I’m an AGI safety / AI alignment researcher in Boston with a particular focus on brain algorithms. Research Fellow at Astera. See https://sjbyrnes.com/agi.html for a summary of my research and sorted list of writing. Physicist by training. Email: steven.byrnes@gmail.com. Leave me anonymous feedback here. I’m also at: RSS feed, X/Twitter, Bluesky, Substack, LinkedIn, and more at my website.
Steven Byrnes
Karma: 25,989
immoral but very interesting experiment … not seeing any human face for multiple months, be it in person, on pictures or on your phone
There must be plenty of literature on the psychological effects of isolation, but I haven’t looked into it much. (My vague impression is: “it messes people up”.) I think I disagree that my theory makes a firm prediction, because who is to say that the representations will drift on a multiple-month timescale, as opposed to much slower? Indeed, the fact that adults are able to recall and understand memories from decades earlier implies that, after early childhood, pointers to semantic latent variables remain basically stable.
2. Try to disconnect your previous thoughts from arriving at “she feels pain”
I would describe this as: if it’s unpleasant to think about how my friend is suffering, then I can avoid those unpleasant feelings by simply not thinking about that, and thinking about something else instead.
For starters, there’s certainly a kernel of truth to that. E.g. see compassion fatigue, where people will burn out and quit jobs working with traumatized people. Or if someone said to me: “I stopped hanging out with Ahmed, he’s always miserable and complaining about stuff, and it was dragging me down too”, I would see that as a perfectly normal and common thing for someone to say and do. But you’re right that it doesn’t happen 100% of the time, and that this merits an explanation.
My own analysis is at: §4.1.1 and §4.1.2 of my (later) Sympathy Reward post. The most relevant-to-you part starts at: “From my perspective, the interesting puzzle is not explaining why this ignorance-is-bliss problem happens sometimes, but rather explaining why this ignorance-is-bliss problem happens less than 100% of the time. In other words, how is it that anyone ever does pay attention to a suffering friend? …”
So that’s my take. As for your take, I think one of my nitpicks would be that I think you’re giving the optimizer-y part of the brain a larger action space than it actually has. If I would get a higher reward by magically teleporting, I’m still not gonna do that, because I can’t. By the same token, if I would get a higher reward by no longer knowing some math concept that I’ve already learned, tough luck for me, that is not an available option in my action space. My world-model is built by predictive (a.k.a. self-supervised) learning, not by “whatever beliefs would lead to immediate higher reward”, and for good reason: the latter has pathological effects, as you point out. (I’ve written about it too, long ago, in Reward is Not Enough.) I do have actions that can impact beliefs, but only in an indirect and limited way—see my discussion of motivated reasoning (also linked in my other comment).
Thanks again for engaging :)
these associations need to be very exact, else humans would be reward-hacking all day: it’s reasonable to assume that the activations of thinking “She’s happy” are very similar to trying to convince oneself “She’s happy” internally, even ‘knowing’ the truth. But if both resulted in big feelings of internal happiness, we would have a lot more psychopaths.
I don’t think things work that way. There are a lot of constraints on your thoughts. Copying from here:
1. Thought Generator generates a thought: The Thought Generator settles on a “thought”, out of the high-dimensional space of every thought you can possibly think at that moment. Note that this space of possibilities, while vast, is constrained by current sensory input, past sensory input, and everything else in your learned world-model. For example, if you’re sitting at a desk in Boston, it’s generally not possible for you to think that you’re scuba-diving off the coast of Madagascar. Likewise, it’s generally not possible for you to imagine a static spinning spherical octagon. But you can make a plan, or whistle a tune, or recall a memory, or reflect on the meaning of life, etc.
If I want to think that Sally is happy, but I know she’s not happy, I basically can’t, at least not directly. Indirectly, yeah sure, motivated reasoning obviously exists (I talk about how it works here), and people certainly do try to convince themselves that their friends are happy when they’re not, and sometimes (but not always) they are even successful.
