative to the kind of timescale that would be expected to create populations on the order of 10^80.
In other words, it fits with the default human assumption that civilization will basically continue as it is for another few centuries or millennia before being wiped out by some great catastrophe.
Do you mind elaborating on this inevitability? It seems like there ought to be other assumptions involved. For example, I can easily imagine that humans will never be able to colonize even this one galaxy, or even any solar system other than this one. Or that they will artificially limit the number of individuals. Or maybe the only consistent CEV is that of a single superintelligence of which human minds will be tiny parts. All of these result in the rather small total number of individuals existing at any point in time.
For example, I can easily imagine that humans will never be able to colonize even this one galaxy, or even any solar system other than this one.
Counts as Doomsday, also doesn’t work because this solar system could support vast numbers of uploads for vast amounts of time (by comparison to previous population).
Or that they will artificially limit the number of individuals.
This is a potential reply to both Doomsday and SA but only if you think that ‘random individual’ has more force than a similar argument from ‘random observer-moment’, i.e. to the second you reply, “What do you mean, why am I near the beginning of a billion-year life rather than the middle? Anyone would think that near the beginning!” (And then you have to not translate that argument back into a beginning-civilization saying the same thing.)
Or maybe the only consistent CEV is that of a single superintelligence of which human minds will be tiny parts.
...whereupon we wonder something about total ‘experience mass’, and, if that argument doesn’t go through, why the original Doomsday Argument / SH should either.
Thanks, I’ll chew on that a bit. I don’t understand the argument in the second and third paragraphs. Also, it’s not clear to me whether by “counts as doomsday” you mean the standard doomsday with the probability estimates attached, or some generalized doomsday, with no clear timeline or total number of people estimated.
Anyway, the feeling I get from your reply is that I’m missing some basic background stuff here I need to go through first, not the usual “this guy is talking out of his ass” impression when someone invokes anthropics in an argument.
No, this is talking-out-of-our-ass anthropics, it’s just that the anthropic part comes in when you start arguing “No, you can’t really be in a position of that much influence”, not when you’re shrugging “Sure, why shouldn’t you have that much influence?” Like, if you’re not arriving at your probability estimate for “Humans will never leave the solar system” just by looking at the costs of interstellar travel, and are factoring in how unique we’d have to be, this is where the talking-out-of-our-ass anthropics comes in.
Though it should be clearly stated that, as always, “We don’t need to talk out of our ass!” is also talking out of your ass, and not necessarily a nicer ass.
it’s just that the anthropic part comes in when you start arguing “No, you can’t really be in a position of that much influence”, not when you’re shrugging “Sure, why shouldn’t you have that much influence?”
Or when you (the generic you) start arguing “Yes, I am indeed in a position of that much influence”, as opposed to “There is an unknown chance of me being in such a position, which I cannot give a ballpark estimate for without talking out of my ass, so I won’t”?
Huh. I don’t understand how refusing to speculate about anthropics counts as anthropics. I guess that’s what you meant by
Though it should be clearly stated that, as always, “We don’t need to talk out of our ass!” is also talking out of your ass, and not necessarily a nicer ass.
I wonder if your definition of anthropics matches mine. I assume that any statement of the sort
All other things equal, an observer should reason as if they are randomly selected from the set of
is anthropics. I do not see how refusing to reason based on some arbitrary set of observers counts as anthropics.
Right. So if you just take everything at face value—the observed laws of physics, the situation we seem to find ourselves in, our default causal model of civilization—and say, “Hm, looks like we’re collectively in a position to influence the future of the galaxy,” that’s non-anthropics. If you reply “But that’s super improbable a priori!” that’s anthropics. If you counter-reply “I don’t believe in all this anthropic stuff!” that’s also an implicit theory of anthropics. If you treat the possibility as more “unknown” than it would be otherwise, that’s anthropics.
OK, I think I understand your point now. I still feel uneasy about the projection like your influencing 10^80 people in some far future, mainly because I think it does not account for the unknown unknowns and so is lost in the noise and ought to be ignored, but I don’t have a calculation to back up this uneasiness at the moment.
if you think that ‘random individual’ has more force than a similar argument from ‘random observer-moment’
I’ve had a vague idea as to why the random observer-moment argument might not be as strong as the random individual one, though I’m not very confident it makes much sense. (But neither argument sounds anywhere near obviously wrong to me.)
I wonder if this argument can be made precise enough to have its premises and all the intermediate assumptions examined. I remain skeptical of any forecast that far into the future. You presumably mean your confidence in the UFAI x-risk within the next 20-100 years as the minimum hurdle to overcome, with the eternal FAI paradise to follow.
My reason for mentioning AMF and global health is that doing so provides a concrete, pretty robustly researched example, rather than as to compare with efforts to improve the far future of humanity.
I think that working in global health in a reflective and goal directed way is probably better for improving global health than “earning to give” to AMF. Similarly, I think that working directly on things that bear on the long term future of humanity is probably a better way of improving the far future of humanity than “earning to give” to efforts along these lines.
I’ll discuss particular opportunities to impact the far future of humanity later on.
My reason for mentioning AMF and global health is that doing so provides a concrete, pretty robustly researched example
That depends on what you want to know, doesn’t it? As far as I know the impact of AMF on x-risk, astronomical waste, and total utilons integrated over the future of the galaxies, is very poorly researched and not at all concrete. Perhaps some other fact about AMF is concrete and robustly researched, but is it the fact I need for my decision-making?
(Yes, let’s talk about this later on. I’m sorry to be bothersome but talking about AMF in the same breath as x-risk just seems really odd. The key issues are going to be very different when you’re trying to do something so near-term, established, without scary ambiguity, etc. as AMF.)
I’m somewhat confused by the direction that this discussion has taken. I might be missing something, but I believe that the points related to AMF that I’ve made are:
GiveWell’s explicit cost-effectiveness estimate for AMF is much higher than the cost per DALY saved implied by the figure that MacAskill cited.
GiveWell’s explicit estimates for the cost-effectiveness of the best giving opportunities in the field of direct global health interventions have steadily gotten lower, and by conservation of expected evidence, one can expect this trend to continue.
The degree of regression to the mean observed in practice suggests that there’s less variance amongst the cost-effectiveness of giving opportunities than may initially appear to be the case.
By choosing an altruistic career path, one can cut down on the number of small probability failure modes associated with what you do.
I don’t remember mentioning AMF and x-risk reduction together at all. I recognize that it’s in principle possible that the “earning to give” route is better for x-risk reduction than it is for improving global health, but I believe the analogy between the two domains is sufficiently strong that my remarks on AMF have relevance (on a meta-level, not on an object level).
Yeah, I also have the feeling that I’m questioning you improperly in some fashion. I’m mostly driven by a sense that AMF is very disanalogous to the choices that face somebody trying to optimize x-risk charity (or rather total utilons over all future time, but x-risk seems to be the word we use for that nowadays). It seems though that we’re trying to have a discussion in an ad-hoc fashion that should be tabled and delayed for explicit discussion in a future post, as you say.
If I may list some differences I perceive between AMF and MIRI:
AMF’s impact is quite certain. MIRI’s impact feels more like a long shot —or even a pipe dream.
AMF’s impact is sizeable. MIRI’s potential impact is astronomic.
AMF’s impact is immediate. MIRI’s impact is long term only.
AMF’s have photos of children. MIRI have science fiction.
In mainstream circles, donating to AMF gives you pats in the back, while donating to MIRI gives you funny looks.
Near mode thinking will most likely direct one to AMF. MIRI probably requires one to shut up and multiply. Which is probably why I’m currently giving a little money to Greenpeace, despite being increasingly certain that it’s far, far from the best choice.
AMF’s impact is very likely to be net positive for the world under all reasonable hypotheses.
MIRI appears to me to have a chance to be massively net negative for humanity. I.e. if AI of the level they predict is actually possible, MIRI might end up creating or assisting in the creation of UFAI that would not otherwise be created, or perhaps not created as soon.
