How Technology Could Revolutionize Refugee Resettlement


Krishnadev Calamur in The Atlantic: “… For nearly 70 years, the process of interviewing, allocating, and accepting refugees has gone largely unchanged. In 1951, 145 countries came together in Geneva, Switzerland, to sign the Refugee Convention, the pact that defines who is a refugee, what refugees’ rights are, and what legal obligations states have to protect them.

This process was born of the idealism of the postwar years—an attempt to make certain that those fleeing war or persecution could find safety so that horrific moments in history, such as the Holocaust, didn’t recur. The pact may have been far from perfect, but in successive years, it was a lifeline to Afghans, Bosnians, Kurds, and others displaced by conflict.

The world is a much different place now, though. The rise of populism has brought with it a concomitant hostility toward immigrants in general and refugees in particular. Last October, a gunman who had previously posted anti-Semitic messages online against HIAS killed 11 worshippers in a Pittsburgh synagogue. Many of the policy arguments over resettlement have shifted focus from humanitarian relief to security threats and cost. The Trump administration has drastically cut the number of refugees the United States accepts, and large parts of Europe are following suit.

If it works, Annie could change that dynamic. Developed at Worcester Polytechnic Institute in Massachusetts, Lund University in Sweden, and the University of Oxford in Britain, the software uses what’s known as a matching algorithm to allocate refugees with no ties to the United States to their new homes. (Refugees with ties to the United States are resettled in places where they have family or community support; software isn’t involved in the process.)

Annie’s algorithm is based on a machine learning model in which a computer is fed huge piles of data from past placements, so that the program can refine its future recommendations. The system examines a series of variables—physical ailments, age, levels of education and languages spoken, for example—related to each refugee case. In other words, the software uses previous outcomes and current constraints to recommend where a refugee is most likely to succeed. Every city where HIAS has an office or an affiliate is given a score for each refugee. The higher the score, the better the match.

This is a drastic departure from how refugees are typically resettled. Each week, HIAS and the eight other agencies that allocate refugees in the United States make their decisions based largely on local capacity, with limited emphasis on individual characteristics or needs….(More)”.

How to Argue with an Algorithm: Lessons from the COMPAS ProPublica Debate


Paper by Anne L. Washington: “The United States optimizes the efficiency of its growing criminal justice system with algorithms however, legal scholars have overlooked how to frame courtroom debates about algorithmic predictions. In State v Loomis, the defense argued that the court’s consideration of risk assessments during sentencing was a violation of due process because the accuracy of the algorithmic prediction could not be verified. The Wisconsin Supreme Court upheld the consideration of predictive risk at sentencing because the assessment was disclosed and the defendant could challenge the prediction by verifying the accuracy of data fed into the algorithm.

Was the court correct about how to argue with an algorithm?

The Loomis court ignored the computational procedures that processed the data within the algorithm. How algorithms calculate data is equally as important as the quality of the data calculated. The arguments in Loomis revealed a need for new forms of reasoning to justify the logic of evidence-based tools. A “data science reasoning” could provide ways to dispute the integrity of predictive algorithms with arguments grounded in how the technology works.

This article’s contribution is a series of arguments that could support due process claims concerning predictive algorithms, specifically the Correctional Offender Management Profiling for Alternative Sanctions (“COMPAS”) risk assessment. As a comprehensive treatment, this article outlines the due process arguments in Loomis, analyzes arguments in an ongoing academic debate about COMPAS, and proposes alternative arguments based on the algorithm’s organizational context….(More)”

Report on Algorithmic Risk Assessment Tools in the U.S. Criminal Justice System


Press release: “The Partnership on AI (PAI) has today published a report gathering the views of the multidisciplinary artificial intelligence and machine learning research and ethics community which documents the serious shortcomings of algorithmic risk assessment tools in the U.S. criminal justice system. These kinds of AI tools for deciding on whether to detain or release defendants are in widespread use around the United States, and some legislatures have begun to mandate their use. Lessons drawn from the U.S. context have widespread applicability in other jurisdictions, too, as the international policymaking community considers the deployment of similar tools.

While criminal justice risk assessment tools are often simpler than the deep neural networks used in many modern artificial intelligence systems, they are basic forms of AI. As such, they present a paradigmatic example of the high-stakes social and ethical consequences of automated AI decision-making….

