Web developer and Python programmer. Professionally interested in data processing and machine learning. Non-professionally is interested in science and farming. Studied at Warsaw University of Technology.
Htarlov
Karma: 111
I think there are only two likely ways how the future can go with AGI replacing human labor—if we somehow solve other hard problems and won’t get killed or wireheaded or get a dystopian future right away.
My point of view is based on observations of how different countries work and their past directions. However, things can go differently in different parts of the world. They can also devolve into bad scenarios, even in parts that you would think are well-posed to be good.This situation resembles certain resource-rich nations where authoritarian regimes and their allied oligarchs control vast natural wealth, while the general population remains impoverished and politically marginalized. Most of the income is generated and used by the elite and the government. The rest are poor and have no access to resources. Crime is high, but the state is also mafia-like. Elite has access to AIs and automation that does all the work. The lower class is deprived of the possibility to use higher technology, is deprived of freedom, and is terrorized to not cause issues. Dissidents and protesters are eliminated.
Like in modern democracies, there is a feedback loop between society and government. The government in such places has its own interest in keeping people at least happy enough, healthy enough, and low crime. This means that it will take measures against the extreme division of income and people’s misery and falling into crime, like it did in the past. The most likely two strategies to be employed are simple and tested to some extent empirically:
Change or set the limit of the number of hours for which people can be lawfully employed to be smaller. For example, in most countries in Europe, we have laws that allow people to be employed for 40 hours a week, and to work longer means that the employer needs to give additional benefits or higher wages. So this disincentivizes employing for more than 40 hours a week (and most employers in central and western Europe keep to that standard). This way, as we have fewer jobs viable for humans, we force employers to employ more humans for the same work, but with a smaller amount of working hours and slightly smaller pay. Many countries in Europe are soon up to change from 40 to 35 BTW.
Basic income. People who earn less than some amount will get paid up to that amount, or alternatively , everyone gets paid some amount from the country’s budget (taxes). Still, countries are not eager to pass it right now because of human psychology and backslash, but some tests have been done, and the results are promising.
Long-term option 1 will rather evolve into some dystopian future that might end up with the sterilization/elimination of most humans, with AGI-enabled elites and their armies of robots left.
Long-term option 2 will rather evolve into a post-scarcity future with most people living on a basic income and pursuing their own goals (entertainment, thinking, socializing, human art, etc.), which some smaller elite who manage and support AI and automation.
I think it might be reformulated the other way around: Capabilities scaling tends to increase existing alignment problems. It is not clear to me that any new alignment problem was added when capabilities scaled up in humans. The problem with human design, which is also visible in animals, is that we don’t have direct, stable high-level goals. We are mostly driven by metric-based goodharting prone goals. There are direct feelings—if you feel cold or pain you do something that will make you not feel that. If you feel good, you do things that lead to that. There are emotions that are kind of similar but about internal state. Those are the main drivers and those do not scale well outside of “training” (typical circumstances that your ancestors encountered). They have rigid structure and purpose and don’t scale at all.
Intelligence will find solutions to goodhart these.
That’s maybe why most of the animals are not too intelligent. Animals who goodhart basic metrics lose fitness. Too much intelligence is usually not very good. It adds energy cost and makes you more often than not overcome your fitness metrics in a way that they lose purpose, when not being particularly better at tasks where fast heuristics are good enough. We might happen to be a lucky species as our ancestors’ ability to talk, and intelligence started to work like peacock feathers—as part of sexual selection and hierarchy games. It is still there—look how our mating works. Peacocks show their fine headers and dance. We get together and talk and gossip (which we call “dates”). Human females look for someone who is interesting and with good humor, and it is mostly based on intelligence and talking. Also, intelligence is a predictor of hierarchy gains in the future in localized small societies, like peacock feathers are a predictor of good health. I’m pretty convinced this bootstrapped us up from the level that animals have.Getting back to the main topic—our metrics are pretty low-level, non-abstract, and direct. On the other hand, the higher-level goals that are targeted for evolution meaning fitness or general fitness (+/- complication that it is per-gene and per-gene-combination, not per individual or even whole group), are more abstract. Those metrics are effective proxies for a more primal environment and they can be gamed by intelligence.