But what if AMF saves a child who grows up to be a biotechnologist and goes on to weaponize malaria and spread it to millions?
If you try hard enough, you can tell a story where any effort to accomplish X somehow turns out to accomplish ~X, but one must distinguish possibility from the balance of probability.
Yes, and the story where the child who grows up to be a biotechnologist and goes on to weaponize malaria and spread it to millions doesn’t pass the balance of probability test. The story that MIRI creates a dangerous AI fails to pass the balance of probability test only to the extent that one believes it is improbable that anyone can create such an AI. I do indeed consider it far more likely than not that there will never be the all-powerful AI you fear. And by that standard donations to MIRI are simply ineffective compared to donations to AMF.
However if I’m wrong about that and powerful FOOMing UFAIs are in fact possible, then I need to consider whether MIRI’s work is wise. If AIs do FOOM, there seems to me to be a very real possibility that MIRI’s work will either create a UFAI while trying to create a FAI, or alternatively enable others to do so. I’m not sure that’s more likely than that MIRI will one day create a FAI, but you can’t just multiply by the value of a very positive and very speculative outcome without including the possibility of a very negative and very speculative outcome.
The story that MIRI creates a dangerous AI fails to pass the balance of probability test only to the extent that one believes it is improbable that anyone can create such an AI.
[...]
However if I’m wrong about that and powerful FOOMing UFAIs are in fact possible, then I need to consider whether MIRI’s work is wise. If AIs do FOOM, there seems to me to be a very real possibility that MIRI’s work will either create a UFAI while trying to create a FAI, or alternatively enable others to do so.
If you increase the probability of uFAI in order for MIRI to kill everyone, the probability of someone else doing it goes up even more.
Maybe. I’m not sure about that though. MIRI is the only person or organization I’m aware of that seems to want to create a world controlling AI; and it’s the world-controlling part that I find especially dangerous. That could send MIRI’s AI in directions others won’t go. Are there other organizations attempting to develop AIs to control the world? Is anyone else trying to build a benevolent dictator?
The Machine Intelligence Research Institute exists to ensure that the creation of smarter-than-human intelligence benefits society.
They are well aware of the dangers of creating a uFAI, and you can be certain they will be real careful before they push a button that have the slightest chance of launching the ultimate ending (good or bad). Even then, they may very well decide that “being real careful” is not enough.
Are there other organizations attempting to develop AIs to control the world?
Anthropomorphic ideas of a “robot rebellion,” in which AIs spontaneously develop primate-like resentments of low tribal status, are the stuff of science fiction. The more plausible danger stems not from malice, but from the fact that human survival requires scarce resources: resources for which AIs may have other uses.
Many AIs will converge toward being optimizing systems, in the sense that, after self-modification, they will act to maximize some goal. For instance, AIs developed under evolutionary pressures would be selected for values that maximized reproductive fitness, and would prefer to allocate resources to reproduction rather than supporting humans.
Are there other organizations attempting to develop AIs to control the world? Is anyone else trying to build a benevolent dictator?
Is MIRI attempting to develop any sort of AI? I understood the current focus of its research to be the logic of Friendly AGI, i.e. given the ability to create a superintelligent entity, how do you build one that we would like to have created? This need not involve working on developing one.
AMF’s impact is very likely to be net positive for the world under all reasonable hypotheses.
That seems like a bizarre belief to hold. Or perhaps just overwhelmingly shortsighted. There are certainly reasonable hypotheses in which more people alive right now result in worse outcomes a single generation down the line, without even considering extinction level threats and opportunities. The world isn’t nearly easy enough to model and optimize for us to be that certain a disruptive influence on that scale will be a net positive under all reasonable hypotheses.
Would you care to cite any such reasonable hypotheses? I.e. under what assumptions do you think that saving a random poor person’s life is likely to be a net negative? Sum over the number of lives saved and even if one person grows up to be a serial killer, the total is still way positive. Can you really defend a situation in which it is preferable to have living people today die from malaria?
The problem with MIRI-hypothesized AI (beyond its implausibility) is that we don’t get to sum over all possible results. We get one result. Even if the chance of a good result is 80%, the chance of a disastrous result is still way too high for comfort.
Would you care to cite any such reasonable hypotheses? I.e. under what assumptions do you think that saving a random poor person’s life is likely to be a net negative? Sum over the number of lives saved and even if one person grows up to be a serial killer, the total is still way positive.
Most obviously it could cause an increase in world GDP without a commensurate acceleration in various risk prevention mechanisms. Species can evolve themselves to extinction and in a similar way humans could easily develop themselves to extinction if they are not careful or lucky. Messing around with various aspects of the human population would influence this… in one direction or another. It’s damn hard to predict.
Having a heuristic “short term lives saved == good” is useful. It massively simplifies calculations and if you have no information either way about side effects of the influence then it works well enough. But it would a significant epistemic error to mistake the heuristic for operating under uncertainty with confidence about the unpredictable (or difficult to predict) system in which you are operating.
Can you really defend a situation in which it is preferable to have living people today die from malaria?
What is socially defensible is not the same thing as what is accurate. But that isn’t the point here. All else being equal I would prefer AMF to have an extra million dollars to spend than to not have that extra million dollars. The expected value is positive. What I criticise is “very likely under all reasonable hypotheses” which is just way off. I do not have the epistemic resources to arrive at that confidence and I believe that you are arriving at that conclusion in error, not because of additional knowledge or probabilistic computational resources.
In fact, I’d expect AMF to have a net-negative impact (and a large one at that) a few decades down the line, unless there are unrealistic, unprecedented, imperialistic-in-scope, gigantic efforts to educate and provide for the dozen then-adult children (and their dozen children) a saved-from-malaria child can typically have.
Here’s Tom Friedman in his recent “Tell Me How This Ends” column:
I’ve been traveling to Yemen, Syria and Turkey to film a documentary on how environmental stresses contributed to the Arab awakening. As I looked back on the trip, it occurred to me that three of our main characters — the leaders of the two Yemeni [different countries, same dynamic] villages that have been fighting over a single water well and the leader of the Free Syrian Army in Raqqa Province, whose cotton farm was wiped out by drought — have 36 children among them: 10, 10 and 16.
It is why you can’t come away from a journey like this without wondering not just who will rule in these countries but how will anyone rule in these countries?
Do you really want to propose that it is better to let children in poor countries die of disease now than to save them, because they might have more children later? My prior on this is that you’re trolling, but if you really believe that and are willing to state it that baldly; then it might be worth having a serious conversation about population.
I’m not trolling. It’s a very touchy subject for sure. I would certainly highly prefer a world in which AMF succeeds if it is coupled with the necessary, massive changes to deal with the consequences of AMF succeeding.
A world in which just AMF succeeds, but in which the changes to deal with the 5 or 6 additional persons for every child surviving malaria do not happen is heading towards even greater disaster. The birth rate is not a “might have more children”, it’s a probabilistic certainty, without the aforementioned new pseudo-imperialism.
However, the task of nation-building and uplifting civil-war ravaged tribal societies is a task that dwarfs AMF (plenty of recent examples), or even the worldwide charity budget. Yet without it, what’s gonna happen, other than mass famines and other catastrophes?
I’m not talking about general Malthusian dynamics, but about countries whose population far exceeds the natural resources to support it, and which often do not offer the political environment, the infrastructure or the skills to exploit and develop what resources they have, other than trade them to the Chinese to prop up the ruling classes.
I’d expect a world in which AMF succeeds, leading to predictable tragedies on a more massive scale down the line, to be off worse than a world without AMF, with tragedies on a smaller scale. (To reiterate: A world with AMF succeeding and a long-term perspective for the survivers would be much better still.)
