Invited Talk 1

“Legitimacy, Authority, and the Political Value of Explanations”

Seth Lazar (Australian National University)
As rapid advances in Artificial Intelligence and the rise of some of history's most potent corporations meet the diminished neoliberal state, we have become increasingly subject to power exercised by means of automated systems. Machine learning, big data, and related computational technologies now underpin vital government services from criminal justice to tax auditing, from public health to social services, from immigration to defence. Twosided markets connecting consumers and producers are shaped by algorithms proprietary to companies such as Google and Amazon. Google's search algorithm determines, for many of us, how we find out about everything from how to vote to where to get vaccinated; Facebook, Twitter and Google decide which of our fellow citizens' speech we get to see—both what gets taken down, and (more importantly) what gets promoted. We sometimes imagine AI as a far off goal—either the handmaiden to a new postscarcity world, or else humanity's apocalyptic 'final invention'. But we are already using AI to shape an increasing proportion of our online and offline lives. As the pandemic economic shock ramifies, and the role of technology in our lives grows exponentially, this will only intensify.
We are increasingly subject to Automatic Authorities—automated computational systems that are used to exercise power over us, by substantially determining what we may know, what we may have, and what our options will be. These computational systems promise radical efficiencies and new abilities. But, as is now widely recognised, they also pose new risks. In this paper I focus on one in particular: that the adoption of Automatic Authorities leads us to base increasingly important decisions on systems whose operations cannot be adequately explained to democratic citizens.
Philosophers have long debated the importance of justifications in morality and politics, but they have not done the same for explanations. What's more, the most prominent Automatic Authorities in our lives today are deployed by nonstate actors like Google and Facebook, and analytical political philosophy has focused much more on state than nonstate power. To make progress on one of the most pressing questions of the age of Automatic Authorities, therefore, we must make substantial firstorder progress, on two fronts, in moral and political philosophy.
That is my goal in this paper. My central claim: only if the powerful can adequately explain their decisions to those on whose behalf or by whose licence they act can they exercise power legitimately and with proper authority, and so overcome presumptive objections to their exercise of power grounded in individual freedom, social equality, and collective selfdetermination. This applies to all authorities, not only automatic ones. But I will demonstrate its application to Automatic Authorities, including those sustained by nonstate actors. I will use this account of why explanations matter to address the urgent regulatory questions of to whom explanations are owed, and what kinds of explanations are owed to them.
Invited Talk 2

“Explanation: from ethics to logic”

Gilles Dowek (INRIA and ENS ParisSaclay)
Abstract: Explaining decisions is an ethical necessity. For example, neither a person nor a piece of software ought to reject a bank loan application, without providing an explanation for this rejection. But, defining the notion of explanation is a challenge. Starting from concrete examples, we attempt to understand what such a definition could look like.
A usual definition assumes that what is explained is a statement and what explains it is a logical proof of this statement. For example, a proof in elementary geometry both shows that the statement is true and explains why it is true. But this definition is not sufficient, as some logical proofs are seen as more explanatory than others.
In this talk, we give two successive definitions of the notion of explanation. The first keeps the idea that an explanation is a logical proof, but the explanatory character of the a proof relies on the fact that it contains a cut: the proof of general statements followed by a specialization to a particular one. So, the fact that a proof is explanatory is measured by the degree of generalization it allows. The second definition uses the algorithmic interpretation of proofs to generalize this definition. An explanation is then a pair formed with a short, fast, and wide algorithm and an input value for this algorithm.
Invited Talk 3

“Justified Representation in ApprovalBased Committee Voting”

Edith Elkind (Oxford University)
Computational social choice is a rapidly growing research field that studies algorithmic aspects of collective decisionmaking and preference aggregation. It studies questions such as: can we quickly compute the outcome of a given voting rule? For a given voting rule, can a strategic voter efficiently compute her optimal strategy? Is there a voting rule that satisfies a particular set of desirable properties (axioms) and admits an efficient winner determination algorithm?
In this talk, we consider these questions in the context of multiwinner voting rules with approval ballots. We formulate a fairness axiom, which we call Justified Representation, as well as a strengthening of this axiom. We identify voting rules that satisfy the JR axiom and investigate their algorithmic complexity. Finally, we propose a tractable rule that satisfies the strong version of the axiom.