Ethical guidelines issued by engineers’ organization fail to gain traction

Curated on Posted on by Stefaan Verhulst

Blogpost by Nicolas Kayser-Bril: “In early 2016, the Institute of Electrical and Electronics Engineers, a professional association known as IEEE, launched a “global initiative to advance ethics in technology.” After almost three years of work and multiple rounds of exchange with experts on the topic, it released last April the first edition of Ethically Aligned Design, a 300-page treatise on the ethics of automated systems.

The general principles issued in the report focus on transparency, human rights and accountability, among other topics. As such, they are not very different from the 83 other ethical guidelines that researchers from the Health Ethics and Policy Lab of the Swiss Federal Institute of Technology in Zurich reviewed in an article published in Nature Machine Intelligence in September. However, one key aspect makes IEEE different from other think-tanks. With over 420,000 members, it is the world’s largest engineers’ association with roots reaching deep into Silicon Valley. Vint Cerf, one of Google’s Vice Presidents, is an IEEE “life fellow.”

Because the purpose of the IEEE principles is to serve as a “key reference for the work of technologists”, and because many technologists contributed to their conception, we wanted to know how three technology companies, Facebook, Google and Twitter, were planning to implement them.

Transparency and accountability

Principle number 5, for instance, requires that the basis of a particular automated decision be “discoverable”. On Facebook and Instagram, the reasons why a particular item is shown on a user’s feed are all but discoverable. Facebook’s “Why You’re Seeing This Post” feature explains that “many factors” are involved in the decision to show a specific item. The help page designed to clarify the matter fails to do so: many sentences there use opaque wording (users are told that “some things influence ranking”, for instance) and the basis of the decisions governing their newsfeeds are impossible to find.

Principle number 6 states that any autonomous system shall “provide an unambiguous rationale for all decisions made.” Google’s advertising systems do not provide an unambiguous rationale when explaining why a particular advert was shown to a user. A click on “Why This Ad” states that an “ad may be based on general factors … [and] information collected by the publisher” (our emphasis). Such vagueness is antithetical to the requirement for explicitness.

AlgorithmWatch sent detailed letters (which you can read below this article) with these examples and more, asking Google, Facebook and Twitter how they planned to implement the IEEE guidelines. This was in June. After a great many emails, phone calls and personal meetings, only Twitter answered. Google gave a vague comment and Facebook promised an answer which never came…(More)”

Algorithmic Impact Assessments under the GDPR: Producing Multi-layered Explanations

Curated on Posted on by Stefaan Verhulst

Paper by Margot E. Kaminski and Gianclaudio Malgieri: “Policy-makers, scholars, and commentators are increasingly concerned with the risks of using profiling algorithms and automated decision-making. The EU’s General Data Protection Regulation (GDPR) has tried to address these concerns through an array of regulatory tools. As one of us has argued, the GDPR combines individual rights with systemic governance, towards algorithmic accountability. The individual tools are largely geared towards individual “legibility”: making the decision-making system understandable to an individual invoking her rights. The systemic governance tools, instead, focus on bringing expertise and oversight into the system as a whole, and rely on the tactics of “collaborative governance,” that is, use public-private partnerships towards these goals. How these two approaches to transparency and accountability interact remains a largely unexplored question, with much of the legal literature focusing instead on whether there is an individual right to explanation.

The GDPR contains an array of systemic accountability tools. Of these tools, impact assessments (Art. 35) have recently received particular attention on both sides of the Atlantic, as a means of implementing algorithmic accountability at early stages of design, development, and training. The aim of this paper is to address how a Data Protection Impact Assessment (DPIA) links the two faces of the GDPR’s approach to algorithmic accountability: individual rights and systemic collaborative governance. We address the relationship between DPIAs and individual transparency rights. We propose, too, that impact assessments link the GDPR’s two methods of governing algorithmic decision-making by both providing systemic governance and serving as an important “suitable safeguard” (Art. 22) of individual rights….(More)”.

A fairer way forward for AI in health care

Curated on Posted on by Stefaan Verhulst

Linda Nordling at Nature: “When data scientists in Chicago, Illinois, set out to test whether a machine-learning algorithm could predict how long people would stay in hospital, they thought that they were doing everyone a favour. Keeping people in hospital is expensive, and if managers knew which patients were most likely to be eligible for discharge, they could move them to the top of doctors’ priority lists to avoid unnecessary delays. It would be a win–win situation: the hospital would save money and people could leave as soon as possible.

