Announcing: Mechanism Design for AI Safety—Reading Group

[Rubi J. Hudson]

We’re starting a new reading group for people interested in applying mechanism design tools to technical AI alignment. If you’re interested in joining, you can apply here by August 22nd (applying takes less than five minutes). If you have recommendations for papers to discuss, please mention them in the comments.

What We’re Doing

Mechanism design is the study of how to reach desirable outcomes or equilibria in the face of differing incentives and incomplete information. Many AI safety researchers have expressed enthusiasm about the potential of using these tools for work in alignment, but relatively little work has been done in the intersection. We believe this is partially due to a lack of potential researchers with expertise in both technical AI safety and mechanism design, and partially due to a lack of shovel-ready problems. The goal of this reading group is to make progress on both fronts.

There are three main areas that this reading group will cover:

1. Work at the intersection of technical AI safety and mechanism design, and where it can be expanded

2. Current work in technical AI safety, and how mechanism design tools can help

3. Current work on mechanism design, and how it can be applied to technical AI safety

The plan is to start with papers in the intersection, then alternate between papers on technical AI safety and papers on mechanism design while keeping the broader perspective in mind. Note that although we believe AI governance work is important and contains many applications for mechanism design, that will not be the focus of this reading group.

Who We Are and Who We Want

I (Rubi) am entering the 2nd year of a PhD in Economics this fall, and am currently working on technical AI safety in Berkeley through the SERI MATS program. Other likely participants include three Economics PhD students at top schools and a Math undergraduate student currently taking part in the SERI Summer Research Fellowship. Our hope for this reading group is to connect with people who have similar interests and create the potential for future collaborations.

Based on current expressions of interest, we expect the modal participant in the reading group to be a PhD student in economics, focusing on economic theory, who has read through the AGI Safety Fundamentals curriculum (or an equivalent, such as Eleuther’s). If that sounds like you, definitely apply! However, these should not be considered necessary qualifications. Talented undergraduates with an interest in both areas or experts in one area who would like to learn more about the other should also apply.

If you’re unsure whether you have the background necessary to keep up with this reading group, a good test is to try skimming The Off-Switch Game. It’s a short paper, and on the more accessible end of papers we will be discussing. If you understand it or predict you would be able to understand it within an hour, then you are likely to be able to process the papers that we will discuss without too much additional work. If you find yourself struggling to understand the mathematical notation and proofs, then that is likely a bottleneck and you should consider prioritizing work to advance your comfort level there.

Participants will be expected to commit approximately eight hours a week for this reading group, which consists of five to seven hours reading the week’s paper and an hour and a half to discuss it. If it becomes apparent that a participant is repeatedly not reading or only skimming the papers, they will be removed from the reading group. Please ensure that you can dedicate the required time before applying.

Logistics

The application form can be found here. The only mandatory fields are a link/​upload of your CV and confirmation that you are willing to make the necessary time commitment, although there are also optional fields if you would like to elaborate on your background in either mechanism design or technical AI safety.

Applications will close on Monday August 22nd at midnight PST, and acceptances will be sent out by August 28th. Discussions will begin in the first week of September and continue weekly for twelve weeks. Meetings will be held online, at a time chosen based on the schedules of participants.

We currently expect one discussion group of between five to eight people. However, if there is sufficient interest then we will run however many groups are required to include all qualified applicants.

Exceptional candidates who cannot commit to attending all meetings can contact me directly about sitting in on the subset of meetings that are relevant to their work.

What We’ll be Reading

A number of people have asked to be provided with the reading list that we will be using. This list will be public once it has been finalized, but due to the small nature of the reading group we plan to customize the papers discussed to the interests of the participants.

To give a taste of the curriculum and to give potential participants a head start on readings, here is what we have planned for the first two weeks:

Week 1

(Double session, 2.5 hours)

Incomplete Contracting and AI Alignment by Dylan Hadfield-Menell and Gillian Hadfield,

The Principal-Agent Alignment Problem in Artificial Intelligence by Dylan Hadfield-Menell

This week will begin with introductions and a short icebreaker. The first paper discusses applying mechanism design to AI safety in broad terms, while the second delves more into specifics. In addition to the two papers, this week’s discussion will cover the areas of AI safety where mechanism could be useful, the limitations of the approach, and the potential upside from success

Week 2

(Normal session, 1.5 hours)

Risks from Learned Optimization in Advanced Machine Learning Systems by Evan Hubinger, Chris van Merwijk, Vladimir Mikulik, Joar Skalse, and Scott Garrabrant

We expect most participants will have already read this paper, which covers the differences between outer and inner alignment. This week’s discussion will involve a brief review of the paper, followed by consideration of which mechanism design tools can help with each form of alignment.

