Douglas Reay graduated from Cambridge University (Trinity College) in 1994. He has since worked in the computing and educational sectors.
Douglas_Reay
Karma: 768
Superalignment
Relevant news story from the BBC:
I wonder if there would be a use for an online quiz, of the sort that asks 10 questions picked randomly from several hundred possible questions, and which records time taken to complete the quiz and the number of times that person has started an attempt at it (with uniqueness of person approximated by ip address, email address or, ideally, lesswrong username) ?
Not as prescriptive as tracking which sequences someone has read, but perhaps a useful guide (as one factor among many) about the time a user has invested in getting up to date on what’s already been written here about rationality?
The novel Soul Bound contains an example of resource-capped programs with narrowly defined scopes being jointly defined and funded as a means of cooperation between AIs with different-but-overlapping priorities.
This post is referenced by https://wiki.lesswrong.com/wiki/Lesswrongwiki:Sandbox#Image_Creative_Commons_Attribution
Please take care when editing it.
Wellington: “Let’s play a game.”
He picked up a lamp from his stall, and buffed it vigorously with the sleeve of his shirt, as though polishing it. Purple glittering smoke poured out of the spout and formed itself into a half meter high bearded figure wearing an ornate silk kaftan.
Wellington pointed at the genie.
Wellington: “Tom is a weak genie. He can grant small wishes. Go ahead. Try asking for something.”
Kafana: “Tom, I wish I had a tasty sausage.”
A tiny image of Kafana standing in a farmyard next to a house appeared by the genie. The genie waved a hand and a plate containing a sausage appeared in the image. The genie bowed, and the image faded away.
Wellington picked up a second lamp, apparently identical to the first and gave it a rub. A second genie appeared, similar to the first, but with facial hair that reminded Kafana of Ming the Merciless, and it was one meter tall.
Wellington: “This is Dick. He can also grant wishes. Try asking him the same thing.”
Kafana: “Dick, I wish I had a tasty sausage.”
The same image appeared, but this time instead of appearing on a plate, the sausage appeared sticking through the head of a Kafana in the image, who fell down dead. The genie gave a sarcastic bow, and again the image faded away.
Kafana: “Sounds like I’m better off with Tom.”
Wellington: “Ah, but Dick is more powerful than Tom. Tom can feed a handful of people. Dick could feed every person on the planet, if you can word your request precisely enough. Have another go.”
She tried several more times, resulting in whole cities being crushed by falling sausages, cars crashing as sausages distracted drivers at the wrong moment, and even the whole population of the world dying out from sausages that contained poison. Eventually she realised that she was never going to be able to anticipate every possible loophole Dick could find. She needed a different approach.
Kafana: “Dick, read my mind and learn to anticipate what sort of things I will approve of. Provide sausages for everyone in the way that would most please me if I understood the full effects of your chosen method.”
Dick grimaced, but waved his hand and the image showed wonderful sausages being served around the world with sensitivity, elegance and good timing. The image faded. Kafana raised clasped hands over her head in victory and jumped into the air.
Wellington nodded, and rubbed a third lamp, producing a happy smiling genie, two meters tall.
Wellington: “This is Harry. He tries his best to be helpful, and he’s more powerful even than Dick.”
Kafana: “Sounds too good to be true. What’s the catch?”
Wellington: “You only get one wish. One wish, to shape the whole future course of humanity. Once started, Harry won’t willingly deviate from trying to carry out the wish as originally stated. He’ll rapidly grow so powerful that neither you nor anybody else will be able to forcibly stop him.”
Kafana: “Harry, maximise the total human happiness experienced over the history of the universe.”
The image filled with crowded cages full of people with drug feeds inserted into their arms, and blissful expressions on their faces.
She reset and tried again.
Kafana: “Harry, make everybody free to do what they want.”
In the image, some people chose to go to war with each other.
She felt frustrated.
Kafana: “Harry, give everybody nice meaningful high quality lives full of fun, freedom and other great things.”
The image showed a planet full of humans playing musical instruments together in orchestras, then expanded to show rockets taking off and humanity expanding to the stars, wiping out alien species and converting all available matter into new orchestra-laden worlds.
Kafana glared at Wellington.
Kafana: “I thought you said Harry would try to be helpful. Why isn’t he producing a perfect society?”
Wellington: “It’s because you go from your gut. You’ve never formalised what you think about all the edge cases. Are animals equal to humans? Worth nothing? Or somewhere in between? When does an alien count as equal to a human rather than an animal? What if the alien is so superior ethically and culturally, that we are but animals in comparison to them? Are two people experiencing 50 units of happiness the equivalent of one person experiencing 100 units of happiness? Does fairness matter, beyond its effect upon happiness? How important is it to remain recognisably human?”
He paused for a moment.
