My current blog is epistem.ink. This one is here just for archival purposes.
AI alarmism and the misinformed position
Let’s begin with a gentle introduction into the field of AI risk, I’ve long felt like when people talk about “AI risk” they are often talking about things which have nothing in common, not even the broad definition of “AI” they are using.
The Various hues of AI risk
I’ve often had a bone to pick against “AI risk” or, as I’ve referred to it, “AI alarmism”. When evaluating AI risk, there are multiple views on the location of the threat and the perceived warning signs.
I think it might help to classify them, I doubt that people actually working in the field would agree with my classification, but you know how the saying goes, either you label yourself or an unsupervised algorithm will do it for you.
1. The Bostromian position
I would call one of these viewpoints the “Bostromian position”, which seems to be mainly promoted by MIRI, philosophers like Nick Bostrom and on forums such as AI Alignment. It’s hard to summarize without apparently straw-man arguments, e.g. “AIXI + Moore’s law means that all-powerful superhuman intelligence is dangerous, inevitable and close.”
That’s partly because I’ve never seen a consistent top-to-bottom reasoning for it. Its proponents always seem to start by assuming things that I wouldn’t hold as given about the ease of data collection, the cost of computing power, the usefulness of intelligence.
I’ve tried to argue against this position, the summary of my view can probably be found in “Artificial general intelligence is here, and it's useless”. Whilst - for the reasons mentioned there - I don’t see it as particularly stable, I think it’s not fundamentally flawed; I could see myself arguing pro or con.
Another way to look at it, which I am shamelessly stealing almost word for word from someone else, is that this position disjunctive.
It's difficult to find a consistent top-to-bottom story because the overall argument is disjunctive. That is, a conjunction is the intersection of different events, whereas a disjunction is the union of different (potentially overlapping) events.
That is to say, it’s based on the idea that there are so many paths for AI to reach a point where it poses a risk to us, that the overall probability adds up to be significant.
However, it fails to give full details about any one of those paths, it fails to give a finite list of paths and it fails to have a Popperian way of being disproved. If you scrutinize one of the ways of getting “superintelligence” as being risk-free or improbable to the point of impossibility, there’s always 100 more ready to take its place.
2. The Standard Position
Advocated by people ranging from my friends, to politicians, to respectable academics, to CEOs of large tech companies. It is perhaps best summarized in Stuart Russell’s book Human Compatible: Artificial Intelligence and the Problem of Control.
This viewpoint is mainly based around real-world use cases for AI (where AI can be understood as “machine learning”).
This position is more or less focused on worrying about the things that are already happening or will be happening in the next 5, 10 or 20 years.
It’s not being worried that self-driving cars will be taken over by a mastermind AI bent to kill all human in order to maximize its reward function, it’s worried about self-driving cars being allowed on the roads without being properly vetted through the standard means by which we can vet an algorithm.
It’s not being worried about us not realizing when the AI decides to transfer our consciousness to a simulation, it’s worried about kids getting addicted to VR games made to look and feel amazing using ML-aided rendering and feedback techniques.
People adopting this perspective are not wrong in being worried, but rather in being worried about the wrong thing.
It’s wrong to be upset by Facebook or Youtube using an algorithm to control and understand user preferences and blaming it on “AI”, rather than on people not being educated enough to use TOR, install a tracking blocker, use Ublock Origin and not center their entire life around conspiracy videos in their youtube feed or in anti-vaccination Facebook groups.
It’s wrong to be alarmed by Amazon making people impulse-buy via a better understanding of their preferences, and thus getting them into inescapable debt, rather than by the legality of providing unethically leveraged debt so easily.
It’s wrong to fuss about automated trading being able to cause sudden large dips in the market, rather than about having markets so unstable and so focused on short-term trading as to make this the starting point of the whole argument. It’s wrong to worry about NLP technology being used to implement preventive policing measures, rather than about governments being allowed to steal their citizens’ data, to request backdoors into devices and to use preventive policing to begin with.
It’s wrong to worry about the Chinese Communist Party using facial recognition and tracking technology to limit civil rights; Worry instead about CCP ruling via a brutal dictatorship that implements such measures without anybody doing something against it.
But I digress, though I ought to give a full rebuttal of this position at some point.
Disclaimer: Stuart Russell’s work doesn’t necessarily disagree with Bostrom’s position, it’s just that it seems more focused on arguing about “current and near-future” threats. Hence why I think Stuart Russell’s work is a good guide to this viewpoint.
