I try to view the world through critical lenses, to challenge existing thoughts/solutions to be a bit more considerate/complete. I do not reply to DMs for non-personal (with respect to the user who reached out directly) discussions, and will post here instead with reference to the user, and will post my reply there.
ZY
Karma: 16
If you meant for current LLMs, some of them could be misuse of current LLM by humans, or risks such as harmful content, harmful hallucination, privacy, memorization, bias, etc. For some other models such as ranking/multiple ranking, I have heard some other worries on deception as well (this is only what I recall of hearing, so it might be completely wrong).
This aligns similarly with my current view. Wanted to add a thought—current LLMs could still have unintended problems/misalignment like factuality or privacy or copyrights or harmful content, which still should be studied/mitigated, together with thinking about other more AGI like models (we don’t know what exactly yet, but could exist.) And a LLM (especially a fine tuned one), if doing increasingly well on generalization ability, should still be monitored. To be prepared for future, having a safety mindset/culture is important for all models.
I also wish to see more safety papers. I guess/from my experience that it might also be—really good quality research takes time, and the papers so far from them seems pretty good. Though I don’t know if they are actively withholding things on purpose which could also be true—any insider/sources for this guess?
I think this is conditioning on one problem with one goal, but I haven’t thought about the other good collectively (more of a discussion on consequentialism).
For best of personal ability, I think the purpose is to distinguish what one can do personally, and what one can do to engage collaboratively/collectively, but I need to think through that better it seems, so that is a good question.
My reason on the na is: have intention, no execution/enough execution, did good is more like an accident, which is the same with no intention, no execution, and did good.
When thinking about deontology and consequentialism in application, I was trying to rate morality of actions based on intention, execution, and outcome. (Some cells are “na” as they are not really logical in real world scenarios.)
In reality, to me, it seems executed “some” intention matters (though I am not sure how much) the most when doing something bad, and executed to the best ability matters the most when doing something good.
It also seems useful to me, when we try to learn about applications of philosophy from law. (I am not an expert though in neither philosophy nor law, except worked for lawyers in securities and finance law for a few years, so these may contain errors.)
Intention to kill the person Executed “some” intention Killed the person ”Bad” level Law Yes Yes Yes 10 murder Yes Yes No 8-10 as an example, attempted first-degree murder is punished by life in state prison (US, CA) Yes No Yes na Yes No No 0-5 no law on this (I can imagine for reasons on “it’s hard to prove”) but personally, assuming multiple “episodes”, or just more time, this leads to murder and attempted murder later anyways; very rare a person can have this thought without executing it in reality. No Yes Yes na No Yes No na No No Yes 0-5 typically not a crime, unless something like negligence No No No 0 Intention to save a person (have limited time)Executed intention to the best of abilitySaved the person ”Good” Level Yes Yes Yes 10 Yes Yes No 10 Yes No Yes na Yes No No 0-5 No Yes Yes na No Yes No na No No Yes 0-5 No No No 0 Intention to do good (have more time)Executed intention to the best of personal abilityDid good ”Good” Level Yes Yes Yes 10 Yes Yes No 8-10 Yes No Yes na Yes No No 0-5 No Yes Yes na No Yes No na No No Yes 0-5 No No No 0
(Mentioned some of these in our chat, but allow me to be repetitive)
On first: I don’t think efforts to reduce rape or discrimination needs 100% abolition, but working towards that currently has huge return at this point of history. Education has snowball effect as well. Just because it is hard to achieve 100%, does not mean there should not be efforts, nor impossible at all. In fact, it is something that is rarely worked on alone; for example, one strategy might actually be to bring up education generally, or economic disparity, and during this process, teach people how to respect other people.
On second: We likely need good value system to align AI on, otherwise, the only alignment AI would know is probably not to overpower the most powerful human. But that does not seem to be the successful outcome of “aligned AI”. I think there are a few posts recently on this as well.
Third: I have seen many people having this confusion/mixed up: rape play/bdsm is consensual, and the definition of rape is non-consensual. Rape is purely about going against the person’s will. If you view it as murder it might be more comparable, but in this case, it is historically one group on to another group due to biological features and power differences that people cannot easily change on, though also a lot of men to men. In my view, it is worse than murder because it is extreme suffering, and that suffering will carry through the victims’ whole lives, and many may end with suicide anyways.
Otherwise, I am glad to see you thinking about these and open to discussion.
Reminded me of the complex systems chapter from a textbook (Center for AI Safety)
Equity vs equality considerations (https://belonging.berkeley.edu/equity-vs-equality-whats-difference):
What caused the differences in outcome?
Historical factors: should definitely apply equity. Conditioning on history is important and corrective efforts are needed.
Is the desired outcome a human “necessity”?
The definition of necessity may be tricky, or even differ by culture. Generally in the US, if it is something like healthcare, or access to education, should move towards/apply equity.