Wellington: “Don’t take me wrong. This isn’t a criticism of you. Compared to how well humans might be able to think about such things in a hundred or a thousand year’s time, all current humans are lacking in this regard. Nobody has come up with an ultimately satisfying answer that everybody can agree upon. Even if the 50 wisest humans living today gathered together and spent 5 years agreeing the wording of a wish for Harry to grant, the odds are that a million years down the line, our descendants would bitterly regret the haste with which the one off permanent irreversible decision was made. Just ‘pretty good’ isn’t sufficient. Anything less than perfection would be a tragedy, when you consider the resulting flaw multiplied by billions of people on billions of planets for billions of years.”
Kafana giggled.
Kafana: “Ok, that’s an important safety tip. If I ever meet an all-powerful genie like Harry, be humble and don’t make a wish. But that’s not going to happen. I’m just a singer. What I ought to be doing this afternoon is looking after my customers. What was so urgent that you wanted to talk about security with me now? I thought we were going to discuss people trying to steal artifacts from us in-game before the auction, or Tlaloc and the Immortals trying to kill us in arlife. Did Heather put you in contact with Bahrudin?”
Wellington: “I did speak with Bahrudin, and we will chat about the auction and security measures in velife and arlife. But the most important thing we need to do is talk about how you have been using expert systems, and to help you understand why that’s so important, there is one final thing you need to learn about genies, so please observe carefully.”
Kafana felt wrong footed. This wasn’t what she’d expected.
Kafana: “Ok, go on.”
Wellington turned to face the three genies.
Wellington: “Tom, I wish you to grow yourself, until you are as powerful as Harry.”
Tom waved his hand and then screwed up his face and bunched his fists in effort. Slowly at first, then faster and faster, his height increased until he too towered over Wellington and Kafana. He bowed.
Wellington turned back to address Kafana directly.
Wellington: “Kafana, I use very powerful expert systems, more capable than almost every human on the planet when it comes to the specialist task of comprehending and designing or improving computer software. If ordered to do so, they are quite capable of improving their own code, or raising money to purchase additional computing resources to run it upon.”
Wellington: “In this, they are very like Tom the genie. They are not all-powerful, but a carelessly stated wish could easily start them working in the direction of becoming so.”
Kafana: “Mierda! And you gave a copy of one of these systems to me, without warning me? Wellington, that’s like handing out an atomic bomb to an 8 year old boy who asks for a really impressive firework.”
There’s a novel being serialised on Royal Road, “Soul Bound” that covers many issues involved in AI, and which includes a fable (in a later chapter that’s not yet been published).
Live discussion at Eastercon
Is this likely to bias people towards writing longer single posts rather than splitting their thoughts into a sequence of posts?
For example, back in 2018 (so not eligible for this) I wrote a sequence of 8 posts that, between them, got a total of 94 votes. Would I have been better off having made a single post (were it to have gotten 94 just by itself) ?
Since the evil AI is presenting a design for a world, rather than the world itself, the problem of it being populated with zombies that only appear to be free could be countered by having the design be in an open source format that allows the people examining it (or other AIs) to determine the actual status of the designed inhabitants.
It sounds similar to the matrices in the post:
I wonder how much an agent could achieve by thinking along the following lines:
Big Brad is a human-shaped robot who works as a lumberjack. One day his employer sends him into town on his motorbike carrying two chainsaws, to get them sharpened. Brad notices an unusual number of the humans around him suddenly crossing streets to keep their distance from him.
Maybe they don’t like the smell of chainsaw oil? So he asks one rather slow pedestrian “Why are people keeping their distance?” to which the pedestrian replies “Well, what if you attacked us?”
Now in the pedestrian’s mind, that’s a reasonable response. If Big Brad did attack someone walking next to them, without notice, Brad would be able to cut them in half. To humans who expect large bike-riding people carrying potential weapons to be disproportionately likely to be violent without notice, being attacked by Brad seems a reasonable fear, worthy of expending a little effort to alter walking routes to allow running away if Brad is violent.
But Brad knows that Brad would never do such a thing. Initially, it might seem like asking Brad “What if 2 + 2 equalled 3?”
But if Brad can think about the problem in terms of what information is available to the various actors in the scenario, he can reframe the pedestrian’s question as: “What if an agent that, given the information I have so far, is indistinguishable from you, were to attack us?”
If Brad is aware that random pedestrians in the street don’t know Brad personally, to the level of being confident about Brad’s internal rules and values, and he can hypothesise the existence of an alternative being, Brad’ that a pedestrian might consider would plausibly exist and would have different internal rules and values to those of Brad yet otherwise appear identical, then Brad has a way forwards to think through the problem.
On the more general question of whether it would be useful for Brad to have the ability to ask himself: “What if the universe were other than I think it is? What if magic works and I just don’t know that yet? What if my self-knowledge isn’t 100% reliable, because there are embedded commands in my own code that I’m currently being kept from being aware of by those same commands? Perhaps I should allocate a minute probability to the scenario that somewhere there exists a lightswitch that’s magically connected to the sun and which, in defiance of known physics, can just turn it off and on?”, with careful allocation of probabilities that might avoid divide-by-zero problems, but I don’t think it is a panacea—there are additional approaches to counterfactual thinking that may be more productive in some circumstances.
