# ➔ Phenomenal World

## ARTIFICIAL AGENCY AND EXPLANATION

### The gray box of XAI

A recent longform piece in the New York Times identifies the problem of explaining artificial intelligence. The stakes are high because of the European Union’s controversial and unclear “right-to-explanation” law, which will become active in May 2018.

“Instead of certainty and cause, A.I. works off probability and correlation. And yet A.I. must nonetheless conform to the society we’ve built — one in which decisions require explanations, whether in a court of law, in the way a business is run or in the advice our doctors give us. The disconnect between how we make decisions and how machines make them, and the fact that machines are making more and more decisions for us, has birthed a new push for transparency and a field of research called explainable A.I., or X.A.I. Its goal is to make machines able to account for the things they learn, in ways that we can understand. But that goal, of course, raises the fundamental question of whether the world a machine sees can be made to match our own.”

An interdisciplinary group addresses the problem:

"Contrary to popular wisdom of AI systems as indecipherable black boxes, we find that this level of explanation should often be technically feasible but may sometimes be practically onerous—there are certain aspects of explanation that may be simple for humans to provide but challenging for AI systems, and vice versa. As an interdisciplinary team of legal scholars, computer scientists, and cognitive scientists, we recommend that for the present, AI systems can and should be held to a similar standard of explanation as humans currently are; in the future we may wish to hold an AI to a different standard."

Full article by FINALE DOSHI-VELEZ et al. here. ht Margarita For the layperson, the most interesting part of the article may be its general overview of societal norms around explanation and explanation in the law.

Michael comments: Human cognitive systems have generated similar questions in vastly different contexts. The problem of chick-sexing (see Part 3) gave rise to a mini-literature within epistemology.

From Michael S. Moore’s book Law and Society: Rethinking the Relationship: “A full explanation in terms of reasons for action requires two premises: the major premise, specifying the agent’s desires (goals, objectives, moral beliefs, purposes, aims, wants, etc.), and the minor premise, specifying the agent’s factual beliefs about the situation he is in and his ability to achieve, through some particular action, the object of his desires.” Link. ht Margarita

• A Medium post with an illustrated summary of some XAI techniques. Link.

## PREDICTIVE JUSTICE

### How to build justice into algorithmic actuarial tools

Key notions of fairness contradict each other—something of an Arrow’s Theorem for criminal justice applications of machine learning.

"Recent discussion in the public sphere about algorithmic classification has involved tension between competing notions of what it means for a probabilistic classification to be fair to different groups. We formalize three fairness conditions that lie at the heart of these debates, and we prove that except in highly constrained special cases, there is no method that can satisfy these three conditions simultaneously. Moreover, even satisfying all three conditions approximately requires that the data lie in an approximate version of one of the constrained special cases identified by our theorem. These results suggest some of the ways in which key notions of fairness are incompatible with each other, and hence provide a framework for thinking about the trade-offs between them."

Full paper from JON KLEINBERG, SENDHIL MULLAINATHAN and MANISH RAGHAVAN here. h/t research fellow Sara, who recently presented on bias in humans, courts, and machine learning algorithms, and who was the source for all the papers in this section.

In a Twitter thread, ARVIND NARAYANAN describes the issue in more casual terms.

"Today in Fairness in Machine Learning class: a comparison of 21 (!) definitions of bias and fairness [...] In CS we're used to the idea that to make progress on a research problem as a community, we should first all agree on a definition. So 21 definitions feels like a sign of failure. Perhaps most of them are trivial variants? Surely there/s one that's 'better' than the rest? The answer is no! Each defn (stat. parity, FPR balance, contextual fairness in RL...) captures something about our fairness intuitions."

Jay comments: Kleinberg et al. describe their result as choosing between conceptions of fairness. It’s not obvious, though, that this is the correct description. The criteria (calibration and balance) discussed aren’t really conceptions of fairness; rather, they’re (putative) tests of fairness. Particular questions about these tests aside, we might have a broader worry: if fairness is not an extensional property that depends upon, and only upon, the eventual judgments rendered by a predictive process, exclusive of the procedures that led to those judgments, then no extensional test will capture fairness, even if this notion is entirely unambiguous and determinate. It’s worth consideringNozick’s objection to “pattern theories” of justice for comparison, and (procedural) due process requirements in US law.

## PROPONENT EXCHANGE

### Policy feedback loops and US old-age policy

Researchers of policy history have long deliberated over explanatory frameworks: institutionalist accounts tend to focus on inherited conditions and path dependency in political development, while others stress the importance of social movements in shaping policy. Among the more dynamic analytical frameworks for the study of welfare politics is that of "policy feedbacks," which looks at the evolution of policy as an iterative process in which new policies change the conditions for further political engagement.

In a 2019 article, EDWIN AMENTA and THOMAS ALAN ELLIOTT look at the utility of these frameworks for explaining the character of old-age pensions in the US from the mid-1930s to 1950s.

From the article:

## FLUIDITY OF MANPOWER

### On contracts and 'intermediate' labor market institutions

The recent boom in
identifying and measuring monopsony in labor markets has brought the question of employers' wage-setting power to the fore of various academic and policy debates. (For an overview, see our blog post by Owen Davis from earlier this year.) Along with its more direct theoretical antecedents, this body of work joins a broader interdisciplinary tradition in examining the relationship between various forms of coercion and the labor contract.

