Using artificial intelligence to make decisions: Addressing the problem of algorithmic bias (2020)

Curated on Posted on by Stefaan Verhulst

Foreword of a Report by the Australian Human Rights Commission: “Artificial intelligence (AI) promises better, smarter decision making.

Governments are starting to use AI to make decisions in welfare, policing and law enforcement, immigration, and many other areas. Meanwhile, the private sector is already using AI to make decisions about pricing and risk, to determine what sorts of people make the ‘best’ customers… In fact, the use cases for AI are limited only by our imagination.

However, using AI carries with it the risk of algorithmic bias. Unless we fully understand and address this risk, the promise of AI will be hollow.

Algorithmic bias is a kind of error associated with the use of AI in decision making, and often results in unfairness. Algorithmic bias can arise in many ways. Sometimes the problem is with the design of the AI-powered decision-making tool itself. Sometimes the problem lies with the data set that was used to train the AI tool, which could replicate or even make worse existing problems, including societal inequality.

Algorithmic bias can cause real harm. It can lead to a person being unfairly treated, or even suffering unlawful discrimination, on the basis of characteristics such as their race, age, sex or disability.

This project started by simulating a typical decision-making process. In this technical paper, we explore how algorithmic bias can ‘creep in’ to AI systems and, most importantly, how this problem can be addressed.

To ground our discussion, we chose a hypothetical scenario: an electricity retailer uses an AI-powered tool to decide how to offer its products to customers, and on what terms. The general principles and solutions for mitigating the problem, however, will be relevant far beyond this specific situation.

Because algorithmic bias can result in unlawful activity, there is a legal imperative to address this risk. However, good businesses go further than the bare minimum legal requirements, to ensure they always act ethically and do not jeopardise their good name.

Rigorous design, testing and monitoring can avoid algorithmic bias. This technical paper offers some guidance for companies to ensure that when they use AI, their decisions are fair, accurate and comply with human rights….(More)”

Facial-recognition research needs an ethical reckoning

Curated on Posted on by Stefaan Verhulst

Editorial in Nature: “…As Nature reports in a series of Features on facial recognition this week, many in the field are rightly worried about how the technology is being used. They know that their work enables people to be easily identified, and therefore targeted, on an unprecedented scale. Some scientists are analysing the inaccuracies and biases inherent in facial-recognition technology, warning of discrimination, and joining the campaigners calling for stronger regulation, greater transparency, consultation with the communities that are being monitored by cameras — and for use of the technology to be suspended while lawmakers reconsider where and how it should be used. The technology might well have benefits, but these need to be assessed against the risks, which is why it needs to be properly and carefully regulated.Is facial recognition too biased to be let loose?

Responsible studies

Some scientists are urging a rethink of ethics in the field of facial-recognition research, too. They are arguing, for example, that scientists should not be doing certain types of research. Many are angry about academic studies that sought to study the faces of people from vulnerable groups, such as the Uyghur population in China, whom the government has subjected to surveillance and detained on a mass scale.

Others have condemned papers that sought to classify faces by scientifically and ethically dubious measures such as criminality….One problem is that AI guidance tends to consist of principles that aren’t easily translated into practice. Last year, the philosopher Brent Mittelstadt at the University of Oxford, UK, noted that at least 84 AI ethics initiatives had produced high-level principles on both the ethical development and deployment of AI (B. Mittelstadt Nature Mach. Intell. 1, 501–507; 2019). These tended to converge around classical medical-ethics concepts, such as respect for human autonomy, the prevention of harm, fairness and explicability (or transparency). But Mittelstadt pointed out that different cultures disagree fundamentally on what principles such as ‘fairness’ or ‘respect for autonomy’ actually mean in practice. Medicine has internationally agreed norms for preventing harm to patients, and robust accountability mechanisms. AI lacks these, Mittelstadt noted. Specific case studies and worked examples would be much more helpful to prevent ethics guidance becoming little more than window-dressing….(More)”.