Really appreciate the post; wondering if you had any thoughts on these since the post was first published? Do think something like RLHF now is an effective enough way?
(Could I also get some advice on why the downvote?)
Relatedly on my first reply to the person’s DM:
On resource scarcity—I also find it difficult for humans to execute altruism without resources, and may be a privilege to even think about these (financially, educationally, health-wise, and information-wise. Would starting with something important but may or may not be the “most” important help? Sometimes I find the process of finding the “most” a bit less rational than people would expect/theorize.
Bias and self-preservation by human nature, but need correction
To expand from there, that is one thing that I am worried about. Our views/beliefs are by nature limited by our experiences, and if humans do not recognize or acknowledge that self-preserving natures, we will be claiming altruism but not actually doing them. It will also prevent us from establishing a really good system that incentivize people to do so.
To give an example, there was a manifold question (https://manifold.markets/Bayesian/would-you-rather-manifold-chooses?r=enlj), asking if manifold wants to “Push someone in front of a trolley to save 5 others (NO) -OR- Be the person someone else pushes in front of the trolley to save 5 others (YES)”. When I saw the stats before close, it was 92% (I chose YES, and lost, I would say my YES is with 85% confidence). While we propose we value people’s life equally, I see self-preservation persist. Then I see that could radiate to I prefer preservation of my family, friends, people you are more familiar with, etc.
Simple majority vs minority situation
This is similar with majority voting. When the population has 80% of people A, and 20% of people B, assuming equal vote/power for everyone, when there are issues that is in conflict between A and B, especially some sort of historical oppression of A towards B, then B’s concerns would rarely be addressed. (Hope humans nowadays are better than this, and more educated, but who knows).
Power
When power (social and economical) is added into the self-preservation, without a proper social contract, this will be even more messed-up. The decision maker of these problems will always the people with some sort of power.
This is true in reality, where the people who get to choose/plan what is the most important in reality and our society, are either the powerful people, or the population outnumbering others. Worse if both. Therefore, we see less passionate caring on minority issues, on a platform like this practicing rationality. With power, everything can go below it, including other people, policies, and law.
How do we create a world with max. freedom to people, with constraint on people do not harm other people, regardless of they have power or not/regardless if they will get caught or not? What level of “constraint” are historically powerful groups willing to accept? I am not sure of right balance/what is most effective yet; I have seen cases where some men were reluctant to receive education materials on consent, claiming they know, but they don’t. This is a very very small cost, but yet there are people who are not willing to even take that. This makes me frustrated.
Regardless, I do believe in the effectiveness of awareness, continuous education, law, and the need to foster a culture to cultivate humans that truly respect other lives.
Real life examples
I was not very aware of many related women’s suffering issues, until recently, from the Korean pop star group
to recent female doctor case (https://www.npr.org/sections/goats-and-soda/2024/08/26/g-s1-18366/rape-murder-doctor-india-sexual-violence-protests). I was believing in 2024, these are things humans should have been able to solve but not yet, but maybe I was in my own bubble. And organizationally, on the country level, I believe you would have seen many news on different wars now.
Thanks for sharing the two pages! Not sure if the above are clear or not, but I try to document all of my thoughts.
Thanks for commenting.
I don’t see the need to remove human properties we cherish to remove/reduce murder nor rape. Maybe war. Rape especially, is never defensive. That is an extremely linear way of connecting these.
Particularly, it is the claim that rape is in any way remotely related to erotic excitement that I found outrageous. That is simply wrong, and is an extremely narrow and perpetrator centered way of thinking. Check out the link that is linked in my comment if you want to learn more on myths about rape.
For the do not care point—my personal belief is that if one believes that those should not be taking up resources to be worked on, even not oneself but also other people, I do not see too much of a difference on with not care from consequentialist point of view/simply practically, or the degree of caring is small/not big enough to agree on allocation of resources/attention, while in my initial message, I have already linked multiple cases recently that has happened. (The person mentioned that working on these are “dangerous distractions”, not for himself, “ourselves”, seems to say society as a whole.)
Finally, the thread post is not on the user alone; it is a bit standalone to reflect on the danger of not having correct awareness or attention on these issues collectively, and how it is related to successful AI alignment as well.
