Archive for year: 2019

David Bounie, Professor of Economics, Head of Economics and Social Sciences at Télécom ParisTech

Patrick Waelbroeck, Professor of Industrial Economy and Econometrics at Télécom ParisTech and co-founder of the Chair Values and Policies of Personal Information

[divider style=”dotted” top=”20″ bottom=”20″]

The original version of this article was published on the website of the Chair Values and Policies of Personal Information. This Chair brings together researchers from Télécom ParisTech, Télécom SudParis and Institut Mines Télécom Business School, and is supported by the Mines-Télécom Foundation.

[divider style=”dotted” top=”20″ bottom=”20″]

[dropcap]A[/dropcap]lgorithms rule our lives. They increasingly intervene in our daily activities – i.e. career paths, adverts, recommendations, scoring, online searches, flight prices – as improvements are made in data science and statistical learning.

Despite being initially considered as neutral, they are now blamed for biasing results and discriminating against people, voluntarily or not, according to their gender, ethnicity or sexual orientation. In the United States, studies have shown that African American people were more penalised in court decisions (Angwin et al., 2016). They are also discriminated against more often on online flat rental platforms (Edelman, Luca and Svirsky, 2017). Finally, online targeted and automated ads promoting job opportunities in the Science, Technology, Engineering and Mathematics (STEM) fields seem to be more frequently shown to men than to women (Lambrecht and Tucker, 2017).

Algorithmic bias raises significant issues in terms of ethics and fairness. Why are algorithms biased? Is bias unpreventable? If so, how can it be limited?

Three sources of bias have been identified, in relation to cognitive, statistical and economic aspects. First, algorithm results vary according to the way programmers, i.e. humans, coded them, and studies in behavioural economics have shown there are cognitive biases in decision-making.

• For instance, a bandwagon bias may lead a programmer to follow popular models without checking whether these are accurate.
• Anticipation and confirmation biases may lead a programmer to favour their own beliefs, even though available data challenges such beliefs.
• Illusory correlation may lead someone to perceive a relationship between two independent variables.
• A framing bias occurs when a person draws different conclusions from a same dataset based on the way the information is presented.

Second, bias can be statistical. The phrase ‘Garbage in, garbage out’ refers to the fact that even the most sophisticated machine will produce incorrect and potentially biased results if the input data provided is inaccurate. After all, it is pretty easy to believe in a score produced by a complex proprietary algorithm and seemingly based on multiple sources. Yet, if the data set based on which the algorithm is trained to learn to categorise or predict is partial or inaccurate, as is often the case with fake news, trolls or fake identities, results are likely to be biased. What happens if the data is incorrect? Or if the algorithm is trained using data from US citizens, who may behave much differently from European citizens? Or even, if certain essential variables are omitted? For instance, how might machines encode relational skills and emotional intelligence (which are hard to get for machines as they do not feel emotions), leadership skills or teamwork in an algorithm? Omitted variables may lead an algorithm to produce a biased result for the simple reason the omitted variables may be correlated with the variables used in the model. Finally, what happens when the training data comes from truncated samples or is not representative of the population that you wish to make predictions for (sample-selection bias)? In his Nobel Memorial Prize-winning research, James Heckman showed that selection bias was related to omitted-variable bias. Credit scoring is a striking example. In order to determine which risk category a borrower belongs to, algorithms rely on data related to people who were eligible for a loan in a particular institution – they ignore files of people who were denied credit, did not need a loan or got one in another institution.

Third, algorithms may bias results for economic reasons. Think of online automated advisors specialised in selling financial services. They can favour the products of the company giving the advice, at the expense of the consumer if these financial products are more expensive than the market average. Such situation is called price discrimination. Besides, in the context of multi-sided platforms, algorithms may favour third parties who have signed agreements with the platform. In the context of e-commerce, the European Commission recently fined Google 2.4bn euros for promoting its own products at the top of search results on Google Shopping, to the detriment of competitors. Other disputes have occurred in relation to the simple delisting of apps in search results on the Apple Store or to apps being downgraded in marketplaces’ search results.

Algorithms thus come with bias, which seems unpreventable. The question now is: how can bias be identified and discrimination be limited? Algorithms and artificial intelligence will indeed only be socially accepted if all actors are capable of meeting the ethical challenges raised by the use of data and following best practice.