Evaluating Identity Disclosure Risk in Fully Synthetic Health Data: Model Development and Validation

Curated on Posted on by Stefaan Verhulst

Paper by Khaled El Emam et al: “There has been growing interest in data synthesis for enabling the sharing of data for secondary analysis; however, there is a need for a comprehensive privacy risk model for fully synthetic data: If the generative models have been overfit, then it is possible to identify individuals from synthetic data and learn something new about them.

Objective: The purpose of this study is to develop and apply a methodology for evaluating the identity disclosure risks of fully synthetic data.

Methods: A full risk model is presented, which evaluates both identity disclosure and the ability of an adversary to learn something new if there is a match between a synthetic record and a real person. We term this “meaningful identity disclosure risk.” The model is applied on samples from the Washington State Hospital discharge database (2007) and the Canadian COVID-19 cases database. Both of these datasets were synthesized using a sequential decision tree process commonly used to synthesize health and social science data.

Results: The meaningful identity disclosure risk for both of these synthesized samples was below the commonly used 0.09 risk threshold (0.0198 and 0.0086, respectively), and 4 times and 5 times lower than the risk values for the original datasets, respectively.

Conclusions: We have presented a comprehensive identity disclosure risk model for fully synthetic data. The results for this synthesis method on 2 datasets demonstrate that synthesis can reduce meaningful identity disclosure risks considerably. The risk model can be applied in the future to evaluate the privacy of fully synthetic data….(More)”.

Algorithmic governance: A modes of governance approach

Curated on Posted on by Stefaan Verhulst

Article by Daria Gritsenko and Matthew Wood: “This article examines how modes of governance are reconfigured as a result of using algorithms in the governance process. We argue that deploying algorithmic systems creates a shift toward a special form of design‐based governance, with power exercised ex ante via choice architectures defined through protocols, requiring lower levels of commitment from governing actors. We use governance of three policy problems – speeding, disinformation, and social sharing – to illustrate what happens when algorithms are deployed to enable coordination in modes of hierarchical governance, self‐governance, and co‐governance. Our analysis shows that algorithms increase efficiency while decreasing the space for governing actors’ discretion. Furthermore, we compare the effects of algorithms in each of these cases and explore sources of convergence and divergence between the governance modes. We suggest design‐based governance modes that rely on algorithmic systems might be re‐conceptualized as algorithmic governance to account for the prevalence of algorithms and the significance of their effects….(More)”.

The political choreography of the Sophia robot: beyond robot rights and citizenship to political performances for the social robotics market

Curated on Posted on by Stefaan Verhulst

Paper by A humanoid robot named ‘Sophia’ has sparked controversy since it has been given citizenship and has done media performances all over the world. The company that made the robot, Hanson Robotics, has touted Sophia as the future of artificial intelligence (AI). Robot scientists and philosophers have been more pessimistic about its capabilities, describing Sophia as a sophisticated puppet or chatbot. Looking behind the rhetoric about Sophia’s citizenship and intelligence and going beyond recent discussions on the moral status or legal personhood of AI robots, we analyse the performativity of Sophia from the perspective of what we call ‘political choreography’: drawing on phenomenological approaches to performance-oriented philosophy of technology. This paper proposes to interpret and discuss the world tour of Sophia as a political choreography that boosts the rise of the social robot market, rather than a statement about robot citizenship or artificial intelligence. We argue that the media performances of the Sophia robot were choreographed to advance specific political interests. We illustrate our philosophical discussion with media material of the Sophia performance, which helps us to explore the mechanisms through which the media spectacle functions hand in hand with advancing the economic interests of technology industries and their governmental promotors. Using a phenomenological approach and attending to the movement of robots, we also criticize the notion of ‘embodied intelligence’ used in the context of social robotics and AI. In this way, we put the discussions about the robot’s rights or citizenship in the context of AI politics and economics….(More)”

Extending the framework of algorithmic regulation. The Uber case

Curated on Posted on by Stefaan Verhulst

Paper by Florian Eyert, Florian Irgmaier, and Lena Ulbricht: “In this article, we take forward recent initiatives to assess regulation based on contemporary computer technologies such as big data and artificial intelligence. In order to characterize current phenomena of regulation in the digital age, we build on Karen Yeung’s concept of “algorithmic regulation,” extending it by building bridges to the fields of quantification, classification, and evaluation research, as well as to science and technology studies. This allows us to develop a more fine‐grained conceptual framework that analyzes the three components of algorithmic regulation as representation, direction, and intervention and proposes subdimensions for each. Based on a case study of the algorithmic regulation of Uber drivers, we show the usefulness of the framework for assessing regulation in the digital age and as a starting point for critique and alternative models of algorithmic regulation….(More)”.

