As autonomous agents proliferate in the real world, both in software and robotic settings, they will increasingly need to band together for cooperative activities with previously unfamiliar teammates. In such "ad hoc" team settings, team strategies cannot be developed a priori.

Rather, an agent must be prepared to cooperate with many types of teammates: it must collaborate without pre-coordination. This talk will cover past and ongoing research on the challenge of building autonomous agents that are capable of robust ad hoc teamwork.

In this talk, I will present the case for the critical role played by third-party enforced rules in the extensive forms of cooperation we see in humans. Cooperation, I’ll argue, cannot be adequately accounted for—or modeled for AI—within the framework of human preferences, coordination incentives or bilateral commitments and reciprocity alone. Cooperation is a group phenomenon and requires group infrastructure to maintain. This insight is critical for training AI agents that can cooperate with humans and, likely, other AI agents. Training environments need to be built with normative infrastructure that enables AI agents to learn and participate in cooperative activities—including the cooperative activity that undergirds all others: collective punishment of agents that violate community norms.

Humans routinely face two types of cooperation problems: How to get to a collectively good outcome given some set of preferences and structural constraints; and how to design, shape, or shove structural constraints and preferences to induce agents to make choices that bring about better collective outcomes. In the terminology of economic theory, the first is a problem of equilibrium selection given a game structure, and the second is a problem of mechanism design by a “social planner.” These two types of problems have been distinguished in and are central to a much longer tradition of political philosophy (e.g., state of nature arguments). It is fairly clear how AI can and might be constructively applied to the first type of problem, while less clear for the second type. How to think about using AI to contribute to optimal design of the terms and parameters – the rules of a game – for other agents? Put differently, could there be an AI of constitutional design?

We consider environments where a set of human workers needs to handle a large set of tasks while interacting with human users. The arriving tasks vary: they may differ in their urgency, their difficulty and the required knowledge and time duration in which to perform them. Our goal is to decrease the number of workers, which we refer to as operators that are handling the tasks while increasing the users’ satisfaction. We present automated intelligent agents that will work together with the human operators in order to improve the overall performance of such systems and increase both operators' and users’ satisfaction. Examples include: home hospitalization environment where remote specialists will instruct and supervise treatments that are carried out at the patients' homes; operators that tele-operate autonomous vehicles when human intervention is needed and bankers that provide online service to customers. The automated agents could support the operators: the machine learning-based agent follows the operator’s work and makes recommendations, helping him interact proficiently with the users. The agents can also learn from the operators and eventually replace the operators in many of their tasks.

Participants can send questions via Sli.do using this link: https://app.sli.do/event/ambolxqi

Participants can send questions via Sli.do using this link: https://app.sli.do/event/02lguhzy

Participants can send questions via Sli.do using this link: https://app.sli.do/event/riko0stp

Participants can send questions via Sli.do using this link: https://app.sli.do/event/50mlx6cq

Participants can send questions via Sli.do using this link: https://app.sli.do/event/9opzmndo

Participants can send questions via Sli.do using this link: https://app.sli.do/event/uqh9pktn

Gather Town link: [ protected link dropped ] /1l0kNMMpqLZvr9Co/CooperativeAI

Authors: Rose Wang, Sarah Wu, James Evans, Joshua Tenenbaum, David Parkes and Max Kleiman-Weiner

Authors: Kamal Ndousse, Douglas Eck, Sergey Levine and Natasha Jaques

Authors: Rohin Shah, Pedro Freire, Neel Alex, Rachel Freedman, Dmitrii Krasheninnikov, Lawrence Chan, Michael Dennis, Pieter Abbeel, Anca Dragan and Stuart Russell

Authors: Xavier Puig, Tianmin Shu, Shuang Li, Zilin Wang, Josh Tenenbaum, Sanja Fidler and Antonio Torralba