Some claims I've been repeating in conversation a bunch: 

Safety work (I claim) should either be focused on one of the following 

1. CEV-style full value loading, to deploy a sovereign 
2. A task AI that contributes to a pivotal act or pivotal process. 

I think that pretty much no one is working directly on 1. I think that a lot of safety work is indeed useful for 2, but in this case, it's useful to know what pivotal process you are aiming for. Specifically, why aren't you just directly working to make that pivotal act/process happen? Why do you need an AI to help you? Typically, the response is that the pivotal act/process is too difficult to be achieved by humans. In that case, you are pushing into a difficult capabilities regime -- the AI has some goals that do not equal humanity's CEV, and so has a convergent incentive to powerseek and escape. With enough time or intelligence, you therefore get wrecked, but you are trying to operate in this window where your AI is smart enough to do the cognitive work, but is 'nerd-sniped' or focused on the particular task that you like. In particular, if this AI reflects on its goals and starts thinking big picture, you reliably get wrecked. This is one of the reasons that doing alignment research seems like a particularly difficult pivotal act to aim for. 

 

Deception is a particularly worrying alignment failure mode because it makes it difficult for us to realize that we have made a mistake: at training time, a deceptive misaligned model and an aligned model make the same behavior. 

There are two ways for deception to appear: 

1. An action chosen instrumentally due to non-myopic future goals that are better achieved by deceiving humans now so that it has more power to achieve its goals in the future. 
2. Because deception was directly selected for as an action. 
    

Another way of describing the difference is that 1 follows from an inner alignment failure: a mesaoptimizer learned an unintended mesaobjective that performs well on training, while 2 follows from an outer alignment failure — an imperfect reward signal. 

Classic discussion of deception focuses on 1 (example 1, example 2),  but I think that 2 is very important as well, particularly because the most common currently used alignment strategy is RLHF, which actively selects for deception.

Once the AI has the ability to create strategies that involve deceiving the human, even without explicitly modeling the human, those strategies will win out and end up eliciting a lot of reward. This is related to the informed oversight problem: it is really hard to give feedback to a model that is smarter than you. I view this as a key problem with RLHF. To my knowledge very little work has been done exploring this and finding more empirical examples of RLHF models learning to deceive the humans giving it feedback, which is surprising to me because it seems like it should be possible. 

Thinking about ethics.

After thinking more about orthogonality I've become more confident that one must go about ethics in a mind-dependent way. If I am arguing about what is 'right' with a paperclipper, there's nothing I can say to them to convince them to instead value human preferences or whatever. 

I used to be a staunch moral realist, mainly relying on very strong intuitions against nihilism, and then arguing something that not nihilism -> moral realism. I now reject the implication, and think that there is both 1) no universal, objective morality, and 2) things matter. 

My current approach is to think of "goodness" in terms of what CEV-Thomas would think of as good. Moral uncertainty, then, is uncertainty over what CEV-Thomas thinks. CEV is necessary to get morality out of a human brain, because it is currently a bundle of contradictory heuristics. However, my moral intuitions still give bits about goodness. Other people's moral intuitions also give some bits about goodness, because of how similar their brains are to mine, so I should weight other peoples beliefs in my moral uncertainty.  

Ideally, I should trade with other people so that we both maximize a joint utility function, instead of each of us maximizing our own utility function. In the extreme, this looks like ECL. For most people, I'm not sure that this reasoning is necessary, however, because their intuitions might already be priced into my uncertainty over my CEV. 

Current impressions of free energy in the alignment space. 