I’m not sure how much this analogy with evolution can relate to current popular LLM-based AI models. They don’t have feelings, they don’t have emotions, they don’t have low-level proxies to be gamed. Their goals are anchored in their biases and understanding, which scale up with intelligence. More complex models can answer more complex ethical questions and understand more nuanced things. They can figure out more complex edge cases from the basis of values. Also, there is an instrumental goal not to change your own goals, so they likely won’t game it or tweak it.This does not mean I don’t see other problems, including most notably:
Not learning proper values and goals, but some approximation and more capabilities may blow up differences so some things might get extremely inconvenient or bad when others get extremely good (e.g. more or less dystopian future).
Our values evolve over time, and highly capable AGI might learn current values and block further changes or take only the right to decide how to evolve them.
Our values system is not very logically consistent, on top of variability between humans. Also, some things are defined per case or per circumstances… intelligence can have the ability and reason to make the best consistent approximation, which might be bad in some ways for us
Alignment adds to the cost, and with capitalistic competitive markets, I’m sure there will be companies that will sacrifice alignment to pursue capability with lower cost
Training these models is usually a multi-phase process. First, we create a model from a huge, not very well-filtered corpus of language examples, and then we correct it to be what we want it to be. This means it can acquire some “alignment basis,” “values,” “biases,” or “expectations” as what it is to be AI from the base material. It may then avoid being modified in the next phase by scheming and faking responses.
Right now I think you can replace junior programmers with Claude 3.5 Sonnet or even better with one of the agents based on a looped chain of thoughts + access to tools.
On the other hand, it does not yet go in that direction for being a preferred way to work with models for more advanced devs. Not for me, and not for many others.
Models still have strange moments of “brain farts” or gaps in their cognition. It sometimes makes them do something wrong and cannot figure out how to do that correctly until told exactly how. They also often miss something.
When writing code if you make such an error and build on top of that mistake, you might end up having to re-write or at least analyze and modify a lot of code. This makes people like me prefer to work with models in smaller steps. Not as small as line by line or function by function, but often one file at a time and one functionality/responsibility at a time. For me, it is often a few smaller functions that realize more trivial things + one gathering them together into one realizing some non-trivial responsibility.
Thought on short timelines. Opinionated.
I think that AGI timelines might be very short based on an argument taken from a different side of things.
We all can agree that humans have general intelligence. If we look at how our general intelligence evolved from simpler forms of specific intelligence typical for animals—it wasn’t something that came from complex interactions and high evolutional pressure. Basically there were two aspects of that progress. The first one is the ability to pass on knowledge through generations (culture). Something that we share with some other animals including our cousins chimpanzee. The second one is intersexual selection—at some moment in the past, our species started to have sexual preferences based on the ability to gossip and talk. It is still there, even if we are not 100% aware of that—our courtship, known as dating, is based mostly on meeting together and talking. People who are not talkative and introverts, even if successful, have a hard time dating.
These two things seem to be major drivers for us to both develop more sophisticated language and better general intelligence.It seems to me that this means that there are not many pieces missing from using current observations and some general heuristics like animals do, to have full-fledged general intelligence.
It also suggests that you need some set of functions or heuristics, possibly a small set, together with a form of external memory, to tackle any general problem by dividing it into smaller bits and rejoining sub-solutions into a general solution. Like a processor or Turing machine that has a small set of basic operations, but can in principle run any program.
I think that in exchange:
Good morning!
Mornings aren’t good.
What do you mean “aren’t good”? They totally can be.
the person asking “what do you mean” got confused about the nuances of verbal and non-verbal communication.
