The Model Is The Message

Curated on Posted on by Stefaan Verhulst

Essay by Benjamin Bratton and Blaise Agüera y Arcas: “An odd controversy appeared in the news cycle last month when a Google engineer, Blake Lemoine, was placed on leave after publicly releasing transcripts of conversations with LaMDA, a chatbot based on a Large Language Model (LLM) that he claims is conscious, sentient and a person.

Like most other observers, we do not conclude that LaMDA is conscious in the ways that Lemoine believes it to be. His inference is clearly based in motivated anthropomorphic projection. At the same time, it is also possible that these kinds of artificial intelligence (AI) are “intelligent” — and even “conscious” in some way — depending on how those terms are defined.

Still, neither of these terms can be very useful if they are defined in strongly anthropocentric ways. An AI may also be one and not the other, and it may be useful to distinguish sentience from both intelligence and consciousness. For example, an AI may be genuinely intelligent in some way but only sentient in the restrictive sense of sensing and acting deliberately on external information. Perhaps the real lesson for philosophy of AI is that reality has outpaced the available language to parse what is already at hand. A more precise vocabulary is essential.

AI and the philosophy of AI have deeply intertwined histories, each bending the other in uneven ways. Just like core AI research, the philosophy of AI goes through phases. Sometimes it is content to apply philosophy (“what would Kant say about driverless cars?”) and sometimes it is energized to invent new concepts and terms to make sense of technologies before, during and after their emergence. Today, we need more of the latter.

We need more specific and creative language that can cut the knots around terms like “sentience,” “ethics,” “intelligence,” and even “artificial,” in order to name and measure what is already here and orient what is to come. Without this, confusion ensues — for example, the cultural split between those eager to speculate on the sentience of rocks and rivers yet dismiss AI as corporate PR vs. those who think their chatbots are persons because all possible intelligence is humanlike in form and appearance. This is a poor substitute for viable, creative foresight. The curious case of synthetic language  — language intelligently produced or interpreted by machines — is exemplary of what is wrong with present approaches, but also demonstrative of what alternatives are possible…(More)”.

Artificial Intelligence in the City: Building Civic Engagement and Public Trust

Curated on Posted on by Stefaan Verhulst

Collection of essays edited by Ana Brandusescu, Ana, and Jess Reia: “After navigating various challenging policy and regulatory contexts over the years, in different regions, we joined efforts to create a space that offers possibilities for engagement focused on the expertise, experiences and hopes to shape the future of technology in urban areas. The AI in the City project emerged as an opportunity to connect people, organizations, and resources in the networks we built over the last decade of work on research and advocacy in tech policy. Sharing non-Western and Western perspectives from five continents, the contributors questioned, challenged, and envisioned ways public trust and meaningful civic engagement can flourish and persist as data and AI become increasingly pervasive in our lives. This collection of essays brings together a group of multidisciplinary scholars, activists, and practitioners working on a diverse range of initiatives to map strategies going forward. Divided into five parts, the collection brings into focus: 1) Meaningful engagement and public participation; 2) Addressing inequalities and building trust; 3) Public and private boundaries in tech policy; 4) Legal perspectives and mechanisms for accountability; and 5) New directions for local and urban governance. The focus on civil society and academia was deliberate: a way to listen to and learn with people who have dedicated many years to public interest advocacy, governance and policy that represents the interests of their communities…(More)”.

Your Boss Is an Algorithm: Artificial Intelligence, Platform Work and Labour

Curated on Posted on by Stefaan Verhulst

Book by Antonio Aloisi and Valerio De Stefano: “What effect do robots, algorithms, and online platforms have on the world of work? Using case studies and examples from across the EU, the UK, and the US, this book provides a compass to navigate this technological transformation as well as the regulatory options available, and proposes a new map for the era of radical digital advancements.

From platform work to the gig-economy and the impact of artificial intelligence, algorithmic management, and digital surveillance on workplaces, technology has overwhelming consequences for everyone’s lives, reshaping the labour market and straining social institutions. Contrary to preliminary analyses forecasting the threat of human work obsolescence, the book demonstrates that digital tools are more likely to replace managerial roles and intensify organisational processes in workplaces, rather than opening the way for mass job displacement.

