DATA – Page 428 – The Living Library

Discrimination by algorithm: scientists devise test to detect AI bias

Curated on December 19, 2016August 3, 2018 by Stefaan Verhulst

Hannah Devlin at the Guardian: “There was the voice recognition software that struggled to understand women, the crime prediction algorithm that targeted black neighbourhoods and the online ad platform which was more likely to show men highly paid executive jobs.

Concerns have been growing about AI’s so-called “white guy problem” and now scientists have devised a way to test whether an algorithm is introducing gender or racial biases into decision-making.

Mortiz Hardt, a senior research scientist at Google and a co-author of the paper, said: “Decisions based on machine learning can be both incredibly useful and have a profound impact on our lives … Despite the need, a vetted methodology in machine learning for preventing this kind of discrimination based on sensitive attributes has been lacking.”

The paper was one of several on detecting discrimination by algorithms to be presented at the Neural Information Processing Systems (NIPS) conference in Barcelona this month, indicating a growing recognition of the problem.

Nathan Srebro, a computer scientist at the Toyota Technological Institute at Chicago and co-author, said: “We are trying to enforce that you will not have inappropriate bias in the statistical prediction.”

The test is aimed at machine learning programs, which learn to make predictions about the future by crunching through vast quantities of existing data. Since the decision-making criteria are essentially learnt by the computer, rather than being pre-programmed by humans, the exact logic behind decisions is often opaque, even to the scientists who wrote the software….“Our criteria does not look at the innards of the learning algorithm,” said Srebro. “It just looks at the predictions it makes.”

Their approach, called Equality of Opportunity in Supervised Learning, works on the basic principle that when an algorithm makes a decision about an individual – be it to show them an online ad or award them parole – the decision should not reveal anything about the individual’s race or gender beyond what might be gleaned from the data itself.

For instance, if men were on average twice as likely to default on bank loans than women, and if you knew that a particular individual in a dataset had defaulted on a loan, you could reasonably conclude they were more likely (but not certain) to be male.

However, if an algorithm calculated that the most profitable strategy for a lender was to reject all loan applications from men and accept all female applications, the decision would precisely confirm a person’s gender.

“This can be interpreted as inappropriate discrimination,” said Srebro….(More)”.

Science Can Restore America’s Faith in Democracy

Curated on December 18, 2016July 19, 2019 by Stefaan Verhulst

Ariel Procaccia in Wired: “…Like most other countries, individual states in the US employ the antiquated plurality voting system, in which each voter casts a vote for a single candidate, and the person who amasses the largest number of votes is declared the winner. If there is one thing that voting experts unanimously agree on, it is that plurality voting is a bad idea, or at least a badly outdated one….. Maine recently became the first US state to adopt instant-runoff voting; the approach will be used for choosing the governor and members of Congress and the state legislature….

So why aren’t we already using cutting-edge voting systems in national elections? Perhaps because changing election systems usually itself requires an election, where short-term political considerations may trump long-term, scientifically grounded reasoning….Despite these difficulties, in the last few years state-of-the-art voting systems have made the transition from theory to practice, through not-for-profit online platforms that focus on facilitating elections in cities and organizations, or even just on helping a group of friends decide where to go to dinner. For example, the Stanford Crowdsourced Democracy Team has created an online tool whereby residents of a city can vote on how to allocate the city’s budget for public projects such as parks and roads. This tool has been used by New York City, Boston, Chicago, and Seattle to allocate millions of dollars. Building on this success, the Stanford team is experimenting with groundbreaking methods, inspired by computational thinking, to elicit and aggregate the preferences of residents.

The Princeton-based project All Our Ideas asks voters to compare pairs of ideas, and then aggregates these comparisons via statistical methods, ultimately providing a ranking of all the ideas. To date, roughly 14 million votes have been cast using this system, and it has been employed by major cities and organizations. Among its more whimsical use cases is the Washington Post’s 2010 holiday gift guide, where the question was “what gift would you like to receive this holiday season”; the disappointingly uncreative top idea, based on tens of thousands of votes, was “money”.

Finally, the recently launched website RoboVote (which I created with collaborators at Carnegie Mellon and Harvard) offers AI-driven voting methods to help groups of people make smart collective decisions. Applications range from selecting a spot for a family vacation or a class president, to potentially high-stakes choices such as which product prototype to develop or which movie script to produce.