I’d rather contribute to charities which do not promise short-term benefits with probable long-term calamities, but rather to e.g. education projects and the development of stable civil institutions in such countries. (The picture gets fuzzied because eliminating certain disruptive diseases also has such positive externalities, but to a smaller degree.)
This ignores the social-scientific consensus that reducing infant mortality leads to reductions in family sizes. The moral dilemma you’re worried about doesn’t exist.
Citations needed. The relevant time horizons here are only 2-3 generations, do you suggest that societal norms will adapt faster than that (Edit: without accompanying larger efforts to build civil institutions)? The population explosion in, say, Bangladesh (1951: 42 million, 2011: 142 million) seems to suggest otherwise.
The phenomenon HaydnB refers to is the demographic transition, the theory of which is perhaps the best-established theory in the field of demography. Hereare two highly-cited reviews of the topic.
The relevant time horizons here are only 2-3 generations, do you suggest that societal norms will adapt faster than that? The population explosion in, say, Bangladesh (1951: 42 million, 2011: 142 million) seems to suggest otherwise.
HaydnB’s referring to family size, you’re referring to population, and it’s quite possible for the second to increase even as the first drops. This appears to be what happened in Bangladesh. I have not found any data stretching back to 1951 for completed family size in Bangladesh, but here is a paper that plots the total fertility rate from 1963 to 1996: it dropped from just under 8 to about 3½. I did find family size data going back to 1951 for neighbouring India: it fell from 6.0 in 1951 to 3.3 in 1997, with a concurrent decrease in infant mortality.
So I’m not HaydnB, but I have to answer your question with a “yes”: fertility norms can change, and have changed, greatly in the course of 2-3 generations. Bangladesh’s population, incidentally, is due to top out in about 40 years at ~200 million, only 40% higher than its current population.
During the transition, first mortality and then fertility declined, causing population growth rates first to accelerate and then to slow again, moving toward low fertility, long life and an old population.
From the second review:
It is true, however, that mortality reductions in poor countries and the consequent
rapid growth of population may impede capital formation and other aspects of
development. (Goes on to call the consequences mostly positive.)
Like Democratic Peace Theory, the demographic transition has historically been modeled after the now developed countries. At least that is where we get the latter “stages” from. Countries in which the reduction in mortality was achieved from within the country, a token of the relative strength of some aspects of its civil society. Not countries in which mortality reduction would be a solely external influence, transplanted from a more developed society into a tribal society.
but I have to answer your question with a “yes”: fertility norms can change, and have changed
Note that the question was whether societal norms will adapt faster than that, not whether they can and have in e.g. European countries. Especially if—and that’s the whole point of the dilemma—there are stark interventions (AMF) only in infant and disease mortality, without the much more difficult and costly interventions in nation building.
Will reducing infant / disease mortality alone thrust a country into a more developed status? Rather the contrary, since even the sources agree that the immediate effect would be even more of the already catastrophic population growth. Once you’re over the brink, a silver lining at the horizon isn’t as relevant.
As with the Bangladesh example, “only 40% higher than its current population” (and Bangladesh is comparatively developed anyways), if that figure translated (which it doesn’t) to Sub-Saharan populations, that would already be a catastrophe right there.
The question is, without nation building, would such countries be equipped to deal with just a 40% population rise over 40 years, let alone the one that’s actually prognosticated?
HaydnB doesn’t see the dilemma, since he seems to say that taking a tribal society, then externally implementing mortality reductions without accompanying large scale nation building will still reduce family sizes drastically, to the point that there are no larger scale catastrophes, even without other measures.
[quotations from reviews about population growth, emphasizing rapid/accelerating population growth]
These are consistent with what I wrote. Moreover, the world has already passed through the phase of accelerating population growth. The world’s population was increasing most rapidly 20-50 years ago (the exact period depends on whether one considers relative or absolute growth rates).
Like Democratic Peace Theory, the demographic transition has historically been modeled after the now developed countries. [...] Not countries in which mortality reduction would be a solely external influence, transplanted from a more developed society into a tribal society.
True enough, but mostly a moot point nowadays, because we’re no longer just predicting a fertility decline based on history; we’re watching it happen before our eyes. The global total fertility rate (not just mortality) has been in freefall for 50 years and even sub-Saharan Africa has had a steadily falling TFR since 1980.
Note that the question was whether societal norms will adapt faster than that, not whether they can and have in e.g. European countries.
Right, but the fact that they can change, have changed, and continue to change (in two large, poor, and very much non-European countries) is good evidence they’ll carry on changing. If medical interventions and other forms of non-institutional aid haven’t arrested the TFR decline so far, why would they arrest it in future?
Will reducing infant / disease mortality alone thrust a country into a more developed status? Rather the contrary, since even the sources agree that the immediate effect would be even more of the already catastrophic population growth.
The long-run effect matters more than the immediate effect (which ended decades ago).
The question is, without nation building, would such countries be equipped to deal with just a 40% population rise over 40 years, let alone the one that’s actually prognosticated?
The question I was addressing was the narrower one of whether reducing infant mortality reduces family sizes. Correlational evidence suggests (though does not prove) it does, maybe with a lag of a few years. I know of no empirical evidence that reductions in infant mortality increase family size in the long run, although they might in the short run.
Still, I might as well comment quickly on the broader question. As far as I know, the First World already focuses on stark interventions (like mass vaccination) more than nation building, and has done since decolonization. This has been accompanied by large declines in infant mortality, TFRs & family sizes, alongside massive population growth. It’s unclear to me why carrying on along this course will unleash disaster, not least because the societies you’re talking about are surely less “tribal” now than they were 10 or 20 or 50 years ago.
I don’t want to come off as Dr. Pangloss here. It’s quite possible global disaster awaits. But if it does happen, I’d be very surprised if it were because of the mechanism you’re proposing.
if development of newer institutions is what you are interested in, you can choose to contribute to charter cities or seasteading. That would be an intermediate risk-reward option between a low risk option like AMF and high risk high reward one like MIRI/FHI.
I’ll grant that MIRI could accelerate the creation of AGI, if their efforts to educate people about UFAI risks are particularly ineffective. But as far as UFAI creation at all is concerned, there are any number of very smart idiots in the world who would love to be on the news as “the first person to program an artificial general intelligence”. Or to be the first person to use a general AI to beat the stock market, as soon as enough parts of the puzzle have been worked out to make one by pasting together published math results. (Maybe a slightly more self-aware variation of AIXI-mc would do the trick.)
In my view, AGI is more or less inevitable, and MIRI is seemingly the only group publically interested in making it safe.
by conservation of expected evidence, one can expect this trend to continue
Not really related to the current discussion, but I want to make sure I understand the above statement. Is this assuming that the trend has not already been taken into account in forming the estimates?
Yes — the cost-effectiveness estimate has been adjusted every time a new issue has arisen, but on a case by case basis, without an attempt to extrapolate based on the historical trend.
I tend to think that if one can make a for-profit entity, that’s the best sort of vehicle to pursue most tasks, though occasionally, churches or governments have some value too.
The top considerations that come into play when I advise someone whether to earn-to-give or work directly on x-risk look like this:
1) Does this person have a large comparative advantage at the direct problem domain? Top-rank math talent can probably do better at MIRI than at a hedge fund, since there are many mathematical talents competing to go into hedge funds and no guarantee of a good job, and the talent we need for inventing new basic math does not translate directly into writing the best QT machine learning programs the fastest.
2) Is this person going to be able to stay motivated if they go off on their own to earn-to-give, without staying plugged into the community? Alternatively, if the person’s possible advantage is at a task that requires a lot of self-direction, will they be able to stay on track without requiring constant labor to keep them on track, since that kind of independent job is much harder to stick at then a 9-to-5 office job with supervision and feedback and cash bonuses?
