#151 – Ajeya Cotra on accidentally teaching AI models to deceive us

Imagine you are an orphaned eight-year-old whose parents left you a $1 trillion company, and no trusted adult to serve as your guide to the world. You have to hire a smart adult to run that company, guide your life the way that a parent would, and administer your vast wealth. You have to hire that adult based on a work trial or interview you come up with. You don’t get to see any resumes or do reference checks. And because you’re so rich, tonnes of people apply for the job — for all sorts of reasons.Today’s guest Ajeya Cotra — senior research analyst at Open Philanthropy — argues that this peculiar setup resembles the situation humanity finds itself in when training very general and very capable AI models using current deep learning methods.As she explains, such an eight-year-old faces a challenging problem. In the candidate pool there are likely some truly nice people, who sincerely want to help and make decisions that are in your interest. But there are probably other characters too — like people who will pretend to care about you while you’re monitoring them, but intend to use the job to enrich themselves as soon as they think they can get away with it.Like a child trying to judge adults, at some point humans will be required to judge the trustworthiness and reliability of machine learning models that are as goal-oriented as people, and greatly outclass them in knowledge, experience, breadth, and speed. Tricky!Can’t we rely on how well models have performed at tasks during training to guide us? Ajeya worries that it won’t work. The trouble is that three different sorts of models will all produce the same output during training, but could behave very differently once deployed in a setting that allows their true colours to come through. She describes three such motivational archetypes:Saints — models that care about doing what we really wantSycophants — models that just want us to say they’ve done a good job, even if they get that praise by taking actions they know we wouldn’t want them toSchemers — models that don’t care about us or our interests at all, who are just pleasing us so long as that serves their own agendaIn principle, a machine learning training process based on reinforcement learning could spit out any of these three attitudes, because all three would perform roughly equally well on the tests we give them, and ‘performs well on tests’ is how these models are selected.But while that’s true in principle, maybe it’s not something that could plausibly happen in the real world. After all, if we train an agent based on positive reinforcement for accomplishing X, shouldn’t the training process spit out a model that plainly does X and doesn’t have complex thoughts and goals beyond that?According to Ajeya, this is one thing we don’t know, and should be trying to test empirically as these models get more capable. For reasons she explains in the interview, the Sycophant or Schemer models may in fact be simpler and easier for the learning algorithm to creep towards than their Saint counterparts.But there are also ways we could end up actively selecting for motivations that we don’t want.For a toy example, let’s say you train an agent AI model to run a small business, and select it for behaviours that make money, measuring its success by whether it manages to get more money in its bank account. During training, a highly capable model may experiment with the strategy of tricking its raters into thinking it has made money legitimately when it hasn’t. Maybe instead it steals some money and covers that up. This isn’t exactly unlikely; during training, models often come up with creative — sometimes undesirable — approaches that their developers didn’t anticipate.If such deception isn’t picked up, a model like this may be rated as particularly successful, and the training process will cause it to develop a progressively stronger tendency to engage in such deceptive behaviour. A model that has the option to engage in deception when it won’t be detected would, in effect, have a competitive advantage.What if deception is picked up, but just some of the time? Would the model then learn that honesty is the best policy? Maybe. But alternatively, it might learn the ‘lesson’ that deception does pay, but you just have to do it selectively and carefully, so it can’t be discovered. Would that actually happen? We don’t yet know, but it’s possible.In today’s interview, Ajeya and Rob discuss the above, as well as:How to predict the motivations a neural network will develop through trainingWhether AIs being trained will functionally understand that they’re AIs being trained, the same way we think we understand that we’re humans living on planet EarthStories of AI misalignment that Ajeya doesn’t buy intoAnalogies for AI, from octopuses to aliens to can openersWhy it’s smarter to have separate planning AIs and doing AIsThe benefits of only following through on AI-generated plans that make sense to human beingsWhat approaches for fixing alignment problems Ajeya is most excited about, and which she thinks are overratedHow one might demo actually scary AI failure mechanismsLearn more and read the full transcript on the 80,000 Hours website.This episode was originally released in May 2023.Chapters:Rob’s intro (00:00:00)The interview begins (00:02:38)How Ajeya’s views have changed since 2020 (00:05:09)Are neural networks more like a sped-up version of evolution, or a slower version of human learning? (00:17:42)Situational awareness (00:26:10)Misalignment stories Ajeya doesn't buy (00:42:03)The orphan heir with a trillion-dollar fortune (00:59:14)Saints, Sycophants, and Schemers (01:03:41)Ways to train safer AI systems (01:23:20)Aliens and other analogies (01:38:22)Moral patienthood (01:53:21)ARC Evaluations (01:55:35)Interpretability research (02:09:25)Rewarding models based on how good and sensible their plans seem to us (02:17:48)Overrated approaches (02:25:49)Demos of actually scary alignment failures (02:30:57)Skills to develop for doing useful work (02:37:23)Rob’s outro (02:47:24)Producer: Keiran HarrisAudio mastering: Ryan Kessler and Ben CordellTranscriptions: Katy Moore