Can flexibility and protection be reconciled so that legal frameworks uphold innovation? How can we address the pervasive power of AI-enabled monitoring? How likely is it that the gig-economy model will emerge as a new organisational paradigm across sectors? And what can social partners and political players do to adopt effective regulation?

Technology is never neutral. It can and must be governed, to ensure that progress favours the many. Digital transformation can be an essential ally, from the warehouse to the office, but it must be tested in terms of social and political sustainability, not only through the lenses of economic convenience. Your Boss Is an Algorithm offers a guide to explore these new scenarios, their promises, and perils…(More)”

Human-centred mechanism design with Democratic AI

Curated on Posted on by Stefaan Verhulst

Paper by Raphael Koster et al: “Building artificial intelligence (AI) that aligns with human values is an unsolved problem. Here we developed a human-in-the-loop research pipeline called Democratic AI, in which reinforcement learning is used to design a social mechanism that humans prefer by majority. A large group of humans played an online investment game that involved deciding whether to keep a monetary endowment or to share it with others for collective benefit. Shared revenue was returned to players under two different redistribution mechanisms, one designed by the AI and the other by humans. The AI discovered a mechanism that redressed initial wealth imbalance, sanctioned free riders and successfully won the majority vote. By optimizing for human preferences, Democratic AI offers a proof of concept for value-aligned policy innovation…(More)”.

Crime Prediction Keeps Society Stuck in the Past

Curated on Posted on by Stefaan Verhulst

Article by Chris Gilliard: “…All of these policing systems operate on the assumption that the past determines the future. In Discriminating Data: Correlation, Neighborhoods, and the New Politics of Recognition, digital media scholar Wendy Hui Kyong Chun argues that the most common methods used by technologies such as PredPol and Chicago’s heat list to make predictions do nothing of the sort. Rather than anticipating what might happen out of the myriad and unknowable possibilities on which the very idea of a future depends, machine learning and other AI-based methods of statistical correlation “restrict the future to the past.” In other words, these systems prevent the future in order to “predict” it—they ensure that the future will be just the same as the past was.

“If the captured and curated past is racist and sexist,” Chun writes, “these algorithms and models will only be verified as correct if they make sexist and racist predictions.” This is partly a description of the familiar garbage-in/garbage-out problem with all data analytics, but it’s something more: Ironically, the putatively “unbiased” technology sold to us by promoters is said to “work” precisely when it tells us that what is contingent in history is in fact inevitable and immutable. Rather than helping us to manage social problems like racism as we move forward, as the McDaniel case shows in microcosm, these systems demand that society not change, that things that we should try to fix instead must stay exactly as they are.

It’s a rather glaring observation that predictive policing tools are rarely if ever (with the possible exception of the parody “White Collar Crime Risk Zone” project) focused on wage theft or various white collar crimes, even though the dollar amounts of those types of offenses far outstrip property crimes in terms of dollar value by several orders of magnitude. This gap exists because of how crime exists in the popular imagination. For instance, news reports in recent weeks bludgeoned readers with reports of a so-called “crime wave” of shoplifting at high-end stores. Yet just this past February, Amazon agreed to pay regulators a whopping $61.7 million, the amount the FTC says the company shorted drivers in a two-and-a-half-year period. That story received a fraction of the coverage, and aside from the fine, there will be no additional charges.

The algorithmic crystal ball that promises to predict and forestall future crimes works from a fixed notion of what a criminal is, where crimes occur, and how they are prosecuted (if at all). Those parameters depend entirely on the power structure empowered to formulate them—and very often the explicit goal of those structures is to maintain existing racial and wealth hierarchies. This is the same set of carceral logics that allow the placement of children into gang databases, or the development of a computational tool to forecast which children will become criminals. The process of predicting the lives of children is about cementing existing realities rather than changing them. Entering children into a carceral ranking system is in itself an act of violence, but as in the case of McDaniel, it also nearly guarantees that the system that sees them as potential criminals will continue to enact violence on them throughout their lifetimes…(More)”.

How Does the Public Sector Identify Problems It Tries to Solve with AI?