No, some of your points are fine, and it is not you do not embrace my goals, but seems you try to prevent me from pursuing mine, by trying convince me or everyone to work on one thing alone, and to not do anything about war, rape nor murder. (though not related to morality yet). First it is not if you not embrace these goals personally, it is that you downplay them, and seem to argue everyone need to pursue a single goal to prevent AI risks, and I felt pushed even when I have already mentioned in this thread that both needs attention. You have mentioned addressing war, rape, and others are distractions.Even that is up to perspectives and though I disagree, I did not find it completely related to morality yet; it could be due to your lack of relatable experiences perhaps and your demographics, and I wrote a very long explanation explaining why that could happen with people’s experience, groups, and about power, and everything. (pasted those under this comment too). It is even fine to argue, that you think they are hard to solve, but you cannot say, that rape is related to something flourishing in humans. Language matters. I will stand by my words unless you acknowledge this claim/language is wrong/a mistake. Note that there is a difference between “low moral standard” (your language) and “if prevalent, the moral standard that does not lead to successful AI alignment” (my language). I have never accused your entire person/view personally, I am criticizing this view/claim. There are many replies on this thread that promote values that you are aligned with, including yourself and I didn’t have problem with, but this language is over the boundary, and I am not posting that quotation here casually.
=> What made your comment itself wrong, is that you seem to have the horrifying interpretation of what rape is. Rape is NOT erotic excitement, and that is very important, and that claim is outrageous. If you didn’t realize that before, and now you have, it does not hurt to admit that. Moral standards always improve as humans learn and gather more information. As I have linked in my link, it is about abuse and power. I have directly pointed this out multiple times in the message, you still repeatedly say it is a sexual instinct, which I have made clear it is not, nor it means nobody should tackle this, nor “dangerous distraction”.
If you mention emotion or something personal as opposed to arguing on the actual thing—I myself was also extremely bothered and frustrated by the comment on rape is erotic excitement, and could not take this off my mind and was talking to my friends about such claims, mostly identifying with women, and we were all very disappointed/horrified about such claims. Many many criminals use that to make excuses. Do you know how many people suffer, from that one sided “erotic excitement” even for someone there is anything sexual to it at all? There is nothing flourishing about this. Life is short, and I need to try my best to correct these (these kinds of comments are exactly reasons why I am worried, and why I try to raise awareness and educate) even if that means I suffer mentally from it. If you do not wish to be criticized, maybe be careful with how you present your arguments, and what is sensible to say, and what is not. If you are wrong/truly didn’t know/believed the wrong thing, then maybe update your belief.
(This should have been attached/replied to my first quick take, as opposed to a separate quick take.)
From someone’s DM to me:
“Regarding war, rape, and murder, it’s one thing to make them taboo, view them with disgust, and make them illegal. It’s another thing to want to see them disappear from the world entirely. They are side effects of human capacities that in other contexts we cherish, like the willingness to follow powerful leaders, to make sacrifices, to protect ourselves by fighting and punishing enemies who threaten or abuse us, and the ability to feel erotic excitement. This is why they keep happening.”
I am seeing—“side effects”, “rape → erotic excitement”—by going against another human’s will by definition. I found it disappointing to have to explain why this is not the case in the first place by linking https://valleycrisiscenter.org/sexual-assault/myths-about-sexual-assault/#:~:text=Reality%3A%20Rape%20is%20not%20about,him%20accountable%20for%20his%20actions. in 2024.
I am not sure if this person is 1) trying to promote “AI risks are more important than anything else” to emphasis its sole importance (based on other parts from this person’s DM) by promoting that other issues are not issues/dangerous, or 2) truly believe in the above about “unavoidable side effects” of the abuse of power such as rape and see it as one sided erotic excitement based on sufferings from another by definition. They also comment that it is similar to having matches in homes without setting homes on fire—which is a poor analogy as matches are useful, but not rape intentions/intentions to harm other people on offense. I am not sure of this person’s conscious/unconscious motivation, and not sure which one is a better motivation to hope for.
I especially hope this moral standard from this comment is not prevalent among our community—we will likely not going to have properly aligned AI, if we don’t care nor understand war, rape, nor murder, or view them as distractions to work on with any resources collectively. Not only we need to know they are wrong, so are AIs.
Finally, I would encourage posting these on the public comments, rather than DM. The person is aware of this. Be open and transparent.
[edited]
If you look into a bit more history on social justice/equality problems, you would see we have actually made many many progress (https://gcdd.org/images/Reports/us-social-movements-web-timeline.pdf), but not enough as the bar was so low. These also have made changes in our law. Before 1879, women cannot be lawyers (https://en.wikipedia.org/wiki/Timeline_of_women_lawyers_in_the_United_States). On war, I don’t have too much knowledge myself, so I will refrain from commenting for now. It is also my belief that we should not stop at attempt, but attempt is the first step (necessary but not sufficient), and they have pushed to real changes as history shown, but it will have to take piles of piles of work, before a significant change. Just because something is very hard to do, does not mean we should stop, nor there will not be a way (just like ensuring there is humanity in the future.) For example, we should not give up on helping people during war nor try to reduce wars in the first place, and we should not give up on preventing women being raped. In my opinion, this is in a way ensuring there is future, as human may very well be destroyed by other humans, or by mistakes by ourselves. (That’s also why in the AI safety case, governance is so important so that we consider the human piece.)
