Stefaan Verhulst

Matt Bencke at TechCrunch: “Most artificial intelligence models are built and trained by humans, and therefore have the potential to learn, perpetuate and massively scale the human trainers’ biases. This is the word of warning put forth in two illuminating articles published earlier this year by Jack Clark at Bloomberg and Kate Crawford at The New York Times.

Tl;dr: The AI field lacks diversity — even more spectacularly than most of our software industry. When an AI practitioner builds a data set on which to train his or her algorithm, it is likely that the data set will only represent one worldview: the practitioner’s. The resulting AImodel demonstrates a non-diverse “intelligence” at best, and a biased or even offensive one at worst….

So what happens when you don’t consider carefully who is annotating the data? What happens when you don’t account for the differing preferences, tendencies and biases among varying humans? We ran a fun experiment to find out….Actually, we didn’t set out to run an experiment. We just wanted to create something fun that we thought our awesome tasking community would enjoy. The idea? Give people the chance to rate puppies’ cuteness in their spare time…There was a clear gender gap — a very consistent pattern of women rating the puppies as cuter than the men did. The gap between women’s and men’s ratings was more narrow for the “less-cute” (ouch!) dogs, and wider for the cuter ones. Fascinating.

I won’t even try to unpack the societal implications of these findings, but the lesson here is this: If you’re training an artificial intelligence model — especially one that you want to be able to perform subjective tasks — there are three areas in which you must evaluate and consider demographics and diversity:

• yourself
• your data
• your annotators

This was a simple example: binary gender differences explaining one subjective numeric measure of an image. Yet it was unexpected and significant. As our industry deploys incredibly complex models that are pushing to the limit chip sets, algorithms and scientists, we risk reinforcing subtle biases, powerfully and at a previously unimaginable scale. Even more pernicious, many AIs reinforce their own learning, so we need to carefully consider “supervised” (aka human) re-training over time.

Artificial intelligence promises to change all of our lives — and it already subtly guides the way we shop, date, navigate, invest and more. But to make sure that it does so for the better, all of us practitioners need to go out of our way to be inclusive. We need to remain keenly aware of what makes us all, well… human. Especially the subtle, hidden stuff….(More)”

Scott Calvert at the Wall Street Journal: “Several states, including Arizona, Kentucky and Ohio, are using their state prescription monitoring databases to send doctors individualized “report cards” that show how their prescribing of addictive opioids and other drugs compares with their peers.

“Arizona probably has the most complete one out there right now—it’s pretty impressive,” said Patrick Knue, director of the Prescription Drug Monitoring Program Training and Technical Assistance Center at Brandeis University, which helps states improve their databases.

Arizona’s quarterly reports rate a doctor’s prescribing of oxycodone and certain other drugs as normal, high, severe or extreme compared with the state’s other doctors in his medical specialty.

During a two-year pilot program, the number of opiate prescriptions fell 10% in five counties while rising in other counties, said Dean Wright, former head of the state’s prescription-monitoring program. The report cards also contributed to a 4% drop in overdose deaths in the pilot counties, he said.

The state now issues the report cards statewide and in June sent notices to more than 13,000 doctors statewide. Mr. Wright said the message is clear: “Stop and think about what you’re prescribing and the impact it can have.”
The report cards list statistics such as how many of a doctor’s patients received controlled substances from five or more doctors. Elizabeth Dodge, Mr. Wright’s successor, said some doctors ask for the patients’ names—information they might have gleaned from the database….(More)”

Topic guide by Liz Carolan: “…introduces evidence and lessons learned about open data, transparency and accountability in the international development context. It discusses the definitions, theories, challenges and debates presented by the relationship between these concepts, summarises the current state of open data implementation in international development, and highlights lessons and resources for designing and implementing open data programmes.

Open data involves the release of data so that anyone can access, use and share it. The Open DataCharter (2015) describes six principles that aim to make data easier to find, use and combine:

• open by default
• timely and comprehensive
• accessible and usable
• comparable and interoperable
• for improved governance and citizen engagement
• for inclusive development and innovation

One of the main objectives of making data open is to promote transparency.