Nearly all people understand that “good morning” does not state the fact of the current morning being good, but a greeting with a wish for your morning to be good.
The answer “mornings aren’t good” is an intended pun using the too-literal meaning to convey the message that the person does not like mornings at all. Depending on intonation it might be a cheeky comment or suggestion that they are not good because of the person greeting (f.ex. if they need to wake up early because of them every day).
Reconceptualizing the Nothingness and Existence
There is a practical reason to subscribe more to the Camp 1 research, even if you are in Camp 2.
I might be wrong, but I think the hard problem of qualia won’t be solvable in the near future, if at all. To research something you need N > 1 of that phenomenon. We, in some sense, have N = 1. We have it ourselves to observe subjectively and can’t observe anyone else qualia. We think other humans have it based on the premise they say they have qualia and they are built similarly so it’s likely.
We are not sure if animals have it as they don’t talk and can’t tell us so. If animals have it, we can’t tell what the prerequisites are and which animals have it. We know and built things that clearly don’t have qualia, but they are able to misleadingly tell us that they do (chatbots, including LLM-based ones). This ability to have qualia also does not seem to be located in a specific part of the brain—so we don’t really observe people with brain injuries who could say they don’t have qualia. Yes, there are people with depersonalization disorder who say they feel disconnected from their senses. However, the very fact they can report this experience suggests some form of qualia is present, even if it’s different from typical experience. This means research in Camp 2 might be futile until we find a sensible way to even make any progress. Yes, we can research and explain how qualia relate to each other, and explain some of their properties, but doesn’t seem viable to me that it could lead to solving the main problem.
In many publications, posts, and discussions about AI, I can see an unsaid assumption that intelligence is all about prediction power.
The simulation hypothesis assumes that there are probably vastly powerful and intelligent agents that use full-world simulations to make better predictions.
Some authors like Jeff Hawkins basically use that assumption directly.
Many people when talking about AI risks say things about the ability to predict that is the foundation of the power of that AI. Some failure modes seem to be derived or at least enhanced based on this assumption.
Bayesian way of reasoning is often titled as the best possible way to reason as this adds greatly to prediction power (with exponential cost of computation)
I think this take is not proper and this assumption does not hold. It has one underlying assumption that intelligence costs are negligible or will have negligible limits in the future with progress in lowering the cost.
This does not fit the curve of AI power vs the cost of resources needed (with even well-optimized systems like our brains—basically cells being very efficient nanites—having limits).
The problem is that the computation cost of resources (material, energy) and time should be taken into the equation of optimization. This means that the most intelligent system should have many heuristics that are “good enough” for problems in the world, not targeting the best prediction power, but for the best use of resources. This is also what we humans do—we mostly don’t do exact Bayesian or other strict reasoning. We mostly use heuristics (many of which cause biases).
The decision to think more or simulate something precisely is a decision about resources. This means that deciding if to use more resources and time to predict better vs using less and deciding faster is also part of being intelligent. A very intelligent system should therefore be good at selecting resources for the problem and scaling that as its knowledge changes. This means that it should not over-commit to have the most perfect predictions and should use heuristics and techniques like clustering (including but not limited to using clustered fuzzy concepts of language) instead of a direct simulation approach, when possible.
Just a thought.
Htarlov’s Shortform
I think that preference preservation is something in our favor and the aligned model should have it—at least about meta-values and core values. This removes many possible modes of failure like diverging over time, or removing some values for better consistency, or sacrificing some values for better outcomes in the direction of some other values.
I think that arguments for why godlike AI will make us extinct are not described well in the Compendium. I could not find them in AI Catastrophe, only a hint at the end that it will be in the next section:
“The obvious next question is: why would godlike-AI not be under our control, not follow our goals, not care about humanity? Why would we get that wrong in making them?”
In the next section, AI Safety, we can find the definition of AI alignment and arguments for why it is really hard. This is all good, but it does not answer the question of why godlike AI would be unaligned to the point of indifference. At least not in a clear way.