These examples show that centuries of research on voting can, at long last, make a societal impact in the internet age. They demonstrate what science can do for democracy, albeit on a relatively small scale, for now….(More)’

Privacy of Public Data

Curated on December 17, 2016August 3, 2018 by Stefaan Verhulst

Paper by Kirsten E. Martin and Helen Nissenbaum: “The construct of an information dichotomy has played a defining role in regulating privacy: information deemed private or sensitive typically earns high levels of protection, while lower levels of protection are accorded to information deemed public or non-sensitive. Challenging this dichotomy, the theory of contextual integrity associates privacy with complex typologies of information, each connected with respective social contexts. Moreover, it contends that information type is merely one among several variables that shape people’s privacy expectations and underpin privacy’s normative foundations. Other contextual variables include key actors – information subjects, senders, and recipients – as well as the principles under which information is transmitted, such as whether with subjects’ consent, as bought and sold, as required by law, and so forth. Prior work revealed the systematic impact of these other variables on privacy assessments, thereby debunking the defining effects of so-called private information.

In this paper, we shine a light on the opposite effect, challenging conventional assumptions about public information. The paper reports on a series of studies, which probe attitudes and expectations regarding information that has been deemed public. Public records established through the historical practice of federal, state, and local agencies, as a case in point, are afforded little privacy protection, or possibly none at all. Motivated by progressive digitization and creation of online portals through which these records have been made publicly accessible our work underscores the need for more concentrated and nuanced privacy assessments, even more urgent in the face of vigorous open data initiatives, which call on federal, state, and local agencies to provide access to government records in both human and machine readable forms. Within a stream of research suggesting possible guard rails for open data initiatives, our work, guided by the theory of contextual integrity, provides insight into the factors systematically shaping individuals’ expectations and normative judgments concerning appropriate uses of and terms of access to information.

Using a factorial vignette survey, we asked respondents to rate the appropriateness of a series of scenarios in which contextual elements were systematically varied; these elements included the data recipient (e.g. bank, employer, friend,.), the data subject, and the source, or sender, of the information (e.g. individual, government, data broker). Because the object of this study was to highlight the complexity of people’s privacy expectations regarding so-called public information, information types were drawn from data fields frequently held in public government records (e.g. voter registration, marital status, criminal standing, and real property ownership).

Our findings are noteworthy on both theoretical and practical grounds. In the first place, they reinforce key assertions of contextual integrity about the simultaneous relevance to privacy of other factors beyond information types. In the second place, they reveal discordance between truisms that have frequently shaped public policy relevant to privacy. …(More)”

How Artificial Intelligence Will Usher in the Next Stage of E-Government

Curated on December 17, 2016August 3, 2018 by Stefaan Verhulst

Daniel Castro at GovTech: “Since the earliest days of the Internet, most government agencies have eagerly explored how to use technology to better deliver services to citizens, businesses and other public-sector organizations. Early on, observers recognized that these efforts often varied widely in their implementation, and so researchers developed various frameworks to describe the different stages of growth and development of e-government. While each model is different, they all identify the same general progression from the informational, for example websites that make government facts available online, to the interactive, such as two-way communication between government officials and users, to the transactional, like applications that allow users to access government services completely online.

However, we will soon see a new stage of e-government: the perceptive.

The defining feature of the perceptive stage will be that the work involved in interacting with government will be significantly reduced and automated for all parties involved. This will come about principally from the integration of artificial intelligence (AI) — computer systems that can learn, reason and decide at levels similar to that of a human — into government services to make it more insightful and intelligent.

Consider the evolution of the Department of Motor Vehicles. The informational stage made it possible for users to find the hours for the local office; the interactive stage made it possible to ask the agency a question by email; and the transactional stage made it possible to renew a driver’s license online.

In the perceptive stage, the user will simply say, “Siri, I need a driver’s license,” and the individual’s virtual assistant will take over — collecting any additional information from the user, coordinating with the government’s system and scheduling any in-person meetings automatically. That’s right: AI might finally end your wait at the DMV.

In general, there are at least three ways that AI will impact government agencies. First, it will enable government workers to be more productive since the technology can be used to automate many tasks. …

Second, AI will create a faster, more responsive government. AI enables the creation of autonomous, intelligent agents — think online chatbots that answer citizens’ questions, real-time fraud detection systems that constantly monitor government expenditures and virtual legislative assistants that quickly synthesize feedback from citizens to lawmakers.

Third, AI will allow people to interact more naturally with digital government services…(More)”

Artificial Intelligence Could Help Colleges Better Plan What Courses They Should Offer

Curated on December 16, 2016May 29, 2019 by Stefaan Verhulst

Jeffrey R. Young at EdSsurge: Big data could help community colleges better predict how industries are changing so they can tailor their IT courses and other programs. After all, if Amazon can forecast what consumers will buy and prestock items in their warehouses to meet the expected demand, why can’t colleges do the same thing when planning their curricula, using predictive analytics to make sure new degree or certificates programs are started just in time for expanding job opportunities?