Researchers first need to design fairer algorithms. Yet what is fairness, and which fairness rules should be applied? There is no easy answer to these questions, as debates have opposed researchers in social science and those in philosophy for centuries. Fairness is a normative concept, many definitions of which are incompatible. For instance, compare individual fairness and group fairness. One simple criterion of individual fairness is that of equal opportunity, the principle according to which individuals with identical capacities should be treated similarly. However, this criterion is incompatible with group fairness, according to which individuals of the same group, such as women, should be treated similarly. In other words, equal opportunity for all individuals cannot exist if a fairness criterion is applied on gender. These two notions of fairness are incompatible.

A second challenge faces companies, policy makers and regulators, whose duty it is to promote ethical practices – transparency and responsibility – through an efficient regulation of the collection and use of personal data. Many issues arise. Should algorithms be transparent and therefore audited? Who should be responsible for the harm caused by discrimination? Is the General Data Protection Regulation fit for algorithmic bias? How could ethical constraints be included? Admittedly they could increase costs for society at the microeconomic level, yet they could help lower the costs of unfairness and inequality stemming from an automated society that wouldn’t comply with the fundamental principles of unbiasedness and lack of systematic discrimination.

Read on I’MTech Ethics, an overlooked aspect on algorithms?

Online services are becoming more and more personalised. This transformation designed to satisfy the end-user might be seen as an economic opportunity, but also as a risk, since personalised services usually require personal data to be efficient. Two perceptions that do not seem compatible. Maryline Laurent and Nesrine Kaâniche, researchers at Telecom SudParis and members of the Chair Values and Policies of Personal Information, tackle this tough issue in this article. They give an overview of how technology can solve this equation by allowing both personalization and privacy. 

[divider style=”normal” top=”20″ bottom=”20″]

This article has initially been published on the Chair Values and Policies of Personal Information website.

[divider style=”normal” top=”20″ bottom=”20″]

[dropcap]P[/dropcap]ersonalised services have become a major stake in the IT sector as they require actors to improve both the quality of the collected data and their ability to use them. Many services are running the innovation race, namely those related to companies’ information systems, government systems, e-commerce, access to knowledge, health, energy management, leisure and entertaining. The point is to offer end-users the best possible quality of experience, which in practice implies qualifying the relevance of the provided information and continuously adapting services to consumers’ uses and preferences.

Personalised services offer many perks, among which targeted recommendations based on interests, events, news, special offers for local services or goods, movies, books, and so on. Search engines return results that are usually personalised based on a user’s profile and actually start personalising as soon as a keyword is entered, by identifying semantics. For instance, the noun ‘mouse’ may refer to a small rodent if you’re a vet, a stay mouse if you’re a sailor, or a device that helps move the cursor on a computer screen if you’re an Internet user. In particular, mobile phone applications use personalisation; health and wellness apps (e.g. the new FitBit and Vivosport trackers) can come in very handy as they offer tips to improve one’s lifestyle, help users receive medical care remotely, or warn them on any possible health issue they detect as being related to a known illness.

How is personalisation technologically carried out?

When surfing on the Internet and using mobile phone services or apps, users are required to authenticate. Authentication allows to connect their digital identity with the personal data that is saved and collected from exchanges. Some software packages also include trackers, such as cookies, which are exchanged between a browser and a service provider or even a third party and allow to track individuals. Once an activity is linked to a given individual, a provider can easily fill up their profile with personal data, e.g. preferences and interests, and run efficient algorithms, often based on artificial intelligence (AI), to provide them with a piece of information, a service or targeted content. Sometimes, although more rarely, personalisation may rely solely on a situation experienced by a user – the simple fact they are geolocated in a certain place can trigger an ad or targeted content to be sent to them.

What risks may arise from enhanced personalisation?

Enhanced personalisation causes risks for users in particular. Based on geolocation data only, a third party may determine that a user goes to a specialised medical centre to treat cancer, or that they often spend time at a legal advice centre, a place of worship or a political party’s local headquarters. If such personal data is sold on a marketplace[1] and thus made accessible to insurers, credit institutions, employers and lessors, their use may breach user privacy and freedom of movement. And this is just one kind of data. If these were to be cross-referenced with a user’s pictures, Internet clicks, credit card purchases and heart rate… What further behavioural conclusions could be drawn? How could those be used?

One example that comes to mind is price discrimination,[2] i.e. charging different prices for the same product or service to different customers according to their location or social group. Democracies can also suffer from personalisation, as the Cambridge Analytica scandal has shown. In April 2018, Facebook confessed that U.S. citizens’ votes had been influenced through targeted political messaging in the 2016 election.