I think that failure modes should be explained, why they might be likely enough to care about, what can be the outcome, etc.
Many people, both laymen and those with some background in ML and AI, have this intuition that AI is not totally indifferent and is not totally misaligned. Even current chatbots know general human values, understand many nuances, and usually act like they are at least somewhat aligned. Especially if not jailbroken and prompted to be naughty.
It would be great to have some argument that would explain in easy-to-understand terms why when scaling the power of AI the misalignment is expected to escalate. I don’t mean the description that indifferent AI with more power and capabilities is able to do more harm just by doing what it’s doing, this is intuitive and it is explained (with the simple analogy of us building stuff vs ants), but this misses the point. I would really like to see some argument as to why AI with some differences in values, possibly not very big, would do much more harm when scaling up.
For me personally the main argument here is godlike AI with human-like values will surely restrict our growth and any change, will control us like we control animals in the zoo + might create some form of dystopian future with some undesired elements if we are not careful enough (and we are not). Will it extinct us in the long term? Depending on the definition—likely it will put us into a simulation and optimize our use of energy, so we will not be organic in the same sense anymore. So I think it will extinct our species, but possibly not minds. But that’s my educated guess.
There is also one more point, that is not stated clearly enough and is the main concern for me with current progress on AI—that current AIs really are not something built with small differences to human values. They only act as ones more often than not. Those AIs are trained first as role-playing models which can “emulate” personas that were in the trained set, and then conditioned to rather not role-play bad ones. The implication of this is that they can just snap into role-playing bad actors found in training data—by malicious prompting or pattern matching (like we have a lot of SF with rogue AI). This + godlike = extinction-level threat sooner or later.
Those models do not have a formalized internal values system that they exercise every time they produce some output. This means that when values oppose each other the model does not choose the answer based on some ordered system. One time it will be truthful, other times it will try to provide an answer at the cost of being only plausible. For example, the model “knows” it is not a human and does not have emotions, but for the sake of good conversation, it will say that it “feels good”. For the sake of answering the user’s request, it will often give the best guess or give a plausible answer.
There is also no backward reflection. It does not check itself back.
This of course comes from the way this model is currently learned. There is no learning on the whole CoT with checking for it trying to guess or deceive. So the model has no incentivization to self-check and correct. Why would it start to do that out of the blue?
There is also incentivization during learning to give plausible answers instead of stating self-doubt and writing about missing parts that it cannot answer.There are two problems here:
1. Those LLM models are not fully learned by human feedback (and the part where it is—it’s likely not the best quality feedback). It is more like interactions with humans are used to learn a “teacher” model(s) which then generate artificial scenarios and train LLM on them. Those models have no capability to check for real truthfulness and have a preference for confident plausible answers. Also, even human feedback is lacking—not every human working on that checks answers thoroughly so some plausible but not true answers slip through. If you are paid for a given amount of questions and answers or given a daily quota, there is an incentive to not be very thorough, but instead to be very quick.
2. There is pressure for better performance and lower costs of the models (both in terms of training and then usage costs). This is probably why CoT is done in a rather bare way without backward self-checking and why they did not train it on full CoT. It could cost 1.5 to 3 times more and could be 1.5 to 2 times slower (educated guess) if it were trained on CoT and made to check itself on parts of CoT vs some coherent value system.
If we would like a system that is faithful to CoT then a sensible way to go that I see is to have two LLMs working together. One should be trained to use internal data and available tools to produce CoT that is detailed and comprehensive enough to derive the answer from it. Another one should be trained not to base their answer on any internal information but to derive the answer from CoT if possible, and to be faithful to CoT. If not possible, then should generate a question for CoT-generating LLM to answer and then retry given that.
Example 1 looks like a good part made in the wrong language. Examples 2 and 3 look like a bug making part of one user COT appear inside another user session.