Across the report, challenges to using these tools fell broadly into three primary categories:

  1. Concerns about the accuracy, bias, and validity in the tools themselves
    • Although the use of these tools is in part motivated by the desire to mitigate existing human fallibility in the criminal justice system, this report suggests that it is a serious misunderstanding to view tools as objective or neutral simply because they are based on data.
  2. Issues with the interface between the tools and the humans who interact with them
    • In addition to technical concerns, these tools must be held to high standards of interpretability and explainability to ensure that users (including judges, lawyers, and clerks, among others) can understand how the tools’ predictions are reached and make reasonable decisions based on these predictions.
  3. Questions of governance, transparency, and accountability
    • To the extent that such systems are adapted to make life-changing decisions, tools and decision-makers who specify, mandate, and deploy them must meet high standards of transparency and accountability.

This report highlights some of the key challenges with the use of risk assessment tools for criminal justice applications. It also raises some deep philosophical and procedural issues which may not be easy to resolve. Surfacing and addressing those concerns will require ongoing research and collaboration between policymakers, the AI research community, civil society groups, and affected communities, as well as new types of data collection and transparency. It is PAI’s mission to spur and facilitate these conversations and to produce research to bridge such gaps….(More)”

AI & Global Governance: Robots Will Not Only Wage Future Wars but also Future Peace


Daanish Masood & Martin Waehlisch at the United Nations University: “At the United Nations, we have been exploring completely different scenarios for AI: its potential to be used for the noble purposes of peace and security. This could revolutionize the way of how we prevent and solve conflicts globally.

Two of the most promising areas are Machine Learning and Natural Language Processing. Machine Learning involves computer algorithms detecting patterns from data to learn how to make predictions and recommendations. Natural Language Processing involves computers learning to understand human languages.

At the UN Secretariat, our chief concern is with how these emerging technologies can be deployed for the good of humanity to de-escalate violence and increase international stability.

This endeavor has admirable precedent. During the Cold War, computer scientists used multilayered simulations to predict the scale and potential outcome of the arms race between the East and the West.

Since then, governments and international agencies have increasingly used computational models and advanced Machine Learning to try to understand recurrent conflict patterns and forecast moments of state fragility.

But two things have transformed the scope for progress in this field.

The first is the sheer volume of data now available from what people say and do online. The second is the game-changing growth in computational capacity that allows us to crunch unprecedented, inconceivable quantities data with relative speed and ease.

So how can this help the United Nations build peace? Three ways come to mind.

Firstly, overcoming cultural and language barriers. By teaching computers to understand human language and the nuances of dialects, not only can we better link up what people write on social media to local contexts of conflict, we can also more methodically follow what people say on radio and TV. As part of the UN’s early warning efforts, this can help us detect hate speech in a place where the potential for conflict is high. This is crucial because the UN often works in countries where internet coverage is low, and where the spoken languages may not be well understood by many of its international staff.

Natural Language Processing algorithms can help to track and improve understanding of local debates, which might well be blind spots for the international community. If we combine such methods with Machine Learning chatbots, the UN could conduct large-scale digital focus groups with thousands in real-time, enabling different demographic segments in a country to voice their views on, say, a proposed peace deal – instantly testing public support, and indicating the chances of sustainability.

Secondly, anticipating the deeper drivers of conflict. We could combine new imaging techniques – whether satellites or drones – with automation. For instance, many parts of the world are experiencing severe groundwater withdrawal and water aquifer depletion. Water scarcity, in turn, drives conflicts and undermines stability in post-conflict environments, where violence around water access becomes more likely, along with large movements of people leaving newly arid areas.

One of the best predictors of water depletion is land subsidence or sinking, which can be measured by satellite and drone imagery. By combining these imaging techniques with Machine Learning, the UN can work in partnership with governments and local communities to anticipate future water conflicts and begin working proactively to reduce their likelihood.

Thirdly, advancing decision making. In the work of peace and security, it is surprising how many consequential decisions are still made solely on the basis of intuition.

Yet complex decisions often need to navigate conflicting goals and undiscovered options, against a landscape of limited information and political preference. This is where we can use Deep Learning – where a network can absorb huge amounts of public data and test it against real-world examples on which it is trained while applying with probabilistic modeling. This mathematical approach can help us to generate models of our uncertain, dynamic world with limited data.

With better data, we can eventually make better predictions to guide complex decisions. Future senior peace envoys charged with mediating a conflict would benefit from such advances to stress test elements of a peace agreement. Of course, human decision-making will remain crucial, but would be informed by more evidence-driven robust analytical tools….(More)”.