As you mentioned political party—it is interesting to see surveys happening here; a side track—I believe general equality problems such as “women can go to school”, is not dependent on political party. And something like “police should not kill a black person randomly” should not be supported just by blacks, but also other races (I am not black).
Thanks for the background otherwise.
I meant people on this platform initially, but also good to reflect people generally, which is also where some of my concerns are.
I agree on attending NGOs, and if by you, you meant me. For awareness—most awareness does not involve detailed stories or images, and those events did leave a strong impression/emotion (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2827459/#:~:text=At%20the%20very%20least%20emotions,thought%20and%20away%20from%20others.).
It is a bit sad to see for a lot of humans, even the ones who are already caring about the world generally (which is already a privilege, because many may need to focus on being alive themselves), if something is not relatable, they don’t deeply care. When that is correlated with power, it is the worst.
Thanks for the thoughtful comments first of all.
What I am sensing or seeing right now is in order to promote AGI risks and X risks, I am seeing people downplay/lower ranking the importance of the people who already cannot see this beautiful world. It does not feel logical to me, that because there are always future issues that affect everyone’s lives, the issues that cause some people to be already on the miserable side should not be addressed.
The problem here is that we are not have the same starting line for “everyone” now, and therefore progress towards saving everyone with future focus might not mean the same thing. I maybe should draw a graph to demonstrate my point. As opposed to only consider problems that concerns everyone, I think we should also focus a bit more on an inclusive finish line that is connected to current realities of not the same starting line. If this world continues to be this way, I also worry if the future generations would like it, or would want to be brought into this.
I understand the utilitarian intentions, but I myself also believe we could incorporate equalitarian views. And in fact, a mindset or rules promoting equality or along similar lines actually helps everyone. In many situations a human will be one of those people at some point in their life in some way. Maybe a person’s home suddenly became war zone. Maybe got disabled suddenly. Maybe experienced sexual assault for self or loved one. Establishing a good system to reduce these and prevent these helps human in the future as well. I would like to formalize this a bit more later.
Both views/also current vs future views should really joint forces, as opposed to exclude each other. There are many tasks that I see are shared such as social good mindsets and governance.
Some background about me; myself believe in looking into both, and believe in value in looking into both. It would be dangerous to focus on only one either way by promoting another, and gradually we overlook/go backwards on things that we have started.
That’s good to hear. Any posts you have encountered that are good and mention these/solutions on these?
Do you think people are firmly aware of this in the first place? I would love to hear that the answer is yes.
On solutions—I am not sure yet on universal solution, as it is very dependent on each case. For some problems, some solutions would be around raising awareness and international cooperation on educating human who hurts other human, pushing for law reforms, and alternative sentencing. Solutions aside, I am not sure if I am seeing enough people even care about these. My power alone is not enough, that’s why we need to join force.
I am worried about how people would down vote on this on this platform. I don’t think worrying about long term is bad, or it should not be looked into, but at the same time, it should not be the only thing we care either. This worries me as “we should only work on long term X risks, but nothing about the people now, and any comments that seek to say otherwise is wrong” type of sentiment is what I am seeing more and more.
There is danger in overlooking current risks. Besides obvious reasons on we are ignoring current people, from the future perspective we would be missing the connection between the future and the present, and missing the opportunity to practice applying solutions to reality. And thanks for the link to effective altruism, and I am aware of the initiative/program after attending an introductory program. It feels to me it was merging a couple different directions, and somehow it felt like recent efforts were mostly on long term X risks. For its original goal, I am also not entirely sure if it considers fairness enough, despite an effort to consider scarcity. An EA concept I see some people not understanding enough is the marginal part—marginal dollar, and marginal labor, which should allow people to invest in a portfolio of things. Would welcome any recommended readings further on this.
I saw someone wrote somewhere: will try their best in their life to address X risks to make sure future humans/generations will see this beautiful world. I respect that. However, many recent news and events make me also think about the current humans/people right now. Who are seeing this beautiful world right now? Whose beautiful world this is?
The people in war didn’t see it. It is not their beautiful world. The female doctor who got brutally gang raped while on duty didn’t see it. It is not their beautiful world. Are there also enough efforts to help these people, to see a beautiful and fair world? Is this really a beautiful world? Does this beautiful world belong to everyone?
A recent thought on AI racing—it may not lead to more intelligent models necessarily especially at a time when low hanging fruits are taken and now more advanced breakthroughs need to come from longer term exploration and research. But this also does not necessarily mean that AI racing (particularly on LLMs in this context, but I think generally too) is not something to be worried about. It may waste a lot of compute/resources to achieve only marginally better models. Additionally the worst side effect of AI racing to me is the potential negligence on safety mitigations, and lack of safety focused mindset/culture.