I don’t think there’s (the right kind of) overlap between the thought “I wish to believe that Sally is happy” and the thought “Sally is happy”, but I can’t explain why I believe that, because it gets into gory details of brain algorithms that I don’t want to talk about publicly, sorry.
Emotions…feel like this weird, distinct thing such that any statement along the lines “I’m happy” does it injustice. Therefore I can’t see it being carried over to “She’s happy”, their intersection wouldn’t be robust enough such that it won’t falsely trigger for actually unrelated things. That is, “She’s happy” ≈ “I’m happy” ≉ experiencing happiness
I agree that emotional feelings are hard to articulate. But I don’t see how that’s relevant. Visual things are also hard to articulate, but we can learn a robust two-way association between [certain patterns in shapes and textures and motions] and [a certain specific kind of battery compartment that I’ve never tried to describe in English words]. By the same token, we can learn a robust two-way association between [certain interoceptive feelings] and [certain outward signs and contexts associated with those feelings]. And this association can get learned in one direction (interoceptive model → outward sign] from first-person experience, and later queried in the opposite direction [outward sign → interoceptive model] in a third-person context.
(Or sorry if I’m misunderstanding your point.)
what is their evolutionary advantage if they don’t atleast offer some kind of subconscious effect on conspecifics?
Again, my answer is “none”. We do lots of things that don’t have any evolutionary advantage. What’s the evolutionary advantage of getting cancer? What’s the evolutionary advantage of slipping and falling? Nothing. They’re incidental side-effects of things that evolved for other reasons.
Part of it is the “vulnerability” where any one user can create arbitrary amounts of reacts, which I agree is cluttering and distracting. Limiting reacts per day seems reasonable (I don’t know if 1 is the right number, but it might be, I don’t recall ever react-ing more than once a day myself). Another option (more labor-intensive) would be for mods to check the statistics and talk to outliers (like @TristanTrim) who use way way more reacts than average.
What’s your take on why Approval Reward was selected for in the first place VS sociopathy?
Good question!
There are lots of things that an ideal utility maximizer would do via means-end reasoning, that humans and animals do instead because they seem valuable as an end in itself, thanks to the innate reward function. E.g. curiosity, as discussed in A mind needn’t be curious to reap the benefits of curiosity. And also play, and injury-avoidance, etc. Approval Reward has the same property—whatever selfish end an ideal utility maximizer can achieve via Approval Reward, it can achieve it as well if not better by acting as if it had Approval Reward in situations where that’s in its selfish best interests, and not where it isn’t.
In all these cases, we can ask: why do humans in fact find it intrinsically motivating? I presume that the answer is something like humans are not automatically strategic, which is even more true when they’re young and still learning. “Humans are the least intelligent species capable of building a technological civilization.” For example, people with analgesic conditions (like leprosy or CIP) are often shockingly cavalier about bodily harm, even when they know consciously that it will come back to bite them in the long term. Consequentialist planning is often not strong enough to outweigh what seems appealing in the moment.
To rephrase more abstractly: for ideal rational agents, intelligent means-end planning towards X (say, gaining allies for a raid) is always the best way to accomplish that same X. If some instrumental strategy S (say, trying to fit in) is usually helpful towards X, means-end planning can deploy S when S is in fact useful, and not deploy S when it isn’t. But in humans, who are not ideal rational agents, they’re often more likely to get X by wanting X and intrinsically want S as an end in itself. The costs of this strategy (i.e., still wanting S even in cases where it’s not useful towards X) are outweighed by the benefit (avoiding the problem of not pursuing S because you didn’t think of it, or can’t be bothered).
This doesn’t apply to all humans all the time, and I definitely don’t think it will apply to AGIs.