Transparency is a characteristic of government, companies, organisations and individuals that are open in the clear disclosure of information, rules, plans, processes and actions. Trans­parency of information is a crucial part of this. Within a development context, transparency and accountability initiatives have emerged over the last decade as a way to address developmental failures and democratic deficits.

There is a strong intersection between open data and transparency as concepts, yet as fields of study and practice, they have remained somewhat separate. This guide draws extensively on analysis and evidence from both sets of literature, beginning by outlining the main concepts and the theories behind the relationships between them.

Data release and transparency are parts of the chain of events leading to accountability.  For open data and transparency initiatives to lead to accountability, the required conditions include:

• getting the right data published, which requires an understanding of the politics of data publication
• enabling actors to find, process and use information, and to act on any outputs, which requires an accountability ecosystem that includes equipped and empowered intermediaries
• enabling institutional or social forms of enforceability or citizens’ ability to choose better services,which requires infrastructure that can impose sanctions, or sufficient choice or official support for citizens

Programmes intended to increase access to information can be impacted by and can affect inequality. They can also pose risks to privacy and may enable the misuse of data for the exploitation of individuals and markets.

Despite a range of international open data initiatives and pressures, developing countries are lagging behind in the implementation of reforms at government level, in the overall availability of data, and in the use of open data for transparency and accountability. What is more, there are signs that ‘open-washing’ –superficial efforts to publish data without full integration with transparency commitments – may be obscuring backsliding in other aspects of accountability.

The topic guide pulls together lessons and guidance from open data, transparency and accountability work,including an outline of technical and non-technical aspects of implementing a government open data initiative. It also lists further resources, tools and guidance….(More)”

Jonathan Bright and Helen Margetts at Policy & Society: “This editorial introduces a special issue resulting from a panel on Internet and policy organized by the Oxford Internet Institute (University of Oxford) at the 2015 International Conference on Public Policy (ICPP) held in Milan. Two main themes emerged from the panel: the challenges of high cost and low participation which many e-participation initiatives have faced; and the potential Big Data seems to hold for remedying these problems. This introduction briefly presents these themes and links them to the papers in the issue. It argues that Big Data can fix some of the problems typically encountered by e-participation initiatives: it can offer a solution to the problem of low turnout which is furthermore accessible to government bodies even if they have low levels of financial resources. However, the use of Big Data in this way is also a radically different approach to the problem of involving citizens in policymaking; and the editorial concludes by reflecting on the significance of this for the policymaking process….(More)”

Press Release: “Across society, from health to agriculture and transport, from energy to climate change and security, practitioners in every discipline recognise the potential of the enormous amounts of data being created every day. The challenge is to capture, manage and process that information to derive meaningful results and make a difference to people’s lives. The Big Data Europe project has just released the first public version of its open source platform designed to do just that. In 7 pilot studies, it is helping to solve societal challenges by putting cutting edge technology in the hands of experts in fields other than IT.

Although many crucial big data technologies are freely available as open source software, they are often difficult for non-experts to integrate and deploy. Big Data Europe solves that problem by providing a package that can readily be installed locally or at any scale in a cloud infrastructure by a systems administrator, and configured via a simple user interface. Tools like Apache Hadoop, Apache Spark, Apache Flink and many others can be instantiated easily….

The tools included in the platform were selected after a process of requirements-gathering across the seven societal challenges identified by the European Commission (Health, Food, Energy, Transport, Climate, Social Sciences and Security). Tasks like message passing are handled using Kafka and Flume, storage by Hive and Cassandra, or publishing through geotriples. The platform uses the Docker system to make it easy to add new tools and, again, for them to operate at a scale limited only by the computing infrastructure….