For the following weeks, a list of relevant topics which may be covered (subject to participant interest) include: the principal-agent problem, cheap talk, multi-agent systems, dynamic mechanisms, robust mechanism design, corrigibility, multi-agent reinforcement learning, cooperative AI and communication, adversarial training and zero-sum mechanisms, causal incentives, and algorithmic mechanism design. Other topics may also be covered, if requested.

Future Plans

Our intention for this reading group is to transition to a working group upon completion. With a shared background, we will be in a good position to provide feedback on each others’ work or collaborate on projects. In addition to a working group, we would also like to have the group produce an agenda in which we lay out what we feel are the most promising research directions, the potential challenges, and the next steps to work on. Ideally (i.e. conditional on funding) this agenda would be hammered out over multiple days at a retreat that includes subject matter experts in both mechanism design and technical AI safety.

Between a reading list, a research agenda, and an active community of researchers, we would be in a position where new members could quickly get up to speed. The long-term goal is to increase the number of people working on technical AI safety by making it easy for mechanism design researchers to contribute, and to improve the quality of technical AI safety research by expanding the set of available tools.

The application form is here, and applications are due by Monday August 22nd at midnight PST. Please pass along this post to anyone who you think would be interested in this reading group.

Thanks to Cecilia Wood and Amelia Michael for reviewing earlier drafts of this post.

[Caspar Oesterheld]

Sounds interesting! Are you going to post the reading list somewhere once it is completed?

(Sorry for self-promotion in the below!)

I have a mechanism design paper that might be of interest: Caspar Oesterheld and Vincent Conitzer: Decision Scoring Rules. WINE 2020. Extended version. Talk at CMID.

Here’s a pitch in the language of incentivizing AI systems—the paper is written in CS-econ style. Imagine you have an AI system that does two things at the same time:
1) It makes predictions about the world.
2) It takes actions that influence the world. (In the paper, we specifically imagine that the agent makes recommendations to a principal who then takes the recommended action.) Note that if the predictions are seen by humanity, they themselves influence the world. So even a pure oracle AI might satisfy 2, as has been discussed before (see end of this comment).
We want to design a reward system for this agent such the agent maximizes its reward by making accurate predictions and taking actions that maximize our, the principals’, utility.

The challenge is that if we reward the accuracy of the agent’s predictions, we may set an incentive on the agent to make the world more predictable, which will generally not be aligned without mazimizing our utility.

So how can we properly incentivize the agent? The paper provides a full and very simple characterization of such incentive schemes, which we call proper decision scoring rules:

We show that proper decision scoring rules cannot give the [agent] strict incentives to report any properties of the outcome distribution [...] other than its expected utility. Intuitively, rewarding the [agent] for getting anything else about the distribution right will make him [take] actions whose outcome is easy to predict as opposed to actions with high expected utility [for the principal]. Hence, the [agent’s] reward can depend only on the reported expected utility for the recommended action. [...] we then obtain four characterizations of proper decision scoring rules, two of which are analogous to existing results on proper affine scoring [...]. One of the [...] characterizations [...] has an especially intuitive interpretation in economic contexts: the principal offers shares in her project to the [agent] at some pricing schedule. The price schedule does not depend on the action chosen. Thus, given the chosen action, the [agent] is incentivized to buy shares up to the point where the price of a share exceeds the expected value of the share, thereby revealing the principal’s expected utility. Moreover, once the [agent] has some positive share in the principal’s utility, it will be (strictly) incentivized to [take] an optimal action.

Also see Johannes Treutlein’s post on “Training goals for large language models”, which also discusses some of the above results among other things that seem like they might be a good fit for the reading group, e.g., Armstrong and O’Rourke’s work.

My motivation for working on this was to address issues of decision making under logical uncertainty. For this I drew inspiration from the fact that Garrabrant et al.’s work on logical induction is also inspired by market design ideas (specifically prediction markets).

[Rubi J. Hudson]

Update: the reading list has now been posted.

[Rubi J. Hudson]

Yeah, the full reading list will be posted publicly once it’s finalized.

Thanks for the recommendation! I was planning on including something from yourself/​Vince/​out of FOCAL, but wasn’t sure which option to go with.