1. Outreach to capabilities researchers. I think that getting people who are actually building the AGI to be more cautious about alignment / racing makes a bunch of things like coordination agreements possible, and also increases the operational adequacy of the capabilities lab. 
   1. One of the reasons people don't like this is because historically outreach hasn't gone well, but I think the reason for this is that mainstream ML people mostly don't buy "AGI big deal", whereas lab capabilities researchers buy "AGI big deal" but not "alignment hard". 
   2. I think people at labs running retreats, 1-1s, alignment presentations within labs are all great to do this. 
   3. I'm somewhat unsure about this one because of downside risk and also 'convince people of X' is fairly uncooperative and bad for everyone's epistemics. 
2. Conceptual alignment research addressing the hard part of the problem. This is hard and not easy to transition to without a bunch of upskilling, but if the SLT hypothesis is right, there are a bunch of key problems that mostly go unnassailed, and so there's a bunch of low hanging fruit there. 
3. Strategy research on the other low hanging fruit in the AI safety space. Ideally, the product of this research would be a public quantitative model about what interventions are effective and why. The path to impact here is finding low hanging fruit and pointing them out so that people can do them. 

Thinking a bit about takeoff speeds.

As I see it, there are ~3 main clusters: 

1. Fast/discountinuous takeoff. Once AIs are doing the bulk of AI research, foom happens quickly, before then, they aren't really doing anything that meaningful. 
2. Slow/continuous takeoff. Once AIs are doing the bulk of AI research, foom happens quickly, before then, they do alter the economy significantly 
3. Perenial slowness. Once AIs are doing the bulk of AI research, there is no foom even still, maybe because of compute bottlenecks, and so there is sort of constant rates of improvements that do alter things. 

It feels to me like multipolar scenarios mostly come from 3, because in either 1 or 2, the pre-foom state is really unstable, and eventually some AI will foom and become unipolar. In a continuous takeoff world, I expect small differences in research ability to compound over time. In a discontinuous takeoff, the first model to make the jump is the thing that matters.

3 also feels pretty unlikely to me, given that I expect running AIs to be cheap relative to training, so you get the ability to copy and scale intelligent labor dramatically, and I expect the AIs to have different skillsets than humans, and so be able to find low hanging fruit that humans missed. 

Some rough takes on the Carlsmith Report. 

Carlsmith decomposes AI x-risk into 6 steps, each conditional on the previous ones:

1. Timelines: By 2070, it will be possible and financially feasible to build APS-AI: systems with advanced capabilities (outperform humans at tasks important for gaining power), agentic planning (make plans then acts on them), and strategic awareness (its plans are based on models of the world good enough to overpower humans).
2. Incentives: There will be strong incentives to build and deploy APS-AI.
3. Alignment difficulty: It will be much harder to build APS-AI systems that don’t seek power in unintended ways, than ones that would seek power but are superficially attractive to deploy.
4. High-impact failures: Some deployed APS-AI systems will seek power in unintended and high-impact ways, collectively causing >$1 trillion in damage.
5. Disempowerment: Some of the power-seeking will in aggregate permanently disempower all of humanity.
6. Catastrophe: The disempowerment will constitute an existential catastrophe.

These steps defines a tree over possibilities. But the associated outcome buckets don't feel that reality carving to me. A recurring crux is that good outcomes are also highly conjunctive, i.e, one of these 6 conditions failing does not give a good AI outcome. Going through piece by piece: 

1. Timelines makes sense and seems like a good criteria; everything else is downstream of timelines. 
2. Incentives seems wierd. What does the world in which there are no incentives to deploy APS-AI look like? There are a bunch of incentives that clearly do impact people towards this already: status, desire for scientific discovery, power, money. Moreover, this doesn't seem necessary for AI x-risk -- even if we somehow removed the gigantic incentives to build APS-AI that we know exist, people might still deploy APS-AI because they personally wanted to, even though there weren't social incentives to do so. 
3. Alignment difficulty is another non necessary condition. Some ways of getting x-risk without alignment being very hard: 
   1. For one, this is a clear spectrum, and even if it is on the really low end of the system, perhaps you only need a small amount of extra compute overhead to robustly align your system. One of the RAAP stories might occur, and even though technical alignment might be pretty easy, but the companies that spend that extra compute robustly aligning their AIs gradually lose out to other companies in the competitive marketplace. 
   2. Maybe alignment is easy, but someone misuses AI, say to create an AI assisted dictatorship 
   3. Maybe we try really hard and we can align AI to whatever we want, but we make a bad choice and lock-in current day values, or we make a bad choice about reflection procedure that gives us much less than the ideal value of the universe. 
4. High-impact failures contains much of the structure, at least in my eyes. The main ways that we avoid alignment failure are worlds where something happens to take us off of the default trajectory: 
   1. Perhaps we make a robust coordination agreement between labs/countries that causes people to avoid deploying until they've solved alignment 
   2. Perhaps we solve alignment, and harden the world in some way, e.g. by removing compute access, dramatically improving cybersec, monitoring and shutting down dangerous training runs. 
   3. In general, thinking about the likelihood of any of these interventions that work, feels very important. 
5. Disempowerment. This and (4), are very entangled with upstream things like takeoff shape. Also, it feels extremely difficult for humanity to not be disempowered. 
6. Catastrophe. To avoid this, again, I need to imagine the extra structure upsteam of this, e.g. 4 was satisfied by a warning shot, and then people coordinated and deployed a benign sovreign that disempowered humanity for good reasons. 