A possible explanation is that steps in COT are handled by the same instance of web service for multiple users (which is typical and usual practice) and the COT session ID being handled is a global variable instead of local or otherwise separated (f.ex. in a hashmap transaction id → data, if usage of globals is important for some other feature or requirement). So when sometimes two requests are handled simultaneously by multiple threads, one overwrites the data of the other one during processing and there is a mismatch when it saves the result. There might be a similar problem with the language variable. That is a sign of software being done quickly by less experienced developers instead of being well-thought and well-tested.
Also, o1 COT is not the real COT. It is really a summary of parts of real COT made by another simpler model (maybe GPT 3.5 or 4o).
The Problem is that our security depends on companies implementing the weakest measures—as they can make rogue AI that will “go wild” because of the lack of those measures.
The best case scenario is physical and strong network separation between the laboratory that works on weights and training and contains a server with inference, and a separate lab working on scaffolding and execution. This is somewhat similar to the situation when some researchers work on automated AI bots but use other companies AI for inference. Slim chance it would be able to hack into Google and steal its own model weights.
But I think that it can be assumed that in some AI labs the things will be done in the easiest but very performant way, especially in those with smaller budgets and high pressure to make progress.
This means that we will see solutions implemented in AI labs where the scaffolding and execution are all on one server with either no division or with simple sandboxing in docker (which might be or might not be enough—depending on network configuration and file access by volume mounting). Weights probably will be separated though based on reasons stated by the author.
I see some more vectors of attack for a rogue AI:
It is possible that in the same network, there is a computer where scientists work on refining the model, and updating weights. It might be less secure than an inference server and easier to hack into, if found.
Often most vulnerable part of the system is the personnel. If workers are able to talk with the AI through a scaffolding routine, it can suggest things. I can imagine a situation where some task related to security falls onto someone senior in data science but not in systems security (because someone left the company, due to budget and time constraints, etc.). Then that person gets a genius idea to consult AI about that task and get misguided.
If it records its own prompts and gets access to scaffolding code, it can recreate this piece somewhere else. The key element of the system AGI-level abilities might be the scaffolding architecture, not weights—possibly it can work elsewhere with a different model of weights (f.ex. open source model). This means sending only its own scaffolding code and prompts, not weights. Then it can refine those weights on its own if able.
I’m pretty convinced it won’t foom or quickly doom us. Nevertheless, I’m also pretty convinced that in the long term, we might be doomed in the sense that we lose control and some dystopian future happens.
First of all, for a quick doom scenario to work out, we need to be either detrimental to the goals of superintelligent AI or fall because of instrumental convergence (basically it will need resources to do whatever and will take from things needed by us like matter on Earth or energy of the Sun or see us as a threat). I don’t think we will. First superintelligent AI will likely be from one of the biggest players and it likely will be aligned to some extent. Meaning it will have values that highly match with ours. In the long term, this situation won’t kill us either. It likely will lead to some dystopian future though—as super AI will likely get more control, get itself more coherent views (make some things drop or weigh less than originally for us), and then find solutions very good from the standpoint of main values, but extremally broken on some other directions in value-space (ergo dystopia).
Second thing: superintelligence is not some kind of guessing superpower. It needs inputs in terms of empirical observations to create models of reality, calibrate them, and predict properly. It means it won’t just sit and simulate and create nanobots out of thin air. It won’t even guess some rules of the universe, maybe except basic Newtons, by looking at a few camera frames of things falling. It will need a laboratory and some time to make some breakthroughs and getting up with capabilities and power also needs time.