Starting their work at the end of 2017, the scientists trained their algorithm on patient data from the University of Chicago academic hospital system. Taking data from the previous three years, they crunched the numbers to see what combination of factors best predicted length of stay. At first they only looked at clinical data. But when they expanded their analysis to other patient information, they discovered that one of the best predictors for length of stay was the person’s postal code. This was puzzling. What did the duration of a person’s stay in hospital have to do with where they lived?

As the researchers dug deeper, they became increasingly concerned. The postal codes that correlated to longer hospital stays were in poor and predominantly African American neighbourhoods. People from these areas stayed in hospitals longer than did those from more affluent, predominantly white areas. The reason for this disparity evaded the team. Perhaps people from the poorer areas were admitted with more severe conditions. Or perhaps they were less likely to be prescribed the drugs they needed.

The finding threw up an ethical conundrum. If optimizing hospital resources was the sole aim of their programme, people’s postal codes would clearly be a powerful predictor for length of hospital stay. But using them would, in practice, divert hospital resources away from poor, black people towards wealthy white people, exacerbating existing biases in the system.

“The initial goal was efficiency, which in isolation is a worthy goal,” says Marshall Chin, who studies health-care ethics at University of Chicago Medicine and was one of the scientists who worked on the project. But fairness is also important, he says, and this was not explicitly considered in the algorithm’s design….(More)”.

The Algorithmic Divide and Equality in the Age of Artificial Intelligence

Curated on Posted on by Stefaan Verhulst

Paper by Peter Yu: “In the age of artificial intelligence, highly sophisticated algorithms have been deployed to detect patterns, optimize solutions, facilitate self-learning, and foster improvements in technological products and services. Notwithstanding these tremendous benefits, algorithms and intelligent machines do not provide equal benefits to all. Just as the digital divide has separated those with access to the Internet, information technology, and digital content from those without, an emerging and ever-widening algorithmic divide now threatens to take away the many political, social, economic, cultural, educational, and career opportunities provided by machine learning and artificial intelligence.

Although policymakers, commentators, and the mass media have paid growing attention to algorithmic bias and the shortcomings of machine learning and artificial intelligence, the algorithmic divide has yet to attract much policy and scholarly attention. To fill this lacuna, this article draws on the digital divide literature to systematically analyze this new inequitable gap between the technology haves and have-nots. Utilizing the analytical framework that the Author developed in the early 2000s, the article discusses the five attributes of the algorithmic divide: awareness, access, affordability, availability, and adaptability.

This article then turns to three major problems precipitated by an emerging and fast-expanding algorithmic divide: (1) algorithmic deprivation; (2) algorithmic discrimination; and (3) algorithmic distortion. While the first two problems affect primarily those on the unfortunate side of the algorithmic divide, the latter impacts individuals on both sides of the divide. This article concludes by proposing seven nonexhaustive clusters of remedial actions to help bridge this emerging and ever-widening algorithmic divide. Combining law, communications policy, ethical principles, institutional mechanisms, and business practices, the article fashions a holistic response to help foster equality in the age of artificial intelligence….(More)”.

The Extended Corporate Mind: When Corporations Use AI to Break the Law

Curated on Posted on by Stefaan Verhulst

Paper by Mihailis Diamantis: “Algorithms may soon replace employees as the leading cause of corporate harm. For centuries, the law has defined corporate misconduct — anything from civil discrimination to criminal insider trading — in terms of employee misconduct. Today, however, breakthroughs in artificial intelligence and big data allow automated systems to make many corporate decisions, e.g., who gets a loan or what stocks to buy. These technologies introduce valuable efficiencies, but they do not remove (or even always reduce) the incidence of corporate harm. Unless the law adapts, corporations will become increasingly immune to civil and criminal liability as they transfer responsibility from employees to algorithms.

This Article is the first to tackle the full extent of the growing doctrinal gap left by algorithmic corporate misconduct. To hold corporations accountable, the law must sometimes treat them as if they “know” information stored on their servers and “intend” decisions reached by their automated systems. Cognitive science and the philosophy of mind offer a path forward. The “extended mind thesis” complicates traditional views about the physical boundaries of the mind. The thesis states that the mind encompasses any system that sufficiently assists thought, e.g. by facilitating recall or enhancing decision-making. For natural people, the thesis implies that minds can extend beyond the brain to include external cognitive aids, like rolodexes and calculators. This Article adapts the thesis to corporate law. It motivates and proposes a doctrinal framework for extending the corporate mind to the algorithms that are increasingly integral to corporate thought. The law needs such an innovation if it is to hold future corporations to account for their most serious harms….(More)”.

