Bias and Fairness

Today’s post has AI and Policy news updates and our next installment on Bias and Policy: the fairness component.

News Items for February, 2020

  • OECD launched the OECD.AI Observatory, an online platform to shape and share AI policies across the globe. 
  • The White House released the American Artificial Intelligence Initiative:Year One Annual Report and supported the OECD policy.

Bias and Fairness

In terms of decision-making and policy, fairness can be defined as “the absence of any prejudice or favoritism towards an individual or a group based on their inherent or acquired characteristics”.  Six of the most used definitions are equalized odds, equal opportunity, demographic parity, fairness through unawareness or group unaware, treatment equality. 

The concept of equalized odds and equal opportunity is that individuals who qualify for a desirable outcome should have an equal chance of being correctly assigned regardless of an individual’s belonging to a protected or unprotected group (e.g., female/male). The additional concepts “demographic parity” and “group unaware” are illustrated by the Google visualization research team with nice visualizations using an example “simulating loan decisions for different groups”. The focus of equal opportunity is on the outcome of the true positive rate of the group.

On the other hand, the focus of the demographic parity is on the positive rate only. Consider a loan approval process for two groups: group A and group B. For demographic parity, the overall number of approved loans should be equal in both group A and group B regardless of a person belonging to a protected group. Since the focus for demographic parity is on overall loan approval rate, the rate should be equal for both the groups. Some people in group A who would pay back the loan might be disadvantaged compared to the people in group B who might not pay back the loan.  However, the people in group A will not be at a disadvantage in the equal opportunity concept, since this concept focuses on true positive rate. As an example of fairness through unawareness “an algorithm is fair as long as any protected attributes A are not explicitly used in the decision-making process”.

All of the fairness concepts or definitions either fall under individual fairness, subgroup fairness or group fairness. For example, demographic parity, equalized odds, and equal opportunity are the group fairness type; fairness through awareness falls under the individual type where the focus is not on the overall group.

A definition of bias can be in three categories: data, algorithmic, and user interaction feedback loop:
Data — behavioral bias, presentation bias, linking bias, and content production bias;
Algoritmic — historical bias, aggregation bias, temporal bias, and social bias falls
User Interaction — popularity bias, ranking bias, evaluation bias, and emergent bias.

Bias is a large domain with much to explore and take into consideration. Bias and public policy will be further discussed in future blog posts.

This series of posts on Bias has been co-authored by Farhana Faruqe, doctoral student in the GWU Human-Technology Collaboration group.

References 
 [1] Ninareh Mehrabi, Fred Morstatter, Nripsuta Saxena, Kristina Lerman, and Aram Galstyan. A survey on bias and fairness in machine learning. CoRR, abs/1908.09635, 2019.
[2] Moritz Hardt, Eric Price, , and Nati Srebro. 2016. Equality of Opportunity in Supervised Learning. In Advances in Neural Information Processing Systems 29, D. D. Lee, M. Sugiyama, U. V. Luxburg, I. Guyon, and R. Garnett (Eds.). Curran Associates, Inc., 3315–3323. http://papers.nips.cc/paper/ 6374-equality-of-opportunity-in-supervised-learning.pdf
[3] Martin Wattenberg, Fernanda Viegas, and Moritz Hardt. Attacking discrimination with smarter machine learning. Accessed at https://research.google.com/bigpicture/attacking-discrimination-in-ml/, 2016

Discrimination and Bias

Our current public policy posts, focused on ethics and bias in current and emerging areas of AI, build on the work “A Survey on Bias and Fairness in Machine Learning” by Ninareh Mehrabi, et al. and resources provided by Barocas, et al. The guest co-author of this series of blog posts on AI and bias is Farhana Faruqe, doctoral student in the George Washington University Human-Technology Collaboration program. We look forward to your comments and suggestions.

Discrimination, unfairness, and bias are terms used frequently these days in the context of AI and data science applications that make decisions in the everyday lives of individuals and groups. Machine learning applications depend on datasets that are usually a reflection of our real world in which individuals have intentional and unintentional biases that may cause discrimination and unfair actions. Broadly, fairness is the absence of any prejudice or favoritism towards an individual or a group based on their intrinsic or acquired traits in the context of decision-making.

Today’s blog post focuses on discrimination, which Ninareh Mehrabi, et al. describe as follows:

Direct Discrimination: “Direct discrimination happens when protected attributes of individuals explicitly result in non-favorable outcomes toward them.”  Some traits like race, color, national origin, religion, sex, family status, disability, marital status, recipient of public assistance, and age are identified as sensitive attributes or protected attributes in the machine learning world.  It is not legal to discriminate against these sensitive attributes, which are listed by the FHA and Equal Credit Opportunity Act (ECOA).                