Every full-time employee at a nonprofit requires at least 10 unusually generous donors or 1 exceptionally generous donor to pay their salary. For any particular person wondering how they should help this implies a strong prior bias toward earning-to-give. There are others competing to have the best advantage for the nonprofit’s exact task, and also there are thousands of job opportunities out there that are competing to be the maximally-earning use of your exact talents—best-fits to direct-task-labor vs. earning-to-give should logically be rare, and they are.
The next-largest issue is motivation, and here again there are two sides to the story. The law student who goes in wanting to be an environmentalist (sigh) and comes out of law school accepting the internship with the highest-paying firm is a common anecdote, though now that I come to write it down, I don’t particularly know of any gathered data. Earning to give can impose improbability in the form of likelihood that the person will actually give. Conversely, a lot of the most important work at the most efficient altruistic organizations is work that requires self-direction, which is also demanding of motivation.
I should pause here to remark that if you constrain yourself to ‘straightforward’ altruistic efforts in which the work done is clearly understandable and repeatable and everyone agrees on how wonderful it is, you will of course be constraining yourself very far away from the most efficient altruism—just like a grant committee that only wants to fund scientific research with a 100% chance of paying off in publications and prestige, or a VC that only wanted to fund companies that were certain to be defensible-appearing decisions, or someone who constrained their investments to assets that had almost no risk of going down. You will end up doing things that are nearly certain never to appear to future historians as a decisive factor in the history of Earth-originating intelligent life; this requires tolerance for not just risk but scary ambiguity. But if you want to work on things that might actually be decisive, you will end up in mostly uncharted territory doing highly self-directed work, and many people cannot do this. Just as many other people cannot sustain altruism without being surrounded by other altruists, but this can possibly be purchased elsewhere via living on the West or East Coast and hanging around with others who are earning-to-give or working directly.
These are the top considerations when someone asks me whether they should work directly or earn to support others working directly—the low prior, whether the exact fit of talent is great enough to overcome that prior, and whether the person can sustain motivation / self-direct.
My main comment on this is that if self-direction is as important as it appears to be, it would seem to me that ‘become self directed’ really should be everyone’s first priority if they can think of any way to do that. My second comment is that it seems to me that if one is self-directed and seeks appropriate mentorship, the expected value of pursuing a conventional career is very low compared to that of pursuing an entrepreneurial career. Conversely, mentorship or advice that doesn’t account for the critical factor of how self-directed someone is, as well as a few other critical factors such at the disposition to explore options, respond to empirical feedback from the market, etc, is likely to be worse than useless.
Can you expand on this? How does one seek appropriate mentorship?
Isn’t $36K/yr the modal MIRI salary? That doesn’t feel like it should be too hard on a $100K/yr software developer salary considering that charitable donations are tax-deductible up to 50% of your income (supposedly). If one donated $36K/yr out of their software developer salary, at $64K/yr they’d still be earning much more than a typical nonprofit employee (heck, much more than a typical college graduate), and if they were to pretend they were working at a nonprofit and subsist on $36K/yr themselves, they could probably subsidize 2 people at the modal MIRI salary (after several years’ worth of promotions/raises).
$36k/yr salary works out to be about $50k-$70k gross expense (including benefits, insurance, taxes etc) for a regular employer, not sure how much it is for a non-profit like MIRI.
10% is the charitable giving limit. There is another thing to be asked about, and that is the impact of the job. If I were to be a tax lawyer, I would be directly harming the ability of the US government to spend on social welfare programs. If I worked on Wall Street anywhere but Vanguard I would be bilking people out of their life savings, and at Vanguard I wouldn’t be making $100 K a year. Someone working as a tobacco farmer to raise money for cancer research has some misplaced priorities.
Where is that 10% number coming from? Looks to me like the limit is at least 20% in the US, and up to 50% for some organizations.
(BTW, can someone from MIRI or anyone else tell us if they’re a 50% organization?)
EDIT: and by the way, that’s just the limit on what’s tax-deductible. There’s no legal limit on how much you can actually give.
MIRI is a 50% organization.
See IRS Exempt Organizations Select Check and click the “Deductibility Status”
Malo knows this, but I’ll say it publicly:
In general, we suspect there are few people for whom it’s healthy to actually be giving away 50% of their income.
I understand why you said this, but most people interested in this are interested in the transition from 10% to >10% (say, 20), not in 10% to 50%. I presume you would estimate a higher number for whom this is healthy?
Yes.
Awesome, thanks!
I guess we also have to worry about state and maybe even city-specific tax laws too, huh?
You could always go work for the IRS. It employs a lot of tax lawyers.
But there’s a bigger issue: you think that the work of a (privately employed) tax lawyer intrinsically harms the ability of the US government to spend? That belief has LOTS of issues. I’ll start with two: One, why do you think the capability of the US government to spend is an unalloyed good thing? And two, do you happen to know the volume (say, in feet of shelf space) of the current tax laws, regulations, and rulings? I’d recommend you find out and then think about whether any moderately complicated business can comply with them without the help of a tax lawyer.
Sigh. First, Vanguard is not part of Wall Street. Second… you really should not believe everything the popular media keeps feeding you.
By “Wall Street” I’m including the Buy Side as well as the Sell Side. The big buyside firms like Fidelity and Charles Schwab sell products that most people shouldn’t buy. Insurance probably has a better case to buy some actively managed products, or some exotic derivatives, but I don’t know why it can’t do it itself.
To the extent that finance reallocates risk it can provide a positive utility benefit. However, the very productive businesses have questionable utility. Promoting active trading, picking hot funds etc, all eat into the returns clients can expect. Justify the existence of Charles Schwab’s S&P 500 index fund, with expense ratio twice that of Vanguards. The most profitable divisions of investment banks tend to be the ones with the least competition, and hence most questionable social benefit.
I’m aware Dodge and Cox is in SF, and Vanguard in Valley Forge, Blackrock in Princeton, etc. However, they are all on “the Street”.
The IRS doesn’t pay well: for government pay one might as well work for NASA and accomplish something fun.
Tax lawyers can not decrease taxes taken by US government in the long run, because US government gets to make the law adjusting for the existence of tax lawyering. This is why I have absolutely no qualms about employing a tax lawyer in the US.
Not in the U.S. (note these are in pre-tax earnings, so they translate into less in foregone consumption than they do in donations made).
Regarding this:
See this essay:
THIS. Although I`m unsure about the particulars you mention here, being an European, people and effective altruists need to realize that your job is INSIDE the world you live in. Estimating how much good you’re producing is not just about how much money/time you’re giving to effective charities, but also how much your way of life is helping/damaging the world.
I’m not convinced. The amount of saved lives, QALYs, or whatever you are counting that the US government welfare program gets per dollar is (or seems to be to me) quite a bit less than the amount that, say, the AMF could get with that money. I don’t know how many dollars per QALY US government welfare manages to get, but I wouldn’t be surprised if it were on the order of $1000-$10000 per QALY. And that’s not even counting the fact that even if the US goverment had that bit more money from you not being a tax lawyer, that money would not all go to welfare and other such efficient (relative to what else the government spends money on) projects. I would imagine a fair portion would go to, say, bombing Syria, or hiring an extra parking-meter enforcer, or such inefficent stuff, that get an even worse $/QALY result.
And that is still not to mention the fact that some of that money would go to, say, funding the NSA to spy on your phone calls and read your email, or to the TSA to harass, strip-search, and detain you, which are net negatives.
And even that is not counting that MIRI may end up having a QALY/$ result far, far higher than anything the AMF or whoever could ever hope of possibly getting.
I’m not saying you’re flat-out wrong, and it is something to take into consideration when figuring out the altruistic impact of your job, but taking into account these objections, it seems highly unlikely that the marginal dollar from the government goes far enough to weigh very heavily in ones analysis.
On the topic of how much it takes to save a QALY in the US:
-from http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1497852/
The first paragraph of this gives more on the cost of QALYs in the US. So, kidney dialysis is an intervention that is paid for by the government in the US, and it comes in at more than $100,000 per QALY saved.