New book by Cass R. Sunstein: “In recent years, ‘Nudge Units’ or ‘Behavioral Insights Teams’ have been created in the United States, the United Kingdom, Germany, and other nations. All over the world, public officials are using the behavioral sciences to protect the environment, promote employment and economic growth, reduce poverty, and increase national security. In this book, Cass R. Sunstein, the eminent legal scholar and best-selling co-author of Nudge (2008), breaks new ground with a deep yet highly readable investigation into the ethical issues surrounding nudges, choice architecture, and mandates, addressing such issues as welfare, autonomy, self-government, dignity, manipulation, and the constraints and responsibilities of an ethical state. Complementing the ethical discussion, The Ethics of Influence: Government in the Age of Behavioral Science contains a wealth of new data on people’s attitudes towards a broad range of nudges, choice architecture, and mandates…(More)”

An exploratory analysis of public uses of New York City open data by Karen Okamoto in Webology: “In 2012, New York City Council passed legislation to make government data open and freely available to the public. By approving this legislation, City Council was attempting to make local government more transparent, accountable, and streamlined in its operations. It was also attempting to create economic opportunities and to encourage the public to identify ways in which to improve government and local communities. The purpose of this study is to explore public uses of New York City open data. Currently, more than 1300 datasets covering broad areas such as health, education, transportation, public safety, housing and business are available on the City’s Open Data Portal. This study found a plethora of maps, visualizations, tools, apps and analyses made by the public using New York City open data. Indeed, open data is inspiring a productive range of creative reuses yet questions remain concerning how useable the data is for users without technical skills and resources….(More)”

Lena Groeger at ProPublica: “A few weeks ago, Snapchat released a new photo filter. It appeared alongside many of the other such face-altering filters that have become a signature of the service. But instead of surrounding your face with flower petals or giving you the nose and ears of a Dalmatian, the filter added slanted eyes, puffed cheeks and large front teeth. A number of Snapchat users decried the filter as racist, saying it mimicked a “yellowface” caricature of Asians. The company countered that they meant to represent anime characters and deleted the filter within a few hours.

“Snapchat is the prime example of what happens when you don’t have enough people of color building a product,” wrote Bay Area software engineer Katie Zhu in an essay she wrote about deleting the app and leaving the service. In a tech world that hires mostly white men, the absence of diverse voices means that companies can be blind to design decisions that are hurtful to their customers or discriminatory.A Snapchat spokesperson told ProPublica that the company has recently hired someone to lead their diversity recruiting efforts.

A Snapchat spokesperson told ProPublica that the company has recently hired someone to lead their diversity recruiting efforts.

But this isn’t just Snapchat’s problem. Discriminatory design and decision-making affects all aspects of our lives: from the quality of our health care and education to where we live to what scientific questions we choose to ask. It would be impossible to cover them all, so we’ll focus on the more tangible and visual design that humans interact with every day.

You can’t talk about discriminatory design without mentioning city planner Robert Moses, whose public works projects shaped huge swaths of New York City from the 1930s through the 1960s. The physical design of the environment is a powerful tool when it’s used to exclude and isolate specific groups of people. And Moses’ design choices have had lasting discriminatory effects that are still felt in modern New York.

A notorious example: Moses designed a number of Long Island Parkway overpasses to be so low that buses could not drive under them. This effectively blocked Long Island from the poor and people of color who tend to rely more heavily on public transportation. And the low bridges continue to wreak havoc in other ways: 64 collisions were recorded in 2014 alone (here’s a bad one).

The design of bus systems, railways, and other forms of public transportation has a history riddled with racial tensions and prejudiced policies. In the 1990s the Los Angeles’ Bus Riders Union went to court over the racial inequity they saw in the city’s public transportation system. The Union alleged that L.A.’s Metropolitan Transportation Authority spent “a disproportionately high share of its resources on commuter rail services, whose primary users were wealthy non-minorities, and a disproportionately low share on bus services, whose main patrons were low income and minority residents.” The landmark case was settled through a court-ordered consent decree that placed strict limits on transit funding and forced the MTA to invest over $2 billion in the bus system.

Of course, the design of a neighborhood is more than just infrastructure. Zoning laws and regulations that determine how land is used or what schools children go to have long been used as a tool to segregate communities. All too often, the end result of zoning is that low-income, often predominantly black and Latino communities are isolated from most of the resources and advantages of wealthy white communities….(More)”