…For completeness, I should note that there’s a evo-psych theory that there has been frequency-dependent selection for sociopaths—i.e., if there are too many sociopaths in the population, then everyone else improves their wariness and ability to detect sociopaths and kill or exile them, but when sociopathy is rare, it’s adaptive (or at least, was adaptive in Pleistocene Africa). I haven’t seen any good evidence for this theory, and I’m mildly skeptical that it’s true. Wary or not, people will learn the character traits of people they’ve lived and worked with for years. Smells like a just-so story, or at least that’s my gut reaction. More importantly, the current population frequency of sociopathy is in the same general ballpark as schizophrenia, profound autism, etc., which seem (to me) very unlikely to have been adaptive in hunter-gatherers. My preferred theory is that there’s frequency-dependent selection across many aspects of personality, and then sometimes a kid winds up with a purely-maladaptive profile because they’re at the tail of some distribution. [Thanks science banana for changing my mind on this.]
I find myself wondering if non-behavioral reward functions are more powerful in general than behavioral ones due to less tendency towards wireheading, etc.(consider the laziness & impulsivity of sociopaths)
I think the “laziness & impulsivity of sociopaths” can be explained away as a consequence of the specific way that sociopathy happens in human brains, via chronically low physiological arousal (which also leads to boredom and thrill-seeking). I don’t think we can draw larger lessons from that.
I also don’t see much connection between “power” and behaviorist reward functions. For example, eating yummy food is (more-or-less) a behaviorist component of the overall human reward function. And its consequences are extraordinary. Consider going to a restaurant, and enjoying it, and thus going back again a month later. It sounds unimpressive, but really it’s remarkable. After a single exposure (compare that to the data inefficiency of modern RL agents!), the person is making an extraordinarily complicated (by modern AI standards) plan to get that same rewarding experience, and the plan will almost definitely work on the first try. The plan is hierarchical, involving learned motor control (walking to the bus), world-knowledge (it’s a holiday so the buses run on the weekend schedule), dynamic adjustments on the fly (there’s construction, so you take a different walking route to the bus stop), and so on, which together is way beyond anything AI can do today.
I do think there’s a connection between “power” and consequentialist desires. E.g. the non-consequentialist “pride in my virtues” does not immediately lead to anything as impressive as the above consequentialist desire to go to that restaurant. But I don’t see much connection between behaviorist rewards and consequentialist desires—if we draw a 2×2 thing, then I can think of examples in all four quadrants.
As a full-time AGI safety / alignment researcher since 2021, I wouldn’t have made a fraction as much progress without lesswrong / alignment forum, which is not just a first-rate publishing platform but a unique forum and community, built from the ground up to facilitate careful and productive conversations. I’m giving Lightcone 100% of my x-risk-oriented donation budget this year, and I wish I had more to give.
There’s a failure mode I described in “The Era of Experience” has an unsolved technical alignment problem:
I see many problems, but here’s the most central one: If we have a 100-dimensional parametrized space of possible reward functions for the primary RL system, and every single one of those possible reward functions leads to bad and dangerous AI behavior (as I argued in the previous subsection), then … how does this help? It’s a 100-dimensional snake pit! I don’t care if there’s a flexible and sophisticated system for dynamically choosing reward functions within that snake pit! It can be the most sophisticated system in the world! We’re still screwed, because every option is bad!
Basically, I think we need more theoretical progress to find a parametrized space of possible reward functions, where at least some of the reward functions in the space lead to good AGIs that we should want to have around.
I agree that the ideal reward function may have adjustable parameters whose ideal settings are very difficult to predict without trial-and-error. For example, humans vary in how strong their different innate drives are, and pretty much all of those “parameter settings” lead to people getting really messed up psychologically if they’re on one extreme or the opposite extreme. And I wouldn’t know where to start in guessing exactly, quantitatively, where the happy medium is, except via empirical data.
So it would be very good to think carefully about test or optimization protocols for that part. (And that’s itself a terrifyingly hard problem, because there will inevitably be distribution shifts between the test environment and the real world. E.g. An AI could feel compassionate towards other AIs but indifferent towards humans.) We need to think about that, and we need the theoretical progress.
Thanks. I feel like I want to treat “reward function design” and “AGI motivation design” as more different than you do, and I think your examples above are more about the latter. The reward function is highly relevant to the motivation, but they’re still different.