LAPD moving away data-driven crime programs over potential racial bias


Mark Puente in The Los Angeles Times: “The Los Angeles Police Department pioneered the controversial use of data to pinpoint crime hot spots and track violent offenders.

Complex algorithms and vast databases were supposed to revolutionize crime fighting, making policing more efficient as number-crunching computers helped to position scarce resources.

But critics long complained about inherent bias in the data — gathered by officers — that underpinned the tools.

They claimed a partial victory when LAPD Chief Michel Moore announced he would end one highly touted program intended to identify and monitor violent criminals. On Tuesday, the department’s civilian oversight panel raised questions about whether another program, aimed at reducing property crime, also disproportionately targets black and Latino communities.

Members of the Police Commission demanded more information about how the agency plans to overhaul a data program that helps predict where and when crimes will likely occur. One questioned why the program couldn’t be suspended.

“There is very limited information” on the program’s impact, Commissioner Shane Murphy Goldsmith said.

The action came as so-called predictive policing— using search tools, point scores and other methods — is under increasing scrutiny by privacy and civil liberties groups that say the tactics result in heavier policing of black and Latino communities. The argument was underscored at Tuesday’s commission meeting when several UCLA academics cast doubt on the research behind crime modeling and predictive policing….(More)”.

How Recommendation Algorithms Run the World


Article by Zeynep Tufekci: “What should you watch? What should you read? What’s news? What’s trending? Wherever you go online, companies have come up with very particular, imperfect ways of answering these questions. Everywhere you look, recommendation engines offer striking examples of how values and judgments become embedded in algorithms and how algorithms can be gamed by strategic actors.

Consider a common, seemingly straightforward method of making suggestions: a recommendation based on what people “like you” have read, watched, or shopped for. What exactly is a person like me? Which dimension of me? Is it someone of the same age, gender, race, or location? Do they share my interests? My eye color? My height? Or is their resemblance to me determined by a whole mess of “big data” (aka surveillance) crunched by a machine-learning algorithm?

Deep down, behind every “people like you” recommendation is a computational method for distilling stereotypes through data. Even when these methods work, they can help entrench the stereotypes they’re mobilizing. They might easily recommend books about coding to boys and books about fashion to girls, simply by tracking the next most likely click. Of course, that creates a feedback cycle: If you keep being shown coding books, you’re probably more likely to eventually check one out.

Another common method for generating recommendations is to extrapolate from patterns in how people consume things. People who watched this then watched that; shoppers who purchased this item also added that one to their shopping cart. Amazon uses this method a lot, and I admit, it’s often quite useful. Buy an electric toothbrush? How nice that the correct replacement head appears in your recommendations. Congratulations on your new vacuum cleaner: Here are some bags that fit your machine.

But these recommendations can also be revealing in ways that are creepy. …

One final method for generating recommendations is to identify what’s “trending” and push that to a broader user base. But this, too, involves making a lot of judgments….(More)”.

The Importance of Data Access Regimes for Artificial Intelligence and Machine Learning


JRC Digital Economy Working Paper by Bertin Martens: “Digitization triggered a steep drop in the cost of information. The resulting data glut created a bottleneck because human cognitive capacity is unable to cope with large amounts of information. Artificial intelligence and machine learning (AI/ML) triggered a similar drop in the cost of machine-based decision-making and helps in overcoming this bottleneck. Substantial change in the relative price of resources puts pressure on ownership and access rights to these resources. This explains pressure on access rights to data. ML thrives on access to big and varied datasets. We discuss the implications of access regimes for the development of AI in its current form of ML. The economic characteristics of data (non-rivalry, economies of scale and scope) favour data aggregation in big datasets. Non-rivalry implies the need for exclusive rights in order to incentivise data production when it is costly. The balance between access and exclusion is at the centre of the debate on data regimes. We explore the economic implications of several modalities for access to data, ranging from exclusive monopolistic control to monopolistic competition and free access. Regulatory intervention may push the market beyond voluntary exchanges, either towards more openness or reduced access. This may generate private costs for firms and individuals. Society can choose to do so if the social benefits of this intervention outweigh the private costs.