Insurance Discrimination and Fairness in Machine Learning: An Ethical Analysis

Curated on Posted on by Stefaan Verhulst

Paper by Michele Loi and Markus Christen: “Here we provide an ethical analysis of discrimination in private insurance to guide the application of non-discriminatory algorithms for risk prediction in the insurance context. This addresses the need for ethical guidance of data-science experts and business managers. The reference to private insurance as a business practice is essential in our approach, because the consequences of discrimination and predictive inaccuracy in underwriting are different from those of using predictive algorithms in other sectors (e.g. medical diagnosis, sentencing). Moreover, the computer science literature has demonstrated the existence of a trade-off in the extent to which one can pursue non- discrimination versus predictive accuracy. Again the moral assessment of this trade-off is related to the context of application…(More)”

AI Global Surveillance Technology

Curated on Posted on by Stefaan Verhulst

Carnegie Endowment: “Artificial intelligence (AI) technology is rapidly proliferating around the world. A growing number of states are deploying advanced AI surveillance tools to monitor, track, and surveil citizens to accomplish a range of policy objectives—some lawful, others that violate human rights, and many of which fall into a murky middle ground.

In order to appropriately address the effects of this technology, it is important to first understand where these tools are being deployed and how they are being used.

To provide greater clarity, Carnegie presents an AI Global Surveillance (AIGS) Index—representing one of the first research efforts of its kind. The index compiles empirical data on AI surveillance use for 176 countries around the world. It does not distinguish between legitimate and unlawful uses of AI surveillance. Rather, the purpose of the research is to show how new surveillance capabilities are transforming the ability of governments to monitor and track individuals or systems. It specifically asks:

  • Which countries are adopting AI surveillance technology?
  • What specific types of AI surveillance are governments deploying?
  • Which countries and companies are supplying this technology?

Learn more about our findings and how AI surveillance technology is spreading rapidly around the globe….(More)”.

Algorithmic Censorship on Social Platforms: Power, Legitimacy, and Resistance

Curated on Posted on by Stefaan Verhulst

Paper by Jennifer Cobbe: “Effective content moderation by social platforms has long been recognised as both important and difficult, with numerous issues arising from the volume of information to be dealt with, the culturally sensitive and contextual nature of that information, and the nuances of human communication. Attempting to scale moderation efforts, various platforms have adopted, or signalled their intention to adopt, increasingly automated approaches to identifying and suppressing content and communications that they deem undesirable. However, algorithmic forms of online censorship by social platforms bring their own concerns, including the extensive surveillance of communications and the use of machine learning systems with the distinct possibility of errors and biases. This paper adopts a governmentality lens to examine algorithmic censorship by social platforms in order to assist in the development of a more comprehensive understanding of the risks of such approaches to content moderation. This analysis shows that algorithmic censorship is distinctive for two reasons: (1) it would potentially bring all communications carried out on social platforms within reach, and (2) it would potentially allow those platforms to take a much more active, interventionist approach to moderating those communications. Consequently, algorithmic censorship could allow social platforms to exercise an unprecedented degree of control over both public and private communications, with poor transparency, weak or non-existent accountability mechanisms, and little legitimacy. Moreover, commercial considerations would be inserted further into the everyday communications of billions of people. Due to the dominance of the web by a small number of social platforms, this control may be difficult or impractical to escape for many people, although opportunities for resistance do exist.

While automating content moderation may seem like an attractive proposition for both governments and platforms themselves, the issues identified in this paper are cause for concern and should be given serious consideration.Jennifer CobbeEffective content moderation by social platforms has long been recognised as both important and difficult, with numerous issues arising from the volume of information to be dealt with, the culturally sensitive and contextual nature of that information, and the nuances of human communication. Attempting to scale moderation efforts, various platforms have adopted, or signalled their intention to adopt, increasingly automated approaches to identifying and suppressing content and communications that they deem undesirable. However, algorithmic forms of online censorship by social platforms bring their own concerns, including the extensive surveillance of communications and the use of machine learning systems with the distinct possibility of errors and biases. This paper adopts a governmentality lens to examine algorithmic censorship by social platforms in order to assist in the development of a more comprehensive understanding of the risks of such approaches to content moderation.