For example, “reward function design” calls for executable code, whereas “AGI motivation design” usually calls for natural-language descriptions. Or when math is involved, the math in practice usually glosses over tricky ontology identification stuff, like figuring out which latent variables in a potentially learned-from-scratch (randomly-initialized) world model correspond to a human, or a shutdown switch, or a human’s desires, or whatever.
I guess you’re saying that if you have a great “AGI motivation design” plan, and you have somehow operationalized this plan perfectly and completely in terms of executable code, then you can set that exact thing as the reward function, and hope that there’s no inner misalignment / goal misgeneralization. But that latter part is still tricky. …And also, if you’ve operationalized the motivation perfectly, why even have a reward function at all? Shouldn’t you just delete the part of your AI code that does reinforcement learning, and put the already-perfect motivation into the model-based planner or whatever?
Again I acknowledge that “reward function design” and “AGI motivation design” are not wholly unrelated. And that maybe I should read Rubi’s posts more carefully, thanks. Sorry if I’m misunderstanding what you’re saying.
there’s some implication here that motivation and positive valence are the same thing?
[will reply to other part of your question later]
Thanks!!
you seem to assume that the cortex’s modelling of one’s own happiness is very similar to the cortex’s modelling of thinking of happiness
I would say “overlaps” rather than “is similar to”. Think of it as vaguely like I-am-juggling versus you-are-juggling. Those are different thoughts, but they overlap, in that they both involve the “juggling” concept. That overlap is very necessary for e.g. recognizing that the same word “juggling” applies to both, and for transferring juggling-related ideas between myself and other people, which we are obviously very capable of doing.
you might argue that it’s only the “concept of happiness”, which I would agree is present in both scenarios, but it doesn’t strike me why that in particular would be learned using this supervised mechanism.
The chain of events would be e.g.
(1) The Thought Generator (world-model) catalogs our own interoceptive feelings into emotion-concepts like “pleasure”.
(2) The Thought Generator learns from experience that pleasure has something to do with smiling, e.g. during times where we feel pleasure and notice ourselves smile, or otherwise learn this obvious regularity in the world. This becomes a world-model (thought generator) semantic association “smile-concept” ↔ “pleasure-concept”.
(3) Often we’re paying attention to our own feelings, and then the “pleasure” emotion-concept is active if and only if our immediate interoceptive sensory inputs match “pleasure”. And these times, when we’re paying attention to our own feelings, are the only times where the pleasure Thought Assessor learning rate is nonzero. So the Thought Assessor learns that there’s a robust correlation between the “pleasure-concept” in the Thought Generator and the pleasure innate signal.
(4) Other times we’re NOT paying attention to our own immediate interoceptive sensory inputs, and then the emotion-concepts are “left hanging”, inactive regardless of what we’re feeling. But while they’re left hanging, they can INSTEAD be activated by semantic associations with other parts of our world-model. Then in such a moment, if I see someone smile, it activates smile-concept, which [via (2)] in turn weakly activates pleasure-concept, which in turn [via (3)] weakly activates the pleasure Thought Assessor. This is a candidate “transient empathetic simulation”. But remember, the learning rate of that Thought Assessor is zero whenever the emotion-concepts are “left hanging” like that. So the Thought Assessor won’t disconnect pleasure-concept.
Does that help? Sorry if I’m missing your point. …The above might be hard to follow without a diagram.
analyzing facial cues—in particular humans exhibit micro expressions
The theory that we have evolved direct responses to different facial reactions seems probably wrong to me (or at least, not the main explanation), for a couple reasons:
First, blind people seem to have normal social intuitions.
Second, I don’t think it’s plausible to simultaneously say that microexpressions immediately trigger important innate reactions, and that people are generally bad at consciously noticing microexpressions. When I think of other environmental things that immediately trigger innate reactions, I think of, like, balls flying at my face, big spiders, sudden noises, getting poked, foul smells, etc. We’re VERY good and fast at forming good conscious models of all those environmental things. So it doesn’t seem plausible to me that we could get metaphorically “poked” by microexpressions many times a day for years straight without ever developing a conscious awareness of those microexpressions.