We briefly discuss the main EU legal instruments that are relevant for data access and ownership, including the General Data Protection Regulation (GDPR) that defines the rights of data subjects with respect to their personal data and the Database Directive (DBD) that grants ownership rights to database producers. These two instruments leave a wide legal no-man’s land where data access is ruled by bilateral contracts and Technical Protection Measures that give exclusive control to de facto data holders, and by market forces that drive access, trade and pricing of data. The absence of exclusive rights might facilitate data sharing and access or it may result in a segmented data landscape where data aggregation for ML purposes is hard to achieve. It is unclear if incompletely specified ownership and access rights maximize the welfare of society and facilitate the development of AI/ML…(More)”

Data Collaboratives as an enabling infrastructure for AI for Good


Blog Post by Stefaan G. Verhulst: “…The value of data collaboratives stems from the fact that the supply of and demand for data are generally widely dispersed — spread across government, the private sector, and civil society — and often poorly matched. This failure (a form of “market failure”) results in tremendous inefficiencies and lost potential. Much data that is released is never used. And much data that is actually needed is never made accessible to those who could productively put it to use.

Data collaboratives, when designed responsibly, are the key to addressing this shortcoming. They draw together otherwise siloed data and a dispersed range of expertise, helping match supply and demand, and ensuring that the correct institutions and individuals are using and analyzing data in ways that maximize the possibility of new, innovative social solutions.

Roadmap for Data Collaboratives

Despite their clear potential, the evidence base for data collaboratives is thin. There’s an absence of a systemic, structured framework that can be replicated across projects and geographies, and there’s a lack of clear understanding about what works, what doesn’t, and how best to maximize the potential of data collaboratives.

At the GovLab, we’ve been working to address these shortcomings. For emerging economies considering the use of data collaboratives, whether in pursuit of Artificial Intelligence or other solutions, we present six steps that can be considered in order to create data collaborative that are more systematic, sustainable, and responsible.

The need for making Data Collaboratives Systematic, Sustainable and Responsible
  • Increase Evidence and Awareness
  • Increase Readiness and Capacity
  • Address Data Supply and Demand Inefficiencies and Uncertainties
  • Establish a New “Data Stewards” Function
  • Develop and strengthen policies and governance practices for data collaboration

Safeguards for human studies can’t cope with big data


Nathaniel Raymond at Nature: “One of the primary documents aiming to protect human research participants was published in the US Federal Register 40 years ago this week. The Belmont Report was commissioned by Congress in the wake of the notorious Tuskegee syphilis study, in which researchers withheld treatment from African American men for years and observed how the disease caused blindness, heart disease, dementia and, in some cases, death.

The Belmont Report lays out core principles now generally required for human research to be considered ethical. Although technically governing only US federally supported research, its influence reverberates across academia and industry globally. Before academics with US government funding can begin research involving humans, their institutional review boards (IRBs) must determine that the studies comply with regulation largely derived from a document that was written more than a decade before the World Wide Web and nearly a quarter of a century before Facebook.

It is past time for a Belmont 2.0. We should not be asking those tasked with protecting human participants to single-handedly identify and contend with the implications of the digital revolution. Technological progress, including machine learning, data analytics and artificial intelligence, has altered the potential risks of research in ways that the authors of the first Belmont report could not have predicted. For example, Muslim cab drivers can be identified from patterns indicating that they stop to pray; the Ugandan government can try to identify gay men from their social-media habits; and researchers can monitor and influence individuals’ behaviour online without enrolling them in a study.

Consider the 2014 Facebook ‘emotional contagion study’, which manipulated users’ exposure to emotional content to evaluate effects on mood. That project, a collaboration with academic researchers, led the US Department of Health and Human Services to launch a long rule-making process that tweaked some regulations governing IRBs.

A broader fix is needed. Right now, data science overlooks risks to human participants by default….(More)”.

Credit denial in the age of AI


Paper by Aaron Klein: “Banks have been in the business of deciding who is eligible for credit for centuries. But in the age of artificial intelligence (AI), machine learning (ML), and big data, digital technologies have the potential to transform credit allocation in positive as well as negative directions. Given the mix of possible societal ramifications, policymakers must consider what practices are and are not permissible and what legal and regulatory structures are necessary to protect consumers against unfair or discriminatory lending practices.

In this paper, I review the history of credit and the risks of discriminatory practices. I discuss how AI alters the dynamics of credit denials and what policymakers and banking officials can do to safeguard consumer lending. AI has the potential to alter credit practices in transformative ways and it is important to ensure that this happens in a safe and prudent manner….(More)”.