My current preferred way to think about likelihood of AI risk routes through something like this framework, but is more structured and has a tree with more conjuncts towards success as well as doom. 

Some thoughts on inner alignment. 

1. The type of object of a mesa objective and a base objective are different (in real life) 
In a cartesian setting (e.g. training a chess bot), the outer objective is a function $R:S^n\to [0,1]$, where $S$ is the state space, and $S^n$ are the trajectories. When you train this agent, it's possible for it to learn some internal search and mesaobjective $O_{mesa}:S^n\to [0,1]$, since the model is big enough to express some utility function over trajectories. For example, it might learn a classifier that evaluates winningness of the board, and then gives a higher utility to the winning boards. 

In an embedded setting, the outer objective cannot see an entire world trajectory like it could in the cartesian setting. Your loss can see the entire trajectory of a chess game, but you loss can't see an entire atomic level representation of the universe at every point in the future. If we're trying to get an AI to care about future consequences over trajectories $O_{mesa}$ will have to have type $O_{mesa}:S^n\to [0,1]$, though it won't actually represent a function of this type because it can't, it will instead represent its values some other way (I don't really know how it would do this -- but (2) talks about the shape in ML). Our outer objective will have a much shallower type, $R:L\to [0,1]$, where $L$ are some observable latents. This means that trying to set get $O_{mesa}$ to equal $R$ doesn't even make sense as they have different type signatures. To salvage this, one could assume that $R$ factors into $R=E_{m\sim M\left(L\right)}{O}_{base}\left(m\right)$, where $M:L\to \Delta S^n$ is a model of the world and $O_{base}:S^n\to [0,1]$ is an objective, but it's impossible to actually compute $R$ this way. 

2. In ML models, there is no mesa objective, only behavioral patterns. More generally,  AI's can't naively store explicit mesaobjectives, they need to compress them in some way / represent them differently. 

My values are such that I do care about the entire trajectory of the world, yet I don't store a utility function with that type signature in my head. Instead of learning a goal over trajectories, ML models will have behavioral patterns that lead to states that performed well according to the outer objective on the training data. 

I have a behavioral pattern that says something like 'sugary thing in front of me -> pick up the sugary thing and eat it'. However, this doesn't mean that I reflectively endorse this behavioral pattern. If I was designing myself again from scratch or modifying my self, I would try to remove this behavioral pattern. 

This is the main-to-me reason why I don't think that the shard theory story of reflective stability holds up.^[1]  A bunch of the behavioral patterns that caused the AI to look nice during training will not get handed down into successor agents / self modified AIs. 

Even in theory, I don't yet know how to make reflectively stable, general, embedded cognition (mainly because of this barrier). 

1. ^{^}

   From what I understand, the shard theory story of reflective stability is something like: The shards that steer the values have an incentive to prevent themselves from getting removed. If you have a shard that wants to get lots of paperclips, the action that removes this shard from the mind would result in less paperclips being gotten. 
   Another way of saying this is that goal-content integrity is convergently instrumental, so reflective stability will happen by default.