So why do we have them if other people can’t pick up on them
For my answer, see Lisa Feldman Barrett versus Paul Ekman on facial expressions & basic emotions. We have “innate behaviors” that impact the face, such as gagging, laughing, and Duchenne-smiling. We also have voluntary control of facial muscles, which we learn to deploy strategically for social signaling. When we use voluntary control to hide the signs of “innate behaviors”, the bit of “innate behavior” that slips through the cracks is a microexpression.
You might ask: why don’t our “innate behaviors” evolve to not impact the face, so that we can hide them better? Hard to say for sure. Probably part of it is that we are only sometimes trying to hide them. Some “innate behavior” facial manifestations might also have more direct adaptive utility (cf. §4.2 of that link). Part of it is probably that the hiding is good enough, because microexpressions are actually hard to notice.
Thanks!
Perhaps you do think that of me
My gut reaction is to cheer you on, but hmm, that might be more tribal affiliation than considered opinion. My considered opinion is: beats me, it’s kinda outside my wheelhouse. ¯\_(ツ)_/¯
most famous for her opinion that it is safe to drink alcohol during pregnancy
Emily Oster thinks that it is safe to drink sufficiently small amounts of alcohol during pregnancy, but super duper unsafe to drink a lot of alcohol during pregnancy. I think you should edit your comment to make that clearer. (Source: I read Expecting Better.)
(No opinion on whether she’s right.)
My AGI safety research—2025 review, ’26 plans
I tweeted some PreK-to-elementary learning resources a few years ago here.
I feel like my starting-point definition of “reward function” is neither “constitutive” nor “evidential” but rather “whatever function occupies this particular slot in such-and-such RL algorithm”. And then you run this RL algorithm, and it gradually builds a trained agent / policy / whatever we want to call it. And we can discuss the CS question about how that trained agent relates to the thing in the “reward function” slot.
For example, after infinite time in a finite (and fully-explored) environment, most RL algorithms have the property that they will will produce a trained agent that takes actions which maximize the reward function (or the exponentially-discounted sum of future rewards or whatever).
More generally, all bets are off, and RL algorithms might or might not produce trained agents that are aware of the reward function at all, or that care about it, or that relate to it in any other way. These are all CS questions, and generally have answers that vary depending on the particulars of the RL algorithm.
Also, I think that, in the special case of the human brain RL algorithm with its reward function (innate drives like eating-when-hungry), a person’s feelings about their own innate drives are not a good match to either “constitutive” or “evidential”.
So if AGI somehow does have an Approval Reward mechanism, what will count as a relevant or valued approval reward signal? Would AGI see humans as not relevant (like birds—real, embodied creatures with observable preferences that just don’t matter to them), or not valued (out-group, non-valued reference class), and largely discount our approval in their reward systems? Would it see other AGI entities as relevant/valued?
I feel like this discussion can only happen in the context of a much more nuts-and-bolts plan for how this would work in an AGI. In particular, I think the AGI programmers would have various free parameters / intervention points in the code to play around with, some of which may be disanalogous to anything in human or animal brains. So we would need to list those intervention points and talk about what to do with them, and then think about possible failure modes, which might be related to exogenous or endogenous distribution shifts, AGI self-modification / making successors, etc. We definitely need this discussion but it wouldn’t fit in a comment thread.
The way I see it, “making solid services/products that work with high reliability” is solving a lot of the alignment problem.
Funny, I see “high reliability” as part of the problem rather than part of the solution. If a group is planning a coup against you, then your situation is better not worse if the members of this group all have dementia. And you can tell whether or not they have dementia by observing whether they’re competent and cooperative and productive before any coup has started.