This analysis shows that algorithmic censorship is distinctive for two reasons: (1) it would potentially bring all communications carried out on social platforms within reach, and (2) it would potentially allow those platforms to take a much more active, interventionist approach to moderating those communications. Consequently, algorithmic censorship could allow social platforms to exercise an unprecedented degree of control over both public and private communications, with poor transparency, weak or non-existent accountability mechanisms, and little legitimacy. Moreover, commercial considerations would be inserted further into the everyday communications of billions of people. Due to the dominance of the web by a small number of social platforms, this control may be difficult or impractical to escape for many people, although opportunities for resistance do exist. While automating content moderation may seem like an attractive proposition for both governments and platforms themselves, the issues identified in this paper are cause for concern and should be given serious consideration….(More)”.

How to Build Artificial Intelligence We Can Trust

Curated on Posted on by Stefaan Verhulst

Gary Marcus and Ernest Davis at the New York Times: “Artificial intelligence has a trust problem. We are relying on A.I. more and more, but it hasn’t yet earned our confidence.

Tesla cars driving in Autopilot mode, for example, have a troubling history of crashing into stopped vehicles. Amazon’s facial recognition system works great much of the time, but when asked to compare the faces of all 535 members of Congress with 25,000 public arrest photos, it found 28 matches, when in reality there were none. A computer program designed to vet job applicants for Amazon was discovered to systematically discriminate against women. Every month new weaknesses in A.I. are uncovered.

The problem is not that today’s A.I. needs to get better at what it does. The problem is that today’s A.I. needs to try to do something completely different.

In particular, we need to stop building computer systems that merely get better and better at detecting statistical patterns in data sets — often using an approach known as deep learning — and start building computer systems that from the moment of their assembly innately grasp three basic concepts: time, space and causality….

We face a choice. We can stick with today’s approach to A.I. and greatly restrict what the machines are allowed to do (lest we end up with autonomous-vehicle crashes and machines that perpetuate bias rather than reduce it). Or we can shift our approach to A.I. in the hope of developing machines that have a rich enough conceptual understanding of the world that we need not fear their operation. Anything else would be too risky….(More)”.

The Ethics of Hiding Your Data From the Machines

Curated on Posted on by Stefaan Verhulst

Molly Wood at Wired: “…But now that data is being used to train artificial intelligence, and the insights those future algorithms create could quite literally save lives.

So while targeted advertising is an easy villain, data-hogging artificial intelligence is a dangerously nuanced and highly sympathetic bad guy, like Erik Killmonger in Black Panther. And it won’t be easy to hate.

I recently met with a company that wants to do a sincerely good thing. They’ve created a sensor that pregnant women can wear, and it measures their contractions. It can reliably predict when women are going into labor, which can help reduce preterm births and C-sections. It can get women into care sooner, which can reduce both maternal and infant mortality.

All of this is an unquestionable good.

And this little device is also collecting a treasure trove of information about pregnancy and labor that is feeding into clinical research that could upend maternal care as we know it. Did you know that the way most obstetricians learn to track a woman’s progress through labor is based on a single study from the 1950s, involving 500 women, all of whom were white?…

To save the lives of pregnant women and their babies, researchers and doctors, and yes, startup CEOs and even artificial intelligence algorithms need data. To cure cancer, or at least offer personalized treatments that have a much higher possibility of saving lives, those same entities will need data….

And for we consumers, well, a blanket refusal to offer up our data to the AI gods isn’t necessarily the good choice either. I don’t want to be the person who refuses to contribute my genetic data via 23andMe to a massive research study that could, and I actually believe this is possible, lead to cures and treatments for diseases like Parkinson’s and Alzheimer’s and who knows what else.

I also think I deserve a realistic assessment of the potential for harm to find its way back to me, because I didn’t think through or wasn’t told all the potential implications of that choice—like how, let’s be honest, we all felt a little stung when we realized the 23andMe research would be through a partnership with drugmaker (and reliable drug price-hiker) GlaxoSmithKline. Drug companies, like targeted ads, are easy villains—even though this partnership actually couldproduce a Parkinson’s drug. But do we know what GSK’s privacy policy looks like? That deal was a level of sharing we didn’t necessarily expect….(More)”.