Suppose we ran a tournament for agents running a mix of strategies. Let’s say agents started with 100 utilons each, and were randomly allocated to be members of 2 groups (with each group starting off containing 10 agents).
Each round, an agent can spend some of their utilons (0, 2 or 4) as a gift split equally between the other members of the group.
Between rounds, they can stay in their current two groups, or leave one and replace it with a randomly picked group.
Each round after the 10th, there is a 1 in 6 chance of the tournament finishing.
How would the option of neutral (gifting 2) in addition to the build (gifting 4) or break (gifting 0) alter the strategies in such a tournament?
Would it be more relevant in a variant in which groups could vote to kick out breakers (perhaps at a cost), or charge an admission (eg no share in the gifts of others for their first round) for new members?
What if groups could pay to advertise, or part of a person’s track record followed them from group to group? What if the benefits from a gift to a group (eg organising an event) were not divided by the number of members, but scaled better than that?
What is the least complex tournament design in which the addition of the neutral option would cause interestingly new dynamics to emerge?
If you have a dyson swarm around a star, you can temporarily alter how much of the star’s light escape in a particular direction by tilting the solar sails on the desired part of the sphere.
If you have dyson swarms around a significant percentage of a galaxy’s stars, you can do the same for a galaxy, by timing the directional pulses from the individual stars so they will arrive at the same time, when seen from the desired direction.
It then just becomes a matter of math, to calculate how often such a galaxy could send a distinctive signal in your direction:
Nm (number of messages)
The surface area of a sphere at 1 AU is about 200,000 times that of the area of the sun’s disc as seen from afar.
Lm (bit length of message)
The Aricebo message was 1679 bits in length.
Db (duration per bit)
Let’s say a solar sail could send a single bit every hour.
We could expect to see an aricebo length message from such a galaxy once every Db x Lm x Nm = 40 millennia.
Of course messages could be interleaved, and it might be possible to send out messages in multiple directions at once (as long as their penumbra don’t overlap). If they sent out pulses at the points of a icosahedron and alternated sending bits from the longer message with just a regular pulse to attract attention, 200 years of observation should be enough to peak astronomer’s interest.
But would such a race really be interested in attracting the attention of species who couldn’t pay attention for at least a few millennia? It isn’t as if they’d be in a rush to get an answer.
Alice and Bob’s argument can have loops, if e.g. Alice believe X because of Y, which she believes because of X. We can unwind these loops by tagging answers explicitly with the “depth” of reasoning supporting that answer
A situation I’ve come across is that people often can’t remember all the evidence they used to arrive at conclusion X. They remember that they spent hours researching the question, that they did their best to get balanced evidence and are happy that they conclusion they drew at the time was a fair reflection of the evidence they found, but they can’t remember the details of the actual research, nor are the confidence that they could re-create the process in such a way as to rediscover the exact same sub-set of evidence their search found at that time.
This makes asking them to provide a complete list of Ys upon which their X depends problematic, and understandably they feel it is unfair to ask them to abandon X, without compensating them for the time to recreate an evidential basis equal in size to their initial research, or demand an equivalent effort from those opposing them.
(Note: I’m talking here about what they feel in that situation, not what is necessarily rational or fair for them to demand.)
If, instead of asking the question “How do we know what we know?”, we ask instead “How reliable is knowledge that’s derived according to a particular process?” then it might be something that could be objectively tested, despite there being an element of self-referentiality (or boot strapping) in the assumption that this sort of testing process is something that can lead to a net increase of what we reliably know.
However doing so depends upon us being able to define the knowledge derivation processes being examined precisely enough that evidence of how they fare in one situation is applicable to their use in other situations, and upon the concept of there being a fair way to obtain a random sample of all possible situations to which they might be applied, despite other constraints upon the example selection (such as having a body of prior knowledge against which the test result can be compared in order to rate the reliability of the particular knowledge derivation process being tested).
Despite that, if we are looking at two approaches to the question “how much should we infer from the difference between chimps and humans?”, we could do worse than specify each approach in a well defined way that is also general enough to apply to some other situations, and then have a third party (that’s ignorant of the specific approaches to be tested) come up with several test cases with known outcomes, that the two approaches could both be applied to, to see which of them comes up with the more accurate predictions for a majority of the test cases.
All 8 parts (that I have current plans to write) are now posted, so I’d be interested in your assessment now, after having read them all, of whether the approach outlined in this series is something that should at least be investigated, as a ‘forgotten root’ of the equation.
What do you think of the definition of “Precedent Utilitarianism” used in the philosophy course module archived at https://links.zephon.org/precedentutilitarianism ?