3. The misinformed position
This is a viewpoint distinct from the previous two, but I don't think many people hold it consciously. It stems from misunderstanding what machine learning systems can already do. It basically consists in panicking over “new developments” which actually have existed for decades.
This view is especially worth fighting against since it’s based on misinformation. Whereas with categories 1 and 2 I can see valid arguments arising for regulating or better understanding machine learning systems (or AI systems in general), people in the third category just don’t understand what’s going on, so they are prone to adopt any view out of sheer fear or need of belonging, without truly understanding the matter.
Until recently I thought this kind of task was better left to PBS. In hindsight, I’ve seen otherwise smart individuals being amazed that “AI” can solve a problem which anyone that has actually worked with machine learning would have been able to tell you is obviously solvable and has been since forever.
Furthermore, I think addressing this viewpoint is relevant, as it’s actually challenging and interesting. The question of “What are the problems we should assume can be solved with machine learning?”, or even narrower and more focused on current developments “What are the problems we should assume a neural network should be able to solve?”, is one I haven’t seen addressed much.
There are theories like PAC learning and AIX which at a glance seem to revolve around this, as it pertains to machine learning in general, but if actually tried in practice won’t yield any meaningful answer.
How people misunderstand what neural networks can do
Let’s look at the general pattern of fear generated by misunderstanding the machine learning capabilities we’ve had for decades.
Show a smart but relatively uninformed person - a philosophy Ph.D. or an older street-smart businessman - a deep learning party trick.
Give them the most convoluted and scary explanation of why it works. E.g. Explain Deep-Dream by using incomplete neurological data about the visual cortex and human image processing, rather than just saying it’s the outputs of a complex edge detector overfit to recognize dog faces.
Wait for them to write an article about it in VOX & co,
An example that originally motivated this article is Scott Alexander’s article about being amazed that GPT-2 is able to learn how to play chess, poorly.
It seems to imply that GPT-2 playing chess well enough not to lose very badly against a mediocre opponent (the author) is impressive and surprising.
Actually, the fact that a 1,500,000,000-parameter model designed for sequential inputs can be trained to kind of play chess is rather unimpressive, to say the least. I would have been astonished if GPT-2 were unable to play chess. Fully connected models a hundred times smaller ( https://github.com/pbaer/neural-chess) could do that more than 2 years ago.
The successful training of GPT-2 is not a feat because if a problem like chess has been already solved using various machine learning models we can assume it can be done with a generic neural network architecture (e.g. any given FC net or an FC net with a few attention layers) hundreds or thousands of times larger in terms of parameters.
It should be our default expectation that neural networks are universal function estimators, so if a given equation can model a problem (e.g. chess), our default expectation should be that a large enough neural network can probably come to approximate that equation or a one that’s analogous to it.
In the GPT-2 example, transformers (i.e. the BERT-like models inspired by the “Attention is all you need” paper’s proposed design) are pretty generic as far as NN architectures go. Not as generic as a fully connected net, arguably; they seem to perform more efficiently (in terms of training time and model size) on many tasks, and they are much better on most sequential input tasks.
Playing chess decently is also a problem already solved. It can be done using small (compared to GPT-2) decision trees and a few very simple heuristics (see for example https://github.com/AdnanZahid/Chess-AI-TDD). If a much smaller model can learn how to play “decently”, we should assume that a fairly generic, exponentially larger neural network can do the same.
Escaping the uninformed view
I’d argue that the only way to escape this position is to basically a large does of information about where machine learning is at, which one can more or less get by thoroughly browsing a website like https://paperswithcode.com/ and looking at the latest developments being sold as a service by companies like Google and Amazon.
Combined this with a heavy dose of skepticism, so that when someone is showing you an already solved problem being solved a different way your default assumption is: This is just snake-oil-salesman, and you’re basically there.
If you want to be more through though, I think the solution is to basically build a model that allows you to look at a problem, even one that hasn’t been touched by machine learning yet, and ask “Should my default hypothesis be that a machine learning algorithm, or even more narrowly, a neural network, can solve this problem”.
I think this is fairly useful - not only for not getting impressed when someone shows us a party trick and tells us it’s AGI - but also for helping us quickly classify a problem as “likely solvable via ML” and “unlikely to be solved by ML”, a skill which I’m pretty sure will become highly valuable in the near future.
To that extent, I would point you to this followup article: When to assume neural networks can solve a problem, which delves more deeply into this problem, mainly based on observations.
If you enjoyed this article you might also like:
Published on: 2020-03-28