Since marginal funding generally goes to pay for interventions which are no more effective than those already being paid for, I wouldn’t expect the cost of a marginal QALY to be below (say) $50,000.
I’m not sure if you were answering my comment or wubbles’s one. What I was saying was that you need to take into account the negative impact your job and way of life have on the world.
I agree that the US government probably is terrible at using tax money to better the world.
Do you actually think that a finance professional who donated a significant portion of their income to effective charity would be doing more harm than good? Even given that you can save the life of a child in the developing world for on the order of $2000?
I don’t think the problem is finance professionals in general—it’s finance professionals in particularly corrupt parts of the industry.
Figuring out in advance that a job is doing particularly corrupt work seems to be something that people are very bad at—I don’t know whether it’s mostly that it would be hard for a neutral observer, or that people don’t want the problems of dealing with the consequences to their own lives if they find that their job is destructive.
Hmm, I was thinking the assumption (which I don’t necessarily entirely agree with) was that finance professionals were simply earning money without providing any benefit to society, and so a net negative. It sounds like your comment assumes that some of them are actually actively doing harm (though perhaps unintentionally), beyond just taking their own paycheck’s worth out of the productive economy. Is that your understanding?
The mortgage crisis was a result of banks being able to sell mortgages to other banks. This meant that the bank making the loan could make money just by the mortgage being initiated—the first bank no longer had a strong interest in the loan being repaid.
There were some other pieces to the situation that I don’t have clear in my mind at the moment, but I think there were incentives for the mortgage to actually not be repaid and the house to be taken by a bank.
One piece that I am clear on is that there were people who decided it wasn’t worth it for banks to keep accurate track of who owned which mortgage, or what had been paid, or what had been agreed to.
This is stealing people’s houses. It’s a degree of damage which it’s hard to imagine being covered by charity.
The other side of the story is that not every bank behaved like that—not all of finance is fraudulent.
This is not true. In fact, this is probably not even wrong...
Its at least somewhat true, if perhaps not well stated- packaged mortgages and derivatives based on packaged mortgages (mortgages sold as investment vehicles to other banks and funds) played a very large role in the crisis.
Without “selling mortgages to other banks” the popping of the housing bubble wouldn’t have turned into the liquidity crunch that started in 2008.
So were mortgages by themselves. Without the widespread availability of mortgages “the popping of the housing bubble wouldn’t have turned into the liquidity crunch” too. Or, for that matter, without the fact that the “standard” mortgage is a 30-year fixed—not, say, a 1⁄1 ARM.
But anyway, the reason for the contagion from mortgages to liquidity wasn’t the ability to sell mortgages. It was the mispricing of mortgage derivatives, specifically the widespread belief that certain tranches of collateralized mortgage obligations (CMOs) were effectively risk-free.
If you want to dig deeper, the real cause was the global asset bubble helped by the too-loose monetary policy in the mid-2000s.
Financial economics are complicated. Snap judgements from popular press rarely have much relationship to reality.
Well, sure. And also without houses themselves…
My point is that the statement wasn’t false on its face. Repackaging and reselling was A proximate cause of the liquidity crunch, its not the only cause but its a part of what happened.
Is it really? I suppose this depends on how many houses any individual is responsible for and how much money they capture per house. I guess that second part is the real issue—any individual who would be giving to charity probably only captures a fraction of what they earn for their firm.
But if you could capture the whole value of a predatory mortgage and convert it into developing world lives saved, it’s not hard to imagine the numbers adding up. (One American family goes bankrupt and 20 Malawian children who otherwise would have don’t die in childhood? On the face of it that looks like a pretty positive net outcome.)
If you can do outsized damage significantly beyond what you can capture as income though, then I suppose it gets a bit tougher to justify.
If we’re talking about donations on the scale of the activities that went into the mortgage crisis, I think you’d start to suffer seriously diminishing returns.
Even if you didn’t, there are other problems you’d run into, such as the limited ability of the Malawian (or other impoverished African) society and economy to accommodate such a sudden spike in children surviving to adulthood. The lives that you save from extermination at the hands of malaria or other preventable causes are probably mostly going to be relatively lousy or short due to other causes, pending much further investment.
As I understood it, the hypothetical was a single individual deciding to work in finance and donate a large portion of their income to efficient charity. In that case I don’t think the diminishing returns are so much of an issue.
I would worry more about negative flow-through effects of a decline in trust and basic decency in society. I think those are much more clear than flow-through effects of positive giving. I’m not sure if this outweighs the 20-to-1 ratio.
Government social welfare spending is notoriously inefficient. So if your client is at all generous with his money you’re coming out ahead. Heck even if he doesn’t give to charity but does use the money to invest in productive enterprises, you’re probably coming out ahead. And that’s before taking into account how you spend your money.
The cost (to an organization) of an employee is more than just his salary, often considerably more. There is health insurance and other benefits, payroll taxes, infrastructure support (e.g. a computer, a desk to put it on, a room to put the desk in), etc.
One potentially relevant note for anyone considering this is that 100 − 36 = 64, not 74.
Thanks, fixed. I appreciate the correction… no need to retract your comment! :)
I know the comment was probably fine, but overall it seemed like it could be read as unnecessarily snarky and hence lower the tone—PMing you the correction would have been a better move.
All in all I think that the standard for discussion here on LessWrong could be increased a lot if people stopped giving “wiseass” replies to things, were more forgiving of minor errors (while still pointing them out), and so on.
Be the change you want to see on LessWrong!
FWIW, it didn’t come off as snarky at all to me—I read it as exactly the sort of polite correction that I like on LW.
If you define “generous” by “amount of capital” then this is tautologically true. But by this standard, extraordinarily wealthy people are capable of being exceptionally exceptionally exceptionally generous. I’d recur to my remark about the Giving Pledge. I believe that the projects of highest humanitarian value will generally get funded.
In principle this could fall under the “unusual values” consideration that I raise above. But I don’t think that the sociological phenomenon that you seem to be implying to exist prevails in practice. I think that there a lot of funders who are not risk-averse, and indeed, many who are actively attracted to high risk projects.
Well, if James Simons wanted to retire from Renaissance and work on FAI full-time, it would not be entirely obvious to me that this was a bad move, but only if Simons had enough in the bank to also pay as much other top-flight math talent as could reasonably be used, and was already so paying, such that there was no marginal return to his further earning power relative to existing funds.
This situation has not yet arisen. Unfortunately.
I think that James Simons is an example of someone with an unusually strong comparative advantage at making money. But this wouldn’t necessarily have been clear a priori: if you put yourself in Simons’ shoes in 1980 the expected earnings of going into finance would be much lower than his actual earnings turned out to be. So it’s not clear that he would have done better to “earn to give” than doing something of direct humanitarian value (though maybe it was clear from the outset that his comparative advantage was in finance.)
Edit: [Moved comment to a different place]
I’ll also highight another point implicit in my post: even if one assumes that there’s not enough funding in the nonprofit world for the projects of highest value, there may be such funding available in other contexts (for-profit, academic and government). This makes the argument for earning to give weaker.
I recognize that I haven’t addressed the specific subject of Friendly AI research, and will do so in future posts.
I understand if your priorities aren’t our priorities. My concrete example reflex was firing, that’s all.
I think that there’s substantial overlap between my values and MIRI staff’s values, and that the difference regarding the relative value of “earning to give” is epistemic rather than normative. But obviously there’s a great deal more that needs to be said about the epistemic side, with reference to the concrete example of Friendly AI.
I can imagine someone thinking that FHI was a better use of money than MIRI, or CFAR, or CSER, or the Foresight Institute, or brain-scanning neuroscience, or rapid-response vaccines, or any number of startups, but considering AMF as being in the running at all seems to require either a value difference or really really different epistemics about what affects the fate of future galaxies.