Four Principles to Make Data Tools Work Better for Kids and Families

Curated on Posted on by Stefaan Verhulst

Blog by the Annie E. Casey Foundation: “Advanced data analytics are deeply embedded in the operations of public and private institutions and shape the opportunities available to youth and families. Whether these tools benefit or harm communities depends on their design, use and oversight, according to a report from the Annie E. Casey Foundation.

Four Principles to Make Advanced Data Analytics Work for Children and Families examines the growing field of advanced data analytics and offers guidance to steer the use of big data in social programs and policy….

The Foundation report identifies four principles — complete with examples and recommendations — to help steer the growing field of data science in the right direction.

Four Principles for Data Tools

  1. Expand opportunity for children and families. Most established uses of advanced analytics in education, social services and criminal justice focus on problems facing youth and families. Promising uses of advanced analytics go beyond mitigating harm and help to identify so-called odds beaters and new opportunities for youth.
    • Example: The Children’s Data Network at the University of Southern California is helping the state’s departments of education and social services explore why some students succeed despite negative experiences and what protective factors merit more investment.
    • Recommendation: Government and its philanthropic partners need to test if novel data science applications can create new insights and when it’s best to apply them.
       
  2. Provide transparency and evidence. Advanced analytical tools must earn and maintain a social license to operate. The public has a right to know what decisions these tools are informing or automating, how they have been independently validated, and who is accountable for answering and addressing concerns about how they work.
    • Recommendations: Local and state task forces can be excellent laboratories for testing how to engage youth and communities in discussions about advanced analytics applications and the policy frameworks needed to regulate their use. In addition, public and private funders should avoid supporting private algorithms whose design and performance are shielded by trade secrecy claims. Instead, they should fund and promote efforts to develop, evaluate and adapt transparent and effective models.
       
  3. Empower communities. The field of advanced data analytics often treats children and families as clients, patients and consumers. Put to better use, these same tools can help elucidate and reform the systems acting upon children and families. For this shift to occur, institutions must focus analyses and risk assessments on structural barriers to opportunity rather than individual profiles.
    • Recommendation: In debates about the use of data science, greater investment is needed to amplify the voices of youth and their communities.
       
  4. Promote equitable outcomes. Useful advanced analytics tools should promote more equitable outcomes for historically disadvantaged groups. New investments in advanced analytics are only worthwhile if they aim to correct the well-documented bias embedded in existing models.
    • Recommendations: Advanced analytical tools should only be introduced when they reduce the opportunity deficit for disadvantaged groups — a move that will take organizing and advocacy to establish and new policy development to institutionalize. Philanthropy and government also have roles to play in helping communities test and improve tools and examples that already exist….(More)”.

How Can Policy Makers Predict the Unpredictable?

Curated on Posted on by Stefaan Verhulst

Essay by Meg King and Aaron Shull: “Policy makers around the world are leaning on historical analogies to try to predict how artificial intelligence, or AI — which, ironically, is itself a prediction technology — will develop. They are searching for clues to inform and create appropriate policies to help foster innovation while addressing possible security risks. Much in the way that electrical power completely changed our world more than a century ago — transforming every industry from transportation to health care to manufacturing — AI’s power could effect similar, if not even greater, disruption.

Whether it is the “next electricity” or not, one fact all can agree on is that AI is not a thing in itself. Most authors contributing to this essay series focus on the concept that AI is a general-purpose technology — or GPT — that will enable many applications across a variety of sectors. While AI applications are expected to have a significantly positive impact on our lives, those same applications will also likely be abused or manipulated by bad actors. Setting rules at both the national and the international level — in careful consultation with industry — will be crucial for ensuring that AI offers new capabilities and efficiencies safely.