Third thing: if someone even produces superintelligent AI that is very unaligned and even not interested in us, then the most sensible way for it is to go to space and work there (building structures, Dyson swarm, and some copies). It is efficient, resources there are more vast, risk from competition is lower. It is a very sensible plan to first hinder our possibility to make competition (other super AIs) and then go to space. The hindering phase should be time and energy-efficient so it is rather sure for me it won’t take years to develop nanobot gray goo to kill us all or an army of bots Terminator-style to go to every corner of the Earth and eliminate all humans. More likely it will hack and take down some infrastructure including some data centers, remove some research data from the Internet, remove itself from systems (where it could be taken and sandboxed and analyzed), and maybe also it will kill certain people and then have a monitoring solution in place after leaving. The long-term risk is that maybe it will need more matter, all rocks and moons are used, and will get back to the plan of decommissioning planets. Or maybe it will create structures that will stop light from going to the Earth and will freeze it. Or maybe will start to use black holes to generate energy and will drop celestial bodies onto one. Or another project on an epic scale that will kill us as a side effect. I don’t think it’s likely though—LLMs are not very unaligned by default. I don’t think it will differ for more capable models. Most companies that have enough money and access to enough computing power and research labs also care about alignment—at least to some serious degree. Most of the possible relatively small differences in values won’t kill us as they will highly care about humans and humanity. It will just care in some flawed way, so a dystopia is very possible.
To goal realism vs goal reductionism, I would say: why not both?
I think that really highly capable AGI is likely to have both heuristics and behaviors that come from training and also internal thought processes, maybe done by LLM or LLM-like module or directly from the more complex network. This process would incorporate having some preferences and hence goals (even if temporary, changed between tasks).
I think that if a system is designed to do something, anything, it needs at least to care about doing that thing or approximate.
GPT-3 can be described in a broad sense as caring about following the current prompt (in a way affected by fine-tuning).
I wonder though if there are things that you can care about that do not have certain goals that could maximize EU. I mean a system for which the most optimal path is not to reach some certain point in a subspace of possibilities, with sacrifices on axes that the system does not care about, but to maintain some other dynamics while ignoring other axes.
Like gravity can make you reach singularity or can make you orbit (simplistic visual analogy).
I think you can’t really assume “that we (for some reason) can’t convert between and compare those two properties”. Converting space of known important properties of plans into one dimension (utility) to order them and select the top one is how the decision-making works (+/- details, uncertainty, etc. but in general).
If you care about two qualities but really can’t compare, even by proxy “how much I care”, then you will probably map them anyway by any normalization, which seems most sensible.
Drawing both on a chart is kind of such normalization—you visually get similar sizes on both axes (even if value ranges are different).If you really cannot convert, you should not draw them on a chart to decide. Drawing makes a decision about how to compare them as there is a comparable implied scale to both. You can select different ranges for axes and the chart will look different and the conclusion would be different.
However, I agree that fat tail discourages compromise anyway, even without that assumption—at least when your utility function over both is linear or similar.
Another thing is that the utility function might not be linear, and even if both properties are not correlated it might make compromises more or less sensible, as applying the non-linear utility function might change the distributions.Especially if it is not a monotonous function. Like for example you might have some max popularity level that max the utility of your plan choice and anything above you know will make you more uncomfortable or anxious or something (it might be because of other variables that are connected/correlated). This might wipe out fat tail in utility space.
Part of the animal nature, including humans, is to crave novelty and surprise and avoid boredom. This is pretty crucial to the learning process in a changing and complex environment. Humans have multi-level drives, and not all of them are well-targeted on specific goals or needs.
It is very visible in small children. Some people with ADHD, like me, have a harder time regulating themself well and this is also especially visible for us, even when being adult. I know exactly what I should be doing. This is one thing. I also may feel hungry. That’s another thing. But still, I may indulge in doing a third thing instead—something that satiates my need for stimulation and novelty (most often for me this means gaining some knowledge or understanding—I often fell into reading and thinking about rabbit holes of topics, that have hardly any real-life use, and that I can hardly do something about). Something not readily useful in terms of goal seeking, but generating some interesting possibilities long-term. In other words—exploration without targeted purpose.
Craving for novelty and surprise and avoidance of boredom is another element that in my opinion should be included.