Realistic amounts of difference in epistemics + the “humans best stick to the mainline probability” heuristic seem enough (where by “realistic” I mean “of the degree actually found in the world”). I.e., I honestly believe that there are many people out there who would care the hell about the fate of future galaxies if they alieved that they had any non-vanishing chance of significantly influencing that fate (and to choose the intervention that influences it in the desired direction).
If you’re one of 10^11 sentients to be born on Ancient Earth with a golden opportunity to influence a roughly 10^80-sized future, what exactly is a ‘vanishing chance’… eh, let’s all save it until later.
I meant that the alieved probability is small in absolute terms, not that it is small compared to the payoff. That’s why I mentioned the “stick to the mainline probability” heuristic. I really do believe that there are many people who, if they alieved that they (or a group effort they could join) could change the probability of a 10^80-sized future by 10%, would really care; but who do not alieve that the probability is large enough to even register, as a probability; and whose brains will not attempt to multiply a not-even-registering probability with a humongous payoff. (By “alieving a probability” I simply mean processing the scenario the way one’s brain processes things it assigns that amount of credence, not a conscious statement about percentages.)
This is meant as a statement about people’s actual reasoning processes, not about what would be reasonable (though I did think that you didn’t feel that multiplying a very small success probability with a very large payoff was a good reason to donate to MIRI; in any case seems to me that the more important unreasonableness is requesting mountains of evidence before alieving a non-vanishing probability for weird-sounding things).
[ETA: I find it hard to put a number on the not-even-registering probability the sort of person I have in mind might actually alieve, but I think a fair comparison is, say, the “LHC will create black holes” thing—I think people will tend to process both in a similar way, and this does not mean that they would shrug it off if somebody counterfactually actually did drop a mountain of evidence about either possibility on their head.]
Because on a planet like this one, there ought to be some medium-probable way for you and a cohort of like-minded people to do something about x-risk, and if a particular path seems low probability, you should look for one that’s at least medium-probability instead.
Ok, fair enough. (I had misunderstood you on that particular point, sorry.)
If there was ever a reliable indicator that you’re wrong about something, it is the belief that you are special to the order of 1 in 10^70.
So do you believe in the Simulation Hypothesis or the Doomsday Argument, then? All attempts to cash out that refusal-to-believe end in one or the other, inevitably.
From where I stand, it’s more like arcane meta-arguments about probability are motivating a refusal-to-doubt the assumptions of a prized scenario.
Yes, I am apriori skeptical of anything which says I am that special. I know there are weird counterarguments (SIA) and I never got to the bottom of that debate. But meta issues aside, why should the “10^80 scenario” be the rational default estimation of Earth’s significance in the universe?
The 10^80 scenario assumes that it’s physically possible to conquer the universe and that nothing would try to stop such a conquest, both enormous assumptions… astronomically naive and optimistic, about the cosmic prospects that await an Earth which doesn’t destroy itself.
Okay, so that’s the Doomsday Argument then: Since being able to conquer the universe implies we’re 10^70 special, we must not be able to conquer the universe.
Calling the converse of this an arcane meta-argument about probability hardly seems fair. You can make a case for Doomsday but it’s not non-arcane.
Perhaps this is hairsplitting but the principle I am employing is not arcane: it is that I should doubt theories which imply astronomically improbable things. The only unusual step is to realize that theories with vast future populations have such an implication.
I am unable to state what the SIA counterargument is.
In the theory that there are astronomically large numbers of people, it is a certainty that some of them came first. The probability that YOU are one of those people equal to the probability that YOU are any one of those other people. However, it does define a certain small narrow equivalence class that you happen to be a member of.
It’s a bit like the difference between theorizing that: A) given that you bought a ticket, you’ll win the lottery, and B) given that the lottery folks gave you a large sum, that you had the winning ticket.
That’s not the “SIA counterargument”, which is what I want to hear (in a compact form, that makes it sound straightforward). You’re just saying “accept the evidence that something ultra-improbable happened to you, because it had to happen to someone”.
I was only replying to the first paragraph, really. Even under the SSA there’s no real problem here. I don’t see how the SIA makes matters worse.
Right. That’s arcane. Mundane theories have no need to measure the population of the universe.
But it’s still a simple idea once you grasp it. I was hoping you could state the counterargument with comparable simplicity. What is the counterargument at the level of principles, which neutralizes this one?
I largely agree with your skepticism. I would go even farther and say that even the 10^80 scenario happens, what we do now can only influence it by random chance, because the uncertainty in the calculations of the consequences of our actions in the near term on the far future overwhelms the calculations themselves. That said, we should still do what we think is best in the near term (defined by our estimates of the uncertainty being reasonably small), just not invoke the 10^80 leverage argument. This can probably be formalized, by assuming that the prediction error grows exponentially with some relevant parameter, like time or the number of choices investigated, and calculating the exponent from historical data.
Doomsday for me, I think. Especially when you consider that it doesn’t mean doomsday is literally imminent, just “imminent” relative to the kind of timescale that would be expected to create populations on the order of 10^80.
In other words, it fits with the default human assumption that civilization will basically continue as it is for another few centuries or millennia before being wiped out by some great catastrophe.
Do you mind elaborating on this inevitability? It seems like there ought to be other assumptions involved. For example, I can easily imagine that humans will never be able to colonize even this one galaxy, or even any solar system other than this one. Or that they will artificially limit the number of individuals. Or maybe the only consistent CEV is that of a single superintelligence of which human minds will be tiny parts. All of these result in the rather small total number of individuals existing at any point in time.
Counts as Doomsday, also doesn’t work because this solar system could support vast numbers of uploads for vast amounts of time (by comparison to previous population).
This is a potential reply to both Doomsday and SA but only if you think that ‘random individual’ has more force than a similar argument from ‘random observer-moment’, i.e. to the second you reply, “What do you mean, why am I near the beginning of a billion-year life rather than the middle? Anyone would think that near the beginning!” (And then you have to not translate that argument back into a beginning-civilization saying the same thing.)
...whereupon we wonder something about total ‘experience mass’, and, if that argument doesn’t go through, why the original Doomsday Argument / SH should either.
Thanks, I’ll chew on that a bit. I don’t understand the argument in the second and third paragraphs. Also, it’s not clear to me whether by “counts as doomsday” you mean the standard doomsday with the probability estimates attached, or some generalized doomsday, with no clear timeline or total number of people estimated.
Anyway, the feeling I get from your reply is that I’m missing some basic background stuff here I need to go through first, not the usual “this guy is talking out of his ass” impression when someone invokes anthropics in an argument.
No, this is talking-out-of-our-ass anthropics, it’s just that the anthropic part comes in when you start arguing “No, you can’t really be in a position of that much influence”, not when you’re shrugging “Sure, why shouldn’t you have that much influence?” Like, if you’re not arriving at your probability estimate for “Humans will never leave the solar system” just by looking at the costs of interstellar travel, and are factoring in how unique we’d have to be, this is where the talking-out-of-our-ass anthropics comes in.
Though it should be clearly stated that, as always, “We don’t need to talk out of our ass!” is also talking out of your ass, and not necessarily a nicer ass.
Or when you (the generic you) start arguing “Yes, I am indeed in a position of that much influence”, as opposed to “There is an unknown chance of me being in such a position, which I cannot give a ballpark estimate for without talking out of my ass, so I won’t”?
When you try to say that there’s something particularly unknown about having lots of influence, you’re using anthropics.
Huh. I don’t understand how refusing to speculate about anthropics counts as anthropics. I guess that’s what you meant by
I wonder if your definition of anthropics matches mine. I assume that any statement of the sort
is anthropics. I do not see how refusing to reason based on some arbitrary set of observers counts as anthropics.
Right. So if you just take everything at face value—the observed laws of physics, the situation we seem to find ourselves in, our default causal model of civilization—and say, “Hm, looks like we’re collectively in a position to influence the future of the galaxy,” that’s non-anthropics. If you reply “But that’s super improbable a priori!” that’s anthropics. If you counter-reply “I don’t believe in all this anthropic stuff!” that’s also an implicit theory of anthropics. If you treat the possibility as more “unknown” than it would be otherwise, that’s anthropics.