That’s the argument made by Gordon Freedman, president of the nonprofit National Laboratory for Education Transformation. He’s part of a new center that will do just that, by building a data warehouse that brings together up-to-date information on what skills employers need and what colleges currently offer—and then applying artificial intelligence to attempt to predict when sectors or certain employment needs might be expanding.

He calls the approach “opportunity engineering,” and the center boasts some heavy-hitting players to assist in the efforts, including the University of Chicago, the San Diego Supercomputing Center and Argonne National Laboratory. It’s called the National Center for Opportunity Engineering & Analysis.

Ian Roark, vice president of workforce development at Pima Community College in Arizona, is among those eager for this kind of “opportunity engineering” to emerge.

He explains when colleges want to start new programs, they face a long haul—it takes time to develop a new curriculum, put it through an internal review, and then send it through an accreditor….

Other players are already trying to translate the job market into a giant data set to spot trends. LinkedIn sits on one of the biggest troves of data, with hundreds of millions of job profiles, and ambitions to create what it calls the “economic graph” of the economy. But not everyone is on LinkedIn, which attracts mainly those in white-collar jobs. And companies such as Burning Glass Technologies have scanned hundreds of thousands of job listings and attempt to provide real-time intelligence on what employers say they’re looking for. Those still don’t paint the full picture, Freedman argues, such as what jobs are forming at companies.

“We need better information from the employer, better information from the job seeker and better information from the college, and that’s what we’re going after,” Freedman says…(More)”.

Can you crowdsource water quality data?

Curated on December 16, 2016May 29, 2019 by Stefaan Verhulst

Pratibha Mistry at The Water Blog (Worldbank): “The recently released Contextual Framework for Crowdsourcing Water Quality Data lays out a strategy for citizen engagement in decentralized water quality monitoring, enabled by the “mobile revolution.”

According to the WHO, 1.8 billion people lack access to safe drinking water worldwide. Poor source water quality, non-existent or insufficient treatment, and defects in water distribution systems and storage mean these consumers use water that often doesn’t meet the WHO’s Guidelines for Drinking Water Quality.

The crowdsourcing framework develops a strategy to engage citizens in measuring and learning about the quality of their own drinking water. Through their participation, citizens provide utilities and water supply agencies with cost-effective water quality data in near-real time. Following a typical crowdsourcing model: consumers use their mobile phones to report water quality information to a central service. That service receives the information, then repackages and shares it via mobile phone messages, websites, dashboards, and social media. Individual citizens can thus be educated about their water quality, and water management agencies and other stakeholders can use the data to improve water management; it’s a win-win.

*A well-implemented crowdsourcing project both depends on and benefits end users.*Source: Figure modified from Hutchings, M., Dev, A., Palaniappan, M., Srinivasan, V., Ramanathan, N., Taylor, J. 2012. “mWASH: Mobile Phone Applications for the Water, Sanitation, and Hygiene Sector.” Pacific Institute, Oakland, California. 114 p. (*Link to full text*)

Several groups, from the private sector to academia to non-profits, have taken a recent interest in developing a variety of so-called mWASH apps (mobile phone applications for the water, sanitation, and hygiene WASH sector). A recent academic study analyzed how mobile phones might facilitate the flow of water quality data between water suppliers and public health agencies in Africa. USAID has invested in piloting a mobile application in Tanzania to help consumers test their water for E. coli….(More)”

Introducing the Agricultural Open Data Package: BETA Version

Curated on December 14, 2016August 3, 2018 by Stefaan Verhulst

PressRelease: “GODAN, Open Data for Development (OD4D) Network, Open Data Charter, and the Open Data Institute are pleased to announce the release of the Agricultural Open Data Package: BETA version. …The Agriculture Open Data Package (http://AgPack.info) has been designed to help governments get to impact with open data in the agriculture sector. This practical resource provides key policy areas, key data categories, examples datasets, relevant interoperability initiatives, and use cases that policymakers and other stakeholders in the agriculture sector or open data should focus on, in order to address food security challenges.

The Package is meant as a source of inspiration and an invitation to start a national open data for agriculture initiative.

In the Package we identify fourteen key categories of data and discuss the effort it will take for a government to make this data available in a meaningful way. …

The Package also highlights more than ten use cases (the number is growing) demonstrating how open data is being harnessed to address sustainable agriculture and food security around the world. Examples include:

mapping water points to optimise scarce resource allocation in Burkina Faso
surfacing daily price information on multiple food commodities across India
benchmarking agricultural productivity in the Netherlands

Where relevant we also highlight applicable interoperability initiatives, such as open contracting, international aid transparency initiative (IATI), and global product classification (GPC) standards.