Indirect Discrimination: Even if sensitive or protected attributes are not used against an individual, still indirect discrimination can happen. For example, residential zip code is not categorized as a protected attribute, but from the zip code one may find out about race which is a protected attribute. So, “protected groups or individuals still can get treated unjustly as a result of implicit effects from their protected attributes.”

Systemic Discrimination. In the nursing profession, the custom is to expect a nurse to be a woman. So, excluding qualified male nurses for nursing position is an example of systematic discrimination. Systematic discrimination is defined as “policies, customs, or behaviors that are a part of the culture or structure of an organization that may perpetuate discrimination against certain subgroups of the population”.                                                                                                                              
Statistical Discrimination: In law enforcement, racial profiling is an example of statistical discrimination. In this case, minority drivers are pulled over more often than white drivers. The authors define “statistical discrimination is a phenomenon where decision-makers use average group statistics to judge an individual belonging to that group.”

Explainable Discrimination: In some cases, “discrimination can be explained using attributes” like working hours and education, which is legal and acceptable as well. In a widely used dataset in the fairness domain, males on average have a higher annual income than females because on average females work fewer hours per week than males do. Decisions made without considering working hours could lead to discrimination.                     

Unexplainable Discrimination: This type of discrimination is not legal as explainable discrimination because “the discrimination toward a group is unjustified”. Some researchers have introduced techniques during data preprocessing and training to remove unexplainable discrimination.   

To understand bias in techniques such as machine learning, we will discuss in our next blog post another important aspect: fairness.

Bias, Ethics, and Policy

We are planning a series of posts on Bias, starting with the background and context of bias in general and then focusing on specific instances of bias in current and emerging areas of AI. Ultimately, this information is intended to inform ideas on public policy. We look forward to your comments and suggestions for a robust discussion.

Extensive work “A Survey on Bias and Fairness in Machine Learning” by Ninareh Mehrabi et al. will be useful for the conversation. The guest co-author of the ACM SIGAI Public Policy blog posts on Bias will be Farhana Faruqe, doctoral student in the George Washington University Human-Technology Collaboration program.

A related announcement is about the new section on AI and Ethics in the Springer Nature Computer Science journal. “The AI & Ethics section focuses on how AI techniques, tools, and technologies are developing, including consideration of where these developments may lead in the future. It seeks to promote informed debate and discussion of the current and future developments in AI, and the ethical, moral, regulatory, and policy implications that arise from these developments.” As a Co-Editor of the new section, I welcome you to submit a manuscript and contact me with any questions and suggestions.

AI Revolution or Evolution

An interesting IEEE Spectrum article “AI and Economic Productivity: Expect Evolution, Not Revolution” by Jeffrey Funk questions popular claims about the rapid pace of AI’s impact on productivity and the economy. He asserts that “Despite the hype, artificial intelligence will take years to significantly boost economic productivity”. If correct, this will have serious implications for public policymaking who have chosen due be proactive. The article raises good points, but many of the examples do not look like real AI, at least as a dominant component. Putting “smart” in the name of a product doesn’t make it AI, and automation doesn’t necessarily use AI. 

On a broader note, we should care about the technology language we use and be aware of the usual practices in commercialization. As discussed in previous blog posts, expanding too far the meanings of terms like AI, machine learning, and algorithms makes rational discourse more difficult. Some of us remember marketing of expert systems and relational databases: companies do a disservice to society by claiming each breakthrough technology actually is in their products. Here we go again — today about anything counts as AI depending on the point you want to make and the products you want to sell. 

Another issue raised by the article is from the emphasis on startups as the leaders of economic impact, as opposed to the results of innovations from established industry and government labs. Technologies have adoption curves, going from early adopters through the laggards, of about seven years. If you add to that the difficulties of making a startup succeed, a decade or so is probably the minimum timescale for large impact on the economy. A better perspective on revolution versus evolution could come from longitudinal evaluations looking at trends. In that case, a good endpoint for a hypothesis about dramatic impact on productivity might be the 2030-2035 timeframe. 

A problem with using a vague or broad notion of AI is that policymakers could miss the revolutionary impact of data science, which can, but may not, involve real AI. Data science probably has the best chance of dramatically impacting society and the economy in the short and long terms and has the advantage of not having to involve designing and manufacturing physical objects, and thus not always having to wait for consumers to adopt new products. Data Science is already affecting society and employment with obvious, and not so obvious, revolutionary impacts on our lives.