OK, I think I understand your point now. I still feel uneasy about the projection like your influencing 10^80 people in some far future, mainly because I think it does not account for the unknown unknowns and so is lost in the noise and ought to be ignored, but I don’t have a calculation to back up this uneasiness at the moment.
Does he?
Does he what?
I’ve had a vague idea as to why the random observer-moment argument might not be as strong as the random individual one, though I’m not very confident it makes much sense. (But neither argument sounds anywhere near obviously wrong to me.)
I wonder if this argument can be made precise enough to have its premises and all the intermediate assumptions examined. I remain skeptical of any forecast that far into the future. You presumably mean your confidence in the UFAI x-risk within the next 20-100 years as the minimum hurdle to overcome, with the eternal FAI paradise to follow.
My reason for mentioning AMF and global health is that doing so provides a concrete, pretty robustly researched example, rather than as to compare with efforts to improve the far future of humanity.
I think that working in global health in a reflective and goal directed way is probably better for improving global health than “earning to give” to AMF. Similarly, I think that working directly on things that bear on the long term future of humanity is probably a better way of improving the far future of humanity than “earning to give” to efforts along these lines.
I’ll discuss particular opportunities to impact the far future of humanity later on.
That depends on what you want to know, doesn’t it? As far as I know the impact of AMF on x-risk, astronomical waste, and total utilons integrated over the future of the galaxies, is very poorly researched and not at all concrete. Perhaps some other fact about AMF is concrete and robustly researched, but is it the fact I need for my decision-making?
(Yes, let’s talk about this later on. I’m sorry to be bothersome but talking about AMF in the same breath as x-risk just seems really odd. The key issues are going to be very different when you’re trying to do something so near-term, established, without scary ambiguity, etc. as AMF.)
I’m somewhat confused by the direction that this discussion has taken. I might be missing something, but I believe that the points related to AMF that I’ve made are:
GiveWell’s explicit cost-effectiveness estimate for AMF is much higher than the cost per DALY saved implied by the figure that MacAskill cited.
GiveWell’s explicit estimates for the cost-effectiveness of the best giving opportunities in the field of direct global health interventions have steadily gotten lower, and by conservation of expected evidence, one can expect this trend to continue.
The degree of regression to the mean observed in practice suggests that there’s less variance amongst the cost-effectiveness of giving opportunities than may initially appear to be the case.
By choosing an altruistic career path, one can cut down on the number of small probability failure modes associated with what you do.
I don’t remember mentioning AMF and x-risk reduction together at all. I recognize that it’s in principle possible that the “earning to give” route is better for x-risk reduction than it is for improving global health, but I believe the analogy between the two domains is sufficiently strong that my remarks on AMF have relevance (on a meta-level, not on an object level).
Yeah, I also have the feeling that I’m questioning you improperly in some fashion. I’m mostly driven by a sense that AMF is very disanalogous to the choices that face somebody trying to optimize x-risk charity (or rather total utilons over all future time, but x-risk seems to be the word we use for that nowadays). It seems though that we’re trying to have a discussion in an ad-hoc fashion that should be tabled and delayed for explicit discussion in a future post, as you say.
If I may list some differences I perceive between AMF and MIRI:
AMF’s impact is quite certain. MIRI’s impact feels more like a long shot —or even a pipe dream.
AMF’s impact is sizeable. MIRI’s potential impact is astronomic.
AMF’s impact is immediate. MIRI’s impact is long term only.
AMF’s have photos of children. MIRI have science fiction.
In mainstream circles, donating to AMF gives you pats in the back, while donating to MIRI gives you funny looks.
Near mode thinking will most likely direct one to AMF. MIRI probably requires one to shut up and multiply. Which is probably why I’m currently giving a little money to Greenpeace, despite being increasingly certain that it’s far, far from the best choice.
One more difference:
AMF’s impact is very likely to be net positive for the world under all reasonable hypotheses.
MIRI appears to me to have a chance to be massively net negative for humanity. I.e. if AI of the level they predict is actually possible, MIRI might end up creating or assisting in the creation of UFAI that would not otherwise be created, or perhaps not created as soon.
But what if AMF saves a child who grows up to be a biotechnologist and goes on to weaponize malaria and spread it to millions?
If you try hard enough, you can tell a story where any effort to accomplish X somehow turns out to accomplish ~X, but one must distinguish possibility from the balance of probability.
Yes, and the story where the child who grows up to be a biotechnologist and goes on to weaponize malaria and spread it to millions doesn’t pass the balance of probability test. The story that MIRI creates a dangerous AI fails to pass the balance of probability test only to the extent that one believes it is improbable that anyone can create such an AI. I do indeed consider it far more likely than not that there will never be the all-powerful AI you fear. And by that standard donations to MIRI are simply ineffective compared to donations to AMF.
However if I’m wrong about that and powerful FOOMing UFAIs are in fact possible, then I need to consider whether MIRI’s work is wise. If AIs do FOOM, there seems to me to be a very real possibility that MIRI’s work will either create a UFAI while trying to create a FAI, or alternatively enable others to do so. I’m not sure that’s more likely than that MIRI will one day create a FAI, but you can’t just multiply by the value of a very positive and very speculative outcome without including the possibility of a very negative and very speculative outcome.
If you increase the probability of uFAI in order for MIRI to kill everyone, the probability of someone else doing it goes up even more.
Maybe. I’m not sure about that though. MIRI is the only person or organization I’m aware of that seems to want to create a world controlling AI; and it’s the world-controlling part that I find especially dangerous. That could send MIRI’s AI in directions others won’t go. Are there other organizations attempting to develop AIs to control the world? Is anyone else trying to build a benevolent dictator?
MIRI’s stated goal is more meta:
They are well aware of the dangers of creating a uFAI, and you can be certain they will be real careful before they push a button that have the slightest chance of launching the ultimate ending (good or bad). Even then, they may very well decide that “being real careful” is not enough.
It probably doesn’t matter, as any uFAI is likely to emerge by mistake:
Is MIRI attempting to develop any sort of AI? I understood the current focus of its research to be the logic of Friendly AGI, i.e. given the ability to create a superintelligent entity, how do you build one that we would like to have created? This need not involve working on developing one.
That seems like a bizarre belief to hold. Or perhaps just overwhelmingly shortsighted. There are certainly reasonable hypotheses in which more people alive right now result in worse outcomes a single generation down the line, without even considering extinction level threats and opportunities. The world isn’t nearly easy enough to model and optimize for us to be that certain a disruptive influence on that scale will be a net positive under all reasonable hypotheses.
Would you care to cite any such reasonable hypotheses? I.e. under what assumptions do you think that saving a random poor person’s life is likely to be a net negative? Sum over the number of lives saved and even if one person grows up to be a serial killer, the total is still way positive. Can you really defend a situation in which it is preferable to have living people today die from malaria?
The problem with MIRI-hypothesized AI (beyond its implausibility) is that we don’t get to sum over all possible results. We get one result. Even if the chance of a good result is 80%, the chance of a disastrous result is still way too high for comfort.
Most obviously it could cause an increase in world GDP without a commensurate acceleration in various risk prevention mechanisms. Species can evolve themselves to extinction and in a similar way humans could easily develop themselves to extinction if they are not careful or lucky. Messing around with various aspects of the human population would influence this… in one direction or another. It’s damn hard to predict.
Having a heuristic “short term lives saved == good” is useful. It massively simplifies calculations and if you have no information either way about side effects of the influence then it works well enough. But it would a significant epistemic error to mistake the heuristic for operating under uncertainty with confidence about the unpredictable (or difficult to predict) system in which you are operating.