We recognise that the agriculture sector is diverse, with many contextual differences affecting scope of activities, priorities and capacities. In the full version of the Agricultural Open Data Package we discuss important implementation considerations such as inter-agency coordination and resourcing to develop an appropriate data infrastructure and a healthy data ‘ecosystem’ for agriculture….(More)”

Four steps to precision public health

Curated on December 13, 2016August 3, 2018 by Stefaan Verhulst

Scott F. Dowell, David Blazes & Susan Desmond-Hellmann at Nature: “When domestic transmission of Zika virus was confirmed in the United States in July 2016, the entire country was not declared at risk — nor even the entire state of Florida. Instead, precise surveillance defined two at-risk areas of Miami-Dade County, neighbourhoods measuring just 2.6 and 3.9 square kilometres. Travel advisories and mosquito control focused on those regions. Six weeks later, ongoing surveillance convinced officials to lift restrictions in one area and expand the other.

By contrast, a campaign against yellow fever launched this year in sub-Saharan Africa defines risk at the level of entire nations, often hundreds of thousands of square kilometres. More granular assessments have been deemed too complex.

The use of data to guide interventions that benefit populations more efficiently is a strategy we call precision public health. It requires robust primary surveillance data, rapid application of sophisticated analytics to track the geographical distribution of disease, and the capacity to act on such information¹.

The availability and use of precise data is becoming the norm in wealthy countries. But large swathes of the developing world are not reaping its advantages. In Guinea, it took months to assemble enough data to clearly identify the start of the largest Ebola outbreak in history. This should take days. Sub-Saharan Africa has the highest rates of childhood mortality in the world; it is also where we know the least about causes of death…..

The value of precise disease tracking was baked into epidemiology from the start. In 1854, John Snow famously located cholera cases in London. His mapping of the spread of infection through contaminated water dealt a blow to the idea that the disease was caused by bad air. These days, people and pathogens move across the globe swiftly and in great numbers. In 2009, the H1N1 ‘swine flu’ influenza virus took just 35 days to spread from Mexico and the United States to China, South Korea and 12 other countries…

The public-health community is sharing more data faster; expectations are higher than ever that data will be available from clinical trials and from disease surveillance. In the past two years, the US National Institutes of Health, the Wellcome Trust in London and the Gates Foundation have all instituted open data policies for their grant recipients, and leading journals have declared that sharing data during disease emergencies will not impede later publication.

Meanwhile, improved analysis, data visualization and machine learning have expanded our ability to use disparate data sources to decide what to do. A study published last year⁴ used precise geospatial modelling to infer that insecticide-treated bed nets were the single most influential intervention in the rapid decline of malaria.

However, in many parts of the developing world, there are still hurdles to the collection, analysis and use of more precise public-health data. Work towards malaria elimination in South Africa, for example, has depended largely on paper reporting forms, which are collected and entered manually each week by dozens of subdistricts, and eventually analysed at the province level. This process would be much faster if field workers filed reports from mobile phones.

Sources: Ref. 8/Bill & Melinda Gates Foundation

…Frontline workers should not find themselves frustrated by global programmes that fail to take into account data on local circumstances. Wherever they live — in a village, city or country, in the global south or north — people have the right to public-health decisions that are based on the best data and science possible, that minimize risk and cost, and maximize health in their communities…(More)”

neveragain.tech

Curated on December 13, 2016August 3, 2018 by Stefaan Verhulst

neveragain.tech: “We, the undersigned, are employees of tech organizations and companies based in the United States. We are engineers, designers, business executives, and others whose jobs include managing or processing data about people. We are choosing to stand in solidarity with Muslim Americans, immigrants, and all people whose lives and livelihoods are threatened by the incoming administration’s proposed data collection policies. We refuse to build a database of people based on their Constitutionally-protected religious beliefs. We refuse to facilitate mass deportations of people the government believes to be undesirable…..

Today we stand together to say: not on our watch, and never again.

We commit to the following actions:

We refuse to participate in the creation of databases of identifying information for the United States government to target individuals based on race, religion, or national origin.
We will advocate within our organizations:
- to minimize the collection and retention of data that would facilitate ethnic or religious targeting.
- to scale back existing datasets with unnecessary racial, ethnic, and national origin data.
- to responsibly destroy high-risk datasets and backups.
- to implement security and privacy best practices, in particular, for end-to-end encryption to be the default wherever possible.
- to demand appropriate legal process should the government request that we turn over user data collected by our organization, even in small amounts.
If we discover misuse of data that we consider illegal or unethical in our organizations:
- We will work with our colleagues and leaders to correct it.
- If we cannot stop these practices, we will exercise our rights and responsibilities to speak out publicly and engage in responsible whistleblowing without endangering users.
- If we have the authority to do so, we will use all available legal defenses to stop these practices.
- If we do not have such authority, and our organizations force us to engage in such misuse, we will resign from our positions rather than comply.
We will raise awareness and ask critical questions about the responsible and fair use of data and algorithms beyond our organization and our industry….(More)“