Responsible vs. resigned consumers

As pointed out in a survey carried out by the Chair Values and Policies of Personal Information (CVPIP) with French audience measurement company Médiamétrie,[3] some users and consumers have adopted data protection strategies, in particular by using software that prevents tracking or enables anonymous online browsing… Yet this requires them to make certain efforts. According to their purpose, they either choose a personalised service or a generic one to gain a certain control over their informational profile.

What if technology could solve the complex equation opposing personalised services and privacy?

Based on this observation, the Chair’s research team carried out a scientific study on Privacy Enhancing Technologies (PETs). In this study, we list the technologies that are best able to meet needs in terms of personalised services, give technical details about them and analyse them comparatively. As a result, we suggest classifying these solutions into 8 families, which are themselves grouped into the following 3 categories:

• User-oriented solutions. Users manage the protection of their identity by themselves by downloading software that allows them to control outgoing personal data.Protection solutions include attribute disclosure minimisation and noise addition, privacy-preserving certification,[4] and secure multiparty calculations (i.e. distributed among several independent collaborators).
• Server-oriented solutions. Any server we use is strongly involved in personal data processing by nature. Several protection approaches focus on servers, as these can anonymise databases in order to share or sell data, run heavy calculations on encrypted data upon customer request, implement solutions for automatic data self-destruction after a certain amount of time, or Privacy Information Retrieval solutions for non-intrusive content search tools that confidentially return relevant content to customers.
• Channel-oriented solutions. What matters here is the quality of the communication channel that connects users with servers, be it intermediated and/or encrypted, and the quality of the exchanged data, which may be damaged on purpose. There are two approaches to such solutions: securing communications and using trusted third parties as intermediators in a communication.

Some PETs are strictly in line with the ‘data protection by design’ concept as they implement data disclosure minimisation or data anonymisation, as required by Article 25-1 of the General Data Protection Regulation (GDPR).[5] Data and privacy protection methods should be implemented at the earliest possible stages of conceiving and developing IT solutions.

Our state of the art shows that using PETs raises many issues. Through a cross-cutting analysis linking CVPIP specialists’ different fields of expertise, we were able to identify several of these challenges:

• Using AI to better include privacy in personalised services;
• Improving the performance of existing solutions by adapting them to the limited capacities of mobile phone personalised services;
• Looking for the best economic trade-off between privacy, use of personal data and user experience;
• Determining how much it would cost industrials to include PETs in their solutions in terms of development, business model and adjusting their Privacy Impact Assessment (PIA);
• PETs seen as a way of bypassing or enforcing legislation.

Clicking, liking, sharing: all of our digital activities produce data. This information, which is collected and monetized by big digital information platforms, is on its way to becoming the virtual black gold of the 21^st century. Have we all become digital workers? Digital labor specialist and Télécom ParisTech researcher Antonio Casilli has recently published a work entitled En attendant les robots, enquête sur le travail du clic (Waiting for Robots, an Inquiry into Click Work). He sat down with us to shed some light on this exploitation 2.0.

Who we are, what we like, what we do, when and with whom: our virtual personal assistants and other digital contacts know everything about us. The digital space has become the new sphere of our private lives. This virtual social capital is the raw material for tech giants. The profitability of digital platforms like Facebook, Airbnb, Apple and Uber relies on the massive analysis of users’ data for advertising purposes. In his work entitled En attendant les robots, enquête sur le travail du clic (Waiting for Robots, an Inquiry into Click Work), Antonio Casilli explores the emergence of surveillance capitalism, an opaque and invisible form of capitalism marking the advent of a new form of digital proletariat: digital labor – or working with our digits. From the click worker who performs microtasks, who is aware of and paid for his activity, to the user who produces data implicitly, the sociologist analyzes the hidden face of this work carried out outside the world of work, and the all too-tangible reality of this intangible economy.

Read on I’MTech What is digtal labor?

Antonio Casilli focuses particularly on net platforms’ ability to put their users to work, convinced that they are consumers more than producers. “Free access to certain digital services is merely an illusion. Each click fuels a vast advertising market and produces data which is mined to develop artificial intelligence. Every “like”, post, photo, comment and connection fulfils one condition: producing value. This digital labor is either very poorly paid or entirely unpaid, since no one receives compensation that measures up to the value produced. But it is work nevertheless: a source of value that is traced, measured, assessed and contractually-regulated by the platforms’ terms and conditions for use,” explains the sociologist.