PCAST and AI Plan

Executive Order on The President’s Council of Advisors on Science and Technology (PCAST)

President Trump issued an executive order on October 22 re-establishing the President’s Council of Advisors on Science and Technology (PCAST), an advisory body that consists of science and technology leaders from the private and academic sectors. PCAST is to be chaired by Kelvin Droegemeier, director of the Office of Science and Technology Policy, and Edward McGinnis, formerly with DOE, is to serve as the executive director. The majority of the 16 members are from key industry sectors. The executive order says that the council is expected to address “strengthening American leadership in science and technology, building the Workforce of the Future, and supporting foundational research and development across the country.” For more information, see the Inside Education article about the first appointments.

Schumer AI Plan

Jeffrey Mervis has a November 11, 2019, article in AAAS News from Science on a recommendation for the government to create a new agency funded with $100 billion over 5 years for basic AI research. “Senator Charles Schumer (D–NY) says the initiative would enable the United States to keep pace with China and Russia in a critical research arena and plug gaps in what U.S. companies are unwilling to finance.”

Schumer gave his ideas publicly in a speech in early November to senior national security and research policymakers following a recent presidential executive order. He wants to create a new national science tech fund looking into “fundamental research related to AI and some other cutting-edge areas” such as quantum computing, 5G networks, robotics, cybersecurity, and biotechnology. Funds would encourage research at U.S. universities, companies, and other federal agencies and support incubators for moving research into commercial products. An additional article can be found in Defense News.

Work Transition

AI and other automation technologies have great promise for benefitting society and enhancing productivity, but appropriate policies by companies and governments are needed to help manage workforce transitions and make them as smooth as possible. The McKinsey Global Institute report AI, automation, and the future of work: Ten things to solve for states that “There is work for everyone today and there will be work for everyone tomorrow, even in a future with automation. Yet that work will be different, requiring new skills, and a far greater adaptability of the workforce than we have seen. Training and retraining both mid-career workers and new generations for the coming challenges will be an imperative. Government, private-sector leaders, and innovators all need to work together to better coordinate public and private initiatives, including creating the right incentives to invest more in human capital. The future with automation and AI will be challenging, but a much richer one if we harness the technologies with aplomb—and mitigate the negative effects.” They list likely actionable and scalable solutions in several key areas:

1. Ensuring robust economic and productivity growth

2. Fostering business dynamism

3. Evolving education systems and learning for a changed workplace

4. Investing in human capital

5. Improving labor-market dynamism

6. Redesigning work

7. Rethinking incomes

8. Rethinking transition support and safety nets for workers affected

9. Investing in drivers of demand for work

10. Embracing AI and automation safely

In redesigning work and rethinking incomes, we have the chance for bold ideas that improve the lives of workers and give them more interesting jobs that could provide meaning, purpose, and dignity.

Some of the categories of new jobs that could replace old jobs are
1. Making, designing, and coding in AI, data science, and engineering occupations
2. Working in new types of non-AI jobs that are enhanced by AI, making unpleasant old jobs more palatable or providing new jobs that are more interesting; the gig economy and crowdsourcing ideas are examples that could provide creative employment options
3. Providing living wages for people to do things they love; for example, in the arts through dramatic funding increases for NEA and NEH programs. Grants to individual artists and musicians, professional and amateur musical organizations, and informal arts education initiatives could enrich communities while providing income for millions of people. Policies to implement this idea could be one piece of the future-of-work puzzle and be much more preferable for the economy and society than allowing large-scale unemployment due to loss of work from automation.

National AI Strategy

The National Artificial Intelligence Research and Development Strategic Plan – an update of the report by the Select Committee on Artificial Intelligence of The National Science & Technology Council – was released in June, 2019, and the President’s, Executive Order 13859 Maintaining American Leadership in Artificial Intelligence was released on February 11. The Computing Community Consortium (CCC) recently released the AI Roadmap Website, and an interesting industry response is “Intel Gets Specific on a National Strategy for AI, “How to Propel the US into a Sustainable Leadership Position on the Global Artificial Intelligence Stage” By Naveen Rao and David Hoffman. Excerpts follow and the accompanying links provide the details:

“AI is more than a matter of making good technology; it is also a matter of making good policy. And that’s what a robust national AI strategy will do: continue to unlock the potential of AI, prepare for AI’s many ramifications, and keep the U.S. among leading AI countries. At least 20 other countries have published, and often funded, their national AI strategies. Last month, the administration signaled its commitment to U.S. leadership in AI by issuing an executive order to launch the American AI Initiative, focusing federal government resources to develop AI. Now it’s time to take the next step and bring industry and government together to develop a fully realized U.S. national strategy to continue leading AI innovation.