In a 2011 paper, using historical data on contract breaches and game theoretical models, Suresh Naidu and Noam Yuchtman examine how Master and Servant law affected contracting and wages in 19th century Britain. The dynamics examined in the paper provide robust evidence of what the authors call “intermediate” labor market institutions—between the poles of free and forced labor.

"We document that criminal prosecutions were widely applied by employers in response to labor demand shocks: a high marginal revenue product of labor led to greater numbers of prosecutions. We address concerns about endogeneity by using exogenous industry-specific output price shocks for independent variation in labor demand, and examining the resulting prosecutions specifically in areas where affected industries were concentrated. We find that positive labor demand shocks in the coal mining, iron, and textile industries all produced increased prosecutions, precisely in counties where those industries were located. We find further evidence suggesting that employers used penal sanctions as a substitute for paying higher wages in response to positive labor demand shocks, which supported long-term contracting: average wages in high prosecution counties, and the responsiveness of wages to labor demand shocks, increased after the 1875 elimination of criminal prosecutions under Master and Servant law.

Historical labor markets have rarely looked like textbook, perfectly competitive markets. Attempts to manage labor mobility have generated a wide variety of legal institutions, ranging from slavery to employment at will. We believe that the study of intermediate cases, such as 19th century Britain, the American South after the Civil War, and the post-emancipation British Caribbean, illuminates the role of legal institutions in securing the supply of effective labor, and represents a rich area for future work."

• From 2005, Chiaki Moriguchi compares the development of labor law in the United States and Japan during the aftermath of the Great Depression. Link. In the International Review of Social History, Sidney Chalhoub examines the "legal and social ambiguities between slavery and freedom that prevailed in nineteenth-century Brazilian society." Link.
• "Negotiated labor contracts make only minor modifications to a relationship whose normatively critical features have already been set by law independently of the will of both parties." A 2015 paper by Elizabeth Anderson on the role of government in ensuring free labor contracts. Link. See also: Anderson's 2017 book Private Government, which drew from her 2015 Tanner Lectures at Princeton. Link to the lecture manuscript.
• Daron Acemoglu and Alexander Wolitzky on the "economics of labor coercion." Link.
• "Suppose a worker were to refuse to yield to the coercion of any employer, but were to choose instead to remain under the legal duty to abstain from the use of any of the money which anyone owns. He must eat. While there is no law against eating in the abstract, there is a law which forbids him to eat any of the food which actually exists in the community—and that law is the law of property." From legal realist Robert Hale’s classic 1923 paper 'Coercion and Distribution in a Supposedly Non-Coercive State.' Link.

## UPWARD ACCOMODATION

### The history of public housing provision

In recent decades, policy approaches to housing provision have focused on increasing the incomes of subsidy recipients and, due to declining federal investment, promoting tenant mobility both between subsidized housing units and out of the public housing system altogether. But the discourse on housing seems to be shifting. Rather than promoting ever increasing incomes, recent proposals aim instead to control housing costs—both through increasing public housing stock and pegging rent to inflation.

In a 2012 paper, Lawrence J. Vale and Yonah Freemark offer a history of public housing in the United States. Their narrative considers how changing approaches to housing provision reveal changes in the government's definition of “deserving” welfare recipients.

From the paper:

"Public housing is too often conceptualized as a single failed program that tragically concentrated deeply impoverished single-parent minority households in ill-designed and publicly mismanaged slums. Such a viewpoint does little justice to the evolution and contingencies that motivated the growth and directions of the multiphased and multifaceted history of federally supported public housing and public-private housing. Taking a longer view, the concentrated poverty welfare phase of public housing may actually be seen as an aberration, a relatively brief interlude between about 1960 and 1990. This phase, we argue, was out of step with the larger pattern of policy preferences for housing the poor, both before and since.

Seen this way, American public housing consists of a 25-year series of efforts to accommodate the upwardly mobile working class between 1935 and 1960, a 30-year consolidation of the poorest into welfare housing between 1960 and the mid 1980s, coupled by efforts to introduce direct private-sector involvement in public housing and other programs; and a series of programs and policies since the mid 1980s to return more of public housing to a less-poor constituency, while furthering growth in other kinds of both deep and shallow subsidy programs through mixed-finance projects and tax-code intervention. After 75 years of experimentation, much of the rest of public housing operations has become completely privatized. In many cities, housing authorities are regularly turning over their conventional housing stock to private managers and often own nothing more than the land beneath their redevelopment endeavors. In this context, even the basic definitional reason for calling some housing 'public housing' now comes into question."

• From November of last year, Jack Y. Favilukis, Pierre Mabille, and Stijn Van Nieuwerburgh find that "Housing affordability policies create large net welfare gains." Link. See also J. W. Mason's recent public testimony on rent control, which offers an overview of empirical findings and concludes that "there is no evidence that rent regulations reduce the overall supply of housing." Link.
• A report by Peter Gowan and Ryan Cooper at 3P compares housing policy in US metropolitan areas with those of Vienna, Helsinki, and Stockholm. Link. At the Urban Institute, Emily Peiffer discusses the history of housing policy in New York City. Link.
• "Housing Affordability in the U.S.: Trends by Geography, Tenure, and Household Income." By Andrew Dumont at the Federal Reserve. Link.
• Data for Progress maps the diversity of America's public housing communities, accounting for rates of unemployment, poverty, and population density. Link. Another map looks at flood risk, police stops, and segregation in NYCHA buildings. Link.