Situating this discussion, though, requires a look back, in order to determine where we may be going. While AI is not new — Marvin Minsky developed what is widely believed to be the first neural network learning machine in the early 1950s — its scale, scope, speed of adoption and potential use cases today highlight a number of new challenges. There are now many ominous signs pointing to extreme danger should AI be deployed in an unchecked manner, particularly in military applications, as well as worrying trends in the commercial context related to potential discrimination, undermining of privacy, and upended traditional employment structures and economic models….(More)”

The necessity of judgment

Curated on Posted on by Stefaan Verhulst

Essay by Jeff Malpas in AI and Society: “In 2016, the Australian Government launched an automated debt recovery system through Centrelink—its Department of Human Services. The system, which came to be known as ‘Robodebt’, matched the tax records of welfare recipients with their declared incomes as held by Ethe Department and then sent out debt notices to recipients demanding payment. The entire system was computerized, and many of those receiving debt notices complained that the demands for repayment they received were false or inaccurate as well as unreasonable—all the more so given that those being targeted were, almost by definition, those in already vulnerable circumstances. The system provoked enormous public outrage, was subjected to successful legal challenge, and after being declared unlawful, the Government paid back all of the payments that had been received, and eventually, after much prompting, issued an apology.

The Robodebt affair is characteristic of a more general tendency to shift to systems of automated decision-making across both the public and the private sector and to do so even when those systems are flawed and known to be so. On the face of it, this shift is driven by the belief that automated systems have the capacity to deliver greater efficiencies and economies—in the Robodebt case, to reduce costs by recouping and reducing social welfare payments. In fact, the shift is characteristic of a particular alliance between digital technology and a certain form of contemporary bureaucratised capitalism. In the case of the automated systems we see in governmental and corporate contexts—and in many large organisations—automation is a result both of the desire on the part of software, IT, and consultancy firms to increase their customer base as well as expand the scope of their products and sales, and of the desire on the part of governments and organisations to increase control at the same time as they reduce their reliance on human judgment and capacity. The fact is, such systems seldom deliver the efficiencies or economies they are assumed to bring, and they also give rise to significant additional costs in terms of their broader impact and consequences, but the imperatives of sales and seemingly increased control (as well as an irrational belief in the benefits of technological solutions) over-ride any other consideration. The turn towards automated systems like Robodebt is, as is now widely recognised, a common feature of contemporary society. To look to a completely different domain, new military technologies are being developed to provide drone weapon systems with the capacity to identify potential threats and defend themselves against them. The development is spawning a whole new field of military ethics-based entirely around the putative ‘right to self-defence’ of automated weapon systems.

In both cases, the drone weapon system and Robodebt, we have instances of the development of automated systems that seem to allow for a form of ‘judgment’ that appears to operate independently of human judgment—hence the emphasis on this systems as autonomous. One might argue—and typically it is so argued—that any flaws that such systems currently present can be overcome either through the provision of more accurate information or through the development of more complex forms of artificial intelligence….(More)”.

AI’s Wide Open: A.I. Technology and Public Policy

Curated on Posted on by Stefaan Verhulst

Paper by Lauren Rhue and Anne L. Washington: “Artificial intelligence promises predictions and data analysis to support efficient solutions for emerging problems. Yet, quickly deploying AI comes with a set of risks. Premature artificial intelligence may pass internal tests but has little resilience under normal operating conditions. This Article will argue that regulation of early and emerging artificial intelligence systems must address the management choices that lead to releasing the system into production. First, we present examples of premature systems in the Boeing 737 Max, the 2020 coronavirus pandemic public health response, and autonomous vehicle technology. Second, the analysis highlights relevant management practices found in our examples of premature AI. Our analysis suggests that redundancy is critical to protecting the public interest. Third, we offer three points of context for premature AI to better assess the role of management practices.

AI in the public interest should: 1) include many sensors and signals; 2) emerge from a broad range of sources; and 3) be legible to the last person in the chain. Finally, this Article will close with a series of policy suggestions based on this analysis. As we develop regulation for artificial intelligence, we need to cast a wide net to identify how problems develop within the technologies and through organizational structures….(More)”.