The hidden, human face of machine learning

For Antonio Casilli, digital labor is a new form of work which remains invisible, but is produced from our digital traces. Far from marking the disappearance of human labor with robots replacing the work they once did, this click work challenges the boundaries between work that is produced implicitly and formally recognizable employment. And for good reason: microworkers paid by the task or user-producers like ourselves are indispensable to these platforms. This data serves as the basis for machine learning models: behind the automation of a given task, such as visual or text recognition, humans are actually fueling applications by indicating clouds on images of the sky, for example, or by typing out words.

“As conventional wisdom would have it, these machines learn by themselves. But to train their algorithms to calibrate, or to improve their services, platforms need a huge number of people to train and test them,” says Antonio Casilli. One of the best-known examples is Mechanical Turk, a service offered by the American giant Amazon. Ironically, its name is a reference to a hoax that dates back to the 18^th century. An automaton chess player, called the “Mechanical Turk” was able to win games against human opponents. But the Turk was actually operated by a real human hiding inside.

I never tire of looking at videos of Chinese click farms. It's just so surreal to see hundreds of phones playing the same video for the purposes of fake engagment. pic.twitter.com/bHAGLqRqVb

— Matthew Brennan (@mbrennanchina) December 10, 2018

Likewise, certain so-called “smart” services rely heavily on unskilled workers: a sort of “artificial” artificial intelligence. In this work designed to benefit machines, digital workers are poorly paid to carry out micro-tasks. “Digital labor marks the appearance of a new way of working which can be called “taskified,” since human activity is reduced to a simple click; and “datafied” because it’s a matter of producing data so that digital platforms can obtain value from it,” explains Antonio Casilli. And this is how data can do harm. Alienation and exploitation: in addition to the internet task workers in northern countries, more commonly their counterparts in India, the Philippines and other developing countries with low average earnings, are sometimes paid less than one cent per click.

Legally regulating digital labor?

For now, these new forms of work are exempt from salary standards. Nevertheless, in recent years there has been an increasing number of class action suits against tech platforms to claim certain rights. Following the example of Uber drivers and Deliveroo delivery people, individuals have taken legal action in an attempt to have their commercial contracts reclassified as employment contacts. Antonio Casilli sees three possible ways to help combat job insecurity for digital workers and bring about social, economic and political recognition of digital labor.

“From Uber to platform moderators, traditional labor law—meaning reclassifying workers as salaried employees—could lead to the recognition of their status. But dependent employment may not be a “one-size-fits-all” solution. There are also a growing number of cooperative platforms being developed, where the users become owners of the means of production and algorithms.” Still, for Antonio Casilli, there are limitations to these advances. He sees a third possible solution. “When it comes to our data, we are not small-scale owners or small-scale entrepreneurs. We are small-scale data workers. And this personal data, which is neither private nor public, belongs to everyone and no one. Our privacy must be a collective bargaining tool. Institutions must still be invented and developed to make it into a real common asset. The internet is a new battleground,” says the researcher.

Toward taxation of the digital economy

Would this make our personal data less personal? “We all produce data. But this data is, in effect, a collective resource, which is appropriated and privatized by platforms. Instead of paying individuals for their data on a piecemeal basis, these platforms should return, give back, the value extracted from this data, to national or international authorities, through fair taxation,“ explains Antonio Casilli. In May of 2018, the General Data Protection Regulation (GDPR) came into effect in the European Union. Among other things, this text protects data as a personality attribute instead of as property. Therefore, in theory, everyone can now freely consent—at any moment—to the use of their personal data and withdraw this consent just as easily.

While in its current form, regulation involves a set of protective measures, setting up a tax system like the one put forward by Antonio Casilli would make it possible to establish an unconditional basic income. The very act of clicking or sharing information could give individuals a right to these royalties and allow each user to be paid for any content posted online. This income would not therefore be linked to the tasks carried out but would recognize the value created through these contributions. In 2020, over 20 billion devices will be connected to the Internet of Things. According to some estimates, the data market could reach nearly €430 billion per year by then, which is equivalent to a third of France’s GDP. Data is clearly a commodity unlike any other.

[divider style=”dotted” top=”20″ bottom=”20″]

En attendant les robots, enquête sur le travail du clic (Waiting for Robots, an Inquiry into Click Work)
Antonio A. Casilli
Éditions du Seuil, 2019
400 pages
24 € (paperback) – 16,99 € (e-book)

Original article in French written by Anne-Sophie Boutaud, for I’MTech.