What is socially defensible is not the same thing as what is accurate. But that isn’t the point here. All else being equal I would prefer AMF to have an extra million dollars to spend than to not have that extra million dollars. The expected value is positive. What I criticise is “very likely under all reasonable hypotheses” which is just way off. I do not have the epistemic resources to arrive at that confidence and I believe that you are arriving at that conclusion in error, not because of additional knowledge or probabilistic computational resources.
In fact, I’d expect AMF to have a net-negative impact (and a large one at that) a few decades down the line, unless there are unrealistic, unprecedented, imperialistic-in-scope, gigantic efforts to educate and provide for the dozen then-adult children (and their dozen children) a saved-from-malaria child can typically have.
Here’s Tom Friedman in his recent “Tell Me How This Ends” column:
Do you really want to propose that it is better to let children in poor countries die of disease now than to save them, because they might have more children later? My prior on this is that you’re trolling, but if you really believe that and are willing to state it that baldly; then it might be worth having a serious conversation about population.
I’m not trolling. It’s a very touchy subject for sure. I would certainly highly prefer a world in which AMF succeeds if it is coupled with the necessary, massive changes to deal with the consequences of AMF succeeding.
A world in which just AMF succeeds, but in which the changes to deal with the 5 or 6 additional persons for every child surviving malaria do not happen is heading towards even greater disaster. The birth rate is not a “might have more children”, it’s a probabilistic certainty, without the aforementioned new pseudo-imperialism.
However, the task of nation-building and uplifting civil-war ravaged tribal societies is a task that dwarfs AMF (plenty of recent examples), or even the worldwide charity budget. Yet without it, what’s gonna happen, other than mass famines and other catastrophes?
I’m not talking about general Malthusian dynamics, but about countries whose population far exceeds the natural resources to support it, and which often do not offer the political environment, the infrastructure or the skills to exploit and develop what resources they have, other than trade them to the Chinese to prop up the ruling classes.
I’d expect a world in which AMF succeeds, leading to predictable tragedies on a more massive scale down the line, to be off worse than a world without AMF, with tragedies on a smaller scale. (To reiterate: A world with AMF succeeding and a long-term perspective for the survivers would be much better still.)
I’d rather contribute to charities which do not promise short-term benefits with probable long-term calamities, but rather to e.g. education projects and the development of stable civil institutions in such countries. (The picture gets fuzzied because eliminating certain disruptive diseases also has such positive externalities, but to a smaller degree.)
This ignores the social-scientific consensus that reducing infant mortality leads to reductions in family sizes. The moral dilemma you’re worried about doesn’t exist.
Citations needed. The relevant time horizons here are only 2-3 generations, do you suggest that societal norms will adapt faster than that (Edit: without accompanying larger efforts to build civil institutions)? The population explosion in, say, Bangladesh (1951: 42 million, 2011: 142 million) seems to suggest otherwise.
The phenomenon HaydnB refers to is the demographic transition, the theory of which is perhaps the best-established theory in the field of demography. Here are two highly-cited reviews of the topic.
HaydnB’s referring to family size, you’re referring to population, and it’s quite possible for the second to increase even as the first drops. This appears to be what happened in Bangladesh. I have not found any data stretching back to 1951 for completed family size in Bangladesh, but here is a paper that plots the total fertility rate from 1963 to 1996: it dropped from just under 8 to about 3½. I did find family size data going back to 1951 for neighbouring India: it fell from 6.0 in 1951 to 3.3 in 1997, with a concurrent decrease in infant mortality.
So I’m not HaydnB, but I have to answer your question with a “yes”: fertility norms can change, and have changed, greatly in the course of 2-3 generations. Bangladesh’s population, incidentally, is due to top out in about 40 years at ~200 million, only 40% higher than its current population.
From the first review:
From the second review:
Like Democratic Peace Theory, the demographic transition has historically been modeled after the now developed countries. At least that is where we get the latter “stages” from. Countries in which the reduction in mortality was achieved from within the country, a token of the relative strength of some aspects of its civil society. Not countries in which mortality reduction would be a solely external influence, transplanted from a more developed society into a tribal society.
Note that the question was whether societal norms will adapt faster than that, not whether they can and have in e.g. European countries. Especially if—and that’s the whole point of the dilemma—there are stark interventions (AMF) only in infant and disease mortality, without the much more difficult and costly interventions in nation building.
Will reducing infant / disease mortality alone thrust a country into a more developed status? Rather the contrary, since even the sources agree that the immediate effect would be even more of the already catastrophic population growth. Once you’re over the brink, a silver lining at the horizon isn’t as relevant.
As with the Bangladesh example, “only 40% higher than its current population” (and Bangladesh is comparatively developed anyways), if that figure translated (which it doesn’t) to Sub-Saharan populations, that would already be a catastrophe right there.
The question is, without nation building, would such countries be equipped to deal with just a 40% population rise over 40 years, let alone the one that’s actually prognosticated?
HaydnB doesn’t see the dilemma, since he seems to say that taking a tribal society, then externally implementing mortality reductions without accompanying large scale nation building will still reduce family sizes drastically, to the point that there are no larger scale catastrophes, even without other measures.
These are consistent with what I wrote. Moreover, the world has already passed through the phase of accelerating population growth. The world’s population was increasing most rapidly 20-50 years ago (the exact period depends on whether one considers relative or absolute growth rates).
True enough, but mostly a moot point nowadays, because we’re no longer just predicting a fertility decline based on history; we’re watching it happen before our eyes. The global total fertility rate (not just mortality) has been in freefall for 50 years and even sub-Saharan Africa has had a steadily falling TFR since 1980.
Right, but the fact that they can change, have changed, and continue to change (in two large, poor, and very much non-European countries) is good evidence they’ll carry on changing. If medical interventions and other forms of non-institutional aid haven’t arrested the TFR decline so far, why would they arrest it in future?
The long-run effect matters more than the immediate effect (which ended decades ago).
The question I was addressing was the narrower one of whether reducing infant mortality reduces family sizes. Correlational evidence suggests (though does not prove) it does, maybe with a lag of a few years. I know of no empirical evidence that reductions in infant mortality increase family size in the long run, although they might in the short run.
Still, I might as well comment quickly on the broader question. As far as I know, the First World already focuses on stark interventions (like mass vaccination) more than nation building, and has done since decolonization. This has been accompanied by large declines in infant mortality, TFRs & family sizes, alongside massive population growth. It’s unclear to me why carrying on along this course will unleash disaster, not least because the societies you’re talking about are surely less “tribal” now than they were 10 or 20 or 50 years ago.
I don’t want to come off as Dr. Pangloss here. It’s quite possible global disaster awaits. But if it does happen, I’d be very surprised if it were because of the mechanism you’re proposing.
if development of newer institutions is what you are interested in, you can choose to contribute to charter cities or seasteading. That would be an intermediate risk-reward option between a low risk option like AMF and high risk high reward one like MIRI/FHI.
I’ll grant that MIRI could accelerate the creation of AGI, if their efforts to educate people about UFAI risks are particularly ineffective. But as far as UFAI creation at all is concerned, there are any number of very smart idiots in the world who would love to be on the news as “the first person to program an artificial general intelligence”. Or to be the first person to use a general AI to beat the stock market, as soon as enough parts of the puzzle have been worked out to make one by pasting together published math results. (Maybe a slightly more self-aware variation of AIXI-mc would do the trick.)
In my view, AGI is more or less inevitable, and MIRI is seemingly the only group publically interested in making it safe.
Not really related to the current discussion, but I want to make sure I understand the above statement. Is this assuming that the trend has not already been taken into account in forming the estimates?
Yes — the cost-effectiveness estimate has been adjusted every time a new issue has arisen, but on a case by case basis, without an attempt to extrapolate based on the historical trend.
I tend to think that if one can make a for-profit entity, that’s the best sort of vehicle to pursue most tasks, though occasionally, churches or governments have some value too.