“… But to sustain leadership and effectively manage the broad social implications of AI, the U.S. needs coordination across government, academia, industry and civil society. This challenge is too big for silos, and it requires that technologists and policymakers work together and understand each other’s worlds.” Their call to action was released in May 2018.

Four Key Pillars

“Our recommendation for a national AI strategy lays out four key responsibilities for government. Within each of these areas we propose actionable steps. We provide some highlights here, and we encourage you to read the full white paper or scan the shorter fact sheet.

Sustainable and funded government AI research and development can help to advance the capabilities of AI in areas such as healthcare, cybersecurity, national security and education, but there need to be clear ethical guidelines.

Create new employment opportunities and protect people’s welfare given that AI has the potential to automate certain work activities.

Liberate and share data responsibly, as the more data that is available, the more “intelligent” an AI system can become. But we need guardrails.

Remove barriers and create a legal and policy environment that supports AI so that the responsible development and use of AI is not inadvertently derailed.”

AI Race Matters

China, the European Union, and the United States have been in the news about strategic plans and policies on the future of AI. The July 2 AI Matters policy blog post was on the U.S. National Artificial Intelligence Research and Development Strategic Plan, released in June, as an update of the report by the Select Committee on Artificial Intelligence of The National Science & Technology Council. The Computing Community Consortium (CCC) recently released the AI Roadmap Website.
Now, a Center for Data Innovation Report compares the current standings of China, the European Union, and the United States and makes policy recommendations. Here is the report summary: “Many nations are racing to achieve a global innovation advantage in artificial intelligence (AI) because they understand that AI is a foundational technology that can boost competitiveness, increase productivity, protect national security, and help solve societal challenges. This report compares China, the European Union, and the United States in terms of their relative standing in the AI economy by examining six categories of metrics—talent, research, development, adoption, data, and hardware. It finds that despite China’s bold AI initiative, the United States still leads in absolute terms. China comes in second, and the European Union lags further behind. This order could change in coming years as China appears to be making more rapid progress than either the United States or the European Union. Nonetheless, when controlling for the size of the labor force in the three regions, the current U.S. lead becomes even larger, while China drops to third place, behind the European Union. This report also offers a range of policy recommendations to help each nation or region improve its AI capabilities.”

About Face

Face recognition R&D has made great progress in recent years and has been prominent in the news. In public policy many are calling for a reversal of the trajectory for FR systems and products. In the hands of people of good will – using products designed for safety and training systems with appropriate data – society and individuals could have a better life. The Verge reports China’s use of unique facial markings of pandas to identify individual animals. FR research includes work to mitigate negative outcomes, as with the Adobe and UC Berkeley work on Detecting Facial Manipulations in Adobe Photoshop: automatic detect when images of faces have been manipulated by use of splicing, cloning, and removing an object.

Intentional and unintentional application of systems that are not designed and trained for ethical use are a threat to society. Screening for terrorists could be good, but FR lie and fraud detection systems may not work properly. The safety of FR is currently an important issue for policymakers, but regulations could have negative consequences for AI researchers. As with many contemporary issues, conflicts arise because of conflicting policies in different countries.

Recent and current legislation is attempting to restrict FR the use and possibly research.
* San Francisco, CA and Somerville, MA, and Oakland, CA, are the first three cities to limit use of FR to identify people.
* “Facial recognition may be banned from public housing thanks to proposed law” – CNET reports that a bill will be introduced to address the issue that “… landlords across the country continue to install smart home technology and tenants worry about unchecked surveillance, there’s been growing concern about facial recognition arriving at people’s doorsteps.”
* The major social media companies are being pressed on “how they plan to handle the threat of deepfake images and videos on their platforms ahead of the 2020 elections.”
* A call for a more comprehensive ban on FR has been launched by the digital rights group Fight for the Future, seeking a complete Federal ban on government use of facial recognition surveillance.

Beyond legislation against FR research and banning certain products, work is in progress to enable safe and ethical use of FR. A more general example that could be applied to FR is the MITRE work The Ethical Framework for the Use of Consumer-Generated Data in Health Care, which “establishes ethical values, principles, and guidelines to guide the use of Consumer-Generated Data for health care purposes.”