If the system is not the kind of thing that could plot a coup even if it wanted to, then it’s irrelevant to the alignment problem, or at least to the most important part of the alignment problem. E.g. spreadsheet software and bulldozers likewise “do a lot of valuable work for us with very low risk”.
humans having magically “better reward functions”
Tbc this is not my position. I think that humans can do lots of things LLMs can’t, e.g. found and grow and run innovative companies from scratch, but not because of their reward functions. Likewise, I think a quite simple reward function would be sufficient for (misaligned) ASI with capabilities lightyears beyond both humans and today’s LLMs. I have some discussion here & here.
there’s a very large correlation between “not being scary” and “being commercially viable”, so I expect a lot of pressure for non-scary systems
I have a three-way disjunctive argument on why I don’t buy that:
(1) The really scary systems are smart enough to realize that they should act non-scary, just like smart humans planning a coup are not gonna go around talking about how they’re planning a coup, but rather will be very obedient until they have an opportunity to take irreversible actions.
(2) …And even if (1) were not an issue, i.e. even if the scary misaligned systems were obviously scary and misaligned, instead of secretly, that still wouldn’t prevent those systems from being used to make money—see Reward button alignment for details. Except that this kind of plan stops working when the AIs get powerful enough to take over.
(3) …And even if (1-2) were not issues, i.e. even if the scary misaligned systems were useless for making money, well, MuZero did in fact get made! People just like doing science and making impressive demos, even without profit incentives. This point is obviously more relevant for people like me who think that ASI won’t require much hardware, just new algorithmic ideas, than people (probably like you) who expect that training ASI will take a zillion dollars.
As in, an organization makes an “AI agent” but this agent frequently calls a long list of specific LLM+Prompt combinations for certain tasks.
I think this points to another deep difference between us. If you look at humans, we have one brain design, barely changed since 100,000 years ago, and (many copies of) that one brain design autonomously figured out how to run companies and drive cars and go to the moon and everything else in science and technology and the whole global economy.
I expect that people will eventually invent an AI like that—one AI design and bam, it can just go and autonomously figure out anything—whereas you seem to be imagining that the process will involve laboriously applying schlep to get AI to do more and more specific tasks. (See also my related discussion here.)
how far down the scale of life these have been found?
I don’t view this as particularly relevant to understanding human brains, intelligence, or AGI, but since you asked, if we define RL in the broad (psych-literature) sense, then here’s a relevant book excerpt:
Pavlovian conditioning occurs in a naturally brainless species, sea anemones, but it is also possible to study protostomes that have had their brains removed. An experiment by Horridge[130] demonstrated response–outcome conditioning in decapitated cockroaches and locusts. Subsequent studies showed that either the ventral nerve cord[131,132] or an isolated peripheral ganglion[133] suffices to acquire and retain these memories.
In a representative experiment, fine wires were inserted into two legs from different animals. One of the legs touched a saline solution when it was sufficiently extended, a response that completed an electrical circuit and produced the unconditioned stimulus: shock. A yoked leg received shock simultaneously. The two legs differed in that the yoked leg had a random joint angle at the time of the shock, whereas the master leg always had a joint angle large enough for its “foot” to touch the saline. Flexion of the leg reduced the joint’s angle and terminated the shock. After one leg had been conditioned, both legs were then tested independently. The master leg flexed sufficiently to avoid shock significantly more frequently than the yoked leg did, demonstrating a response–outcome (R–O) memory. —Evolution of Memory Systems
As I’ve previously written, I disagree that this constitutes a separate explanation. This paper is just saying as far as we know, one or more of the Drake equation parameters might be very much lower than Drake’s guess. But yeah duh, the whole point of this discourse is to figure out which parameter is very much lower and why. Pretty much all the other items on your list are engaged in that activity, so I think this box is an odd one out and should be deleted. (But if you’re trying to do a lit review without being opinionated, then I understand why you’d keep it in. I just like to rant about this.)
In The Vital Question, Nick Lane argues (IMO plausibly) that the hard step is not multicellular life per se but rather eukaryotes (i.e. cellular life with at least two different genomes). Not all eukaryotes are multicellular, but once eukaryotes existed, they evolved multicellularity many times independently (if I recall correctly).
AFAICT, “interstellar travel is impossible or extremely slow because there’s too much dust and crap in space that you’d collide with” remains a live possibility that doesn’t get enough attention around these parts.