Curated on Posted on by Stefaan Verhulst

Article by Maia Levy Daniel: “A correct analysis of the implementation of AI in a particular field or process needs to start by identifying if there actually is a problem to be solved. For instance, in the case of job matching, the problem would be related to the levels of unemployment in the country, and presumably addressing imbalances in specific fields. Then, would AI be the best way to address this specific problem? Are there any alternatives? Is there any evidence that shows that AI would be a better tool? Building AI systems is expensive and the funds being used by the public sector come from taxpayers. Are there any alternatives that could be less expensive? 

Moreover, governments must understand from the outset that these systems could involve potential risks for civil and human rights. Thus, it should be justified in detail why the government might be choosing a more expensive or riskier option. A potential guide to follow is the one developed by the UK’s Office for Artificial Intelligence on how to use AI in the public sector. This guide includes a section specifically devoted to how to assess whether AI is the right solution to a problem.

AI is such a buzzword that it has become appealing for governments to use as a solution to any public problem, without even starting to look for available alternatives. Although automation could accelerate decision-making processes, speed should not be prioritized over quality or over human rights protection. As Daniel Susser argues in his recent paper, the speed at which automated decisions are reached has normative implications. By incorporating digital technologies in decision-making processes, temporal norms and values that govern them are impacted, disrupting prior norms, re-calibrating balanced trade-offs, or displacing automation’s costs. As Susser suggests, speed is not necessarily bad; however, “using computational tools to speed up (or slow down) certain decisions is not a ‘neutral’ adjustment without further explanations.” 

So, conducting a thorough diagnosis including the identification of the specific problem to address and the best way to address it is key to protecting citizens’ rights. And this is why transparency must be mandatory. As citizens, we have a right to know how these processes are being conceived and designed, the reasons governments choose to implement technologies, as well as the risks involved.

In addition, maybe a good way to ultimately approach the systemic problem and change the structure of incentives is to stop using the pretentious terms “artificial intelligence”, “AI”, and “machine learning”, as Emily Tucker, the Executive Director of the Center on Privacy & Technology at Georgetown Law Center announced the Center would do. As Tucker explained, these terms are confusing for the average person, and the way they are typically employed makes us think it’s a machine rather than human beings making the decisions. By removing marketing terms from the equation and giving more visibility to the humans involved, these technologies may not ultimately seem so exotic…(More)”.

What AI Can Tell Us About Intelligence

Curated on Posted on by Stefaan Verhulst

Essay by Yann LeCun and Jacob Browning: “If there is one constant in the field of artificial intelligence it is exaggeration: there is always breathless hype and scornful naysaying. It is helpful to occasionally take stock of where we stand.

The dominant technique in contemporary AI is deep learning (DL) neural networks, massive self-learning algorithms which excel at discerning and utilizing patterns in data. Since their inception, critics have prematurely argued that neural networks had run into an insurmountable wall — and every time, it proved a temporary hurdle. In the 1960s, they could not solve non-linear functions. That changed in the 1980s with backpropagation, but the new wall was how difficult it was to train the systems. The 1990s saw a rise of simplifying programs and standardized architectures which made training more reliable, but the new problem was the lack of training data and computing power.

In 2012, when contemporary graphics cards could be trained on the massive ImageNet dataset, DL went mainstream, handily besting all competitors. But then critics spied a new problem: DL required too much hand-labelled data for training. The last few years have rendered this criticism moot, as self-supervised learning has resulted in incredibly impressive systems, such as GPT-3, which do not require labeled data.

Today’s seemingly insurmountable wall is symbolic reasoning, the capacity to manipulate symbols in the ways familiar from algebra or logic. As we learned as children, solving math problems involves a step-by-step manipulation of symbols according to strict rules (e.g., multiply the furthest right column, carry the extra value to the column to the left, etc.). Gary Marcus, author of “The Algebraic Mind”and co-author (with Ernie Davis) of “Rebooting AI,”recently argued that DL is incapable of further progress because neural networks struggle with this kind of symbol manipulation. By contrast, many DL researchers are convinced that DL is already engaging in symbolic reasoning and will continue to improve at it.

At the heart of this debate are two different visions of the role of symbols in intelligence, both biological and mechanical: one holds that symbolic reasoning must be hard-coded from the outset and the other holds it can be learned through experience, by machines and humans alike. As such, the stakes are not just about the most practical way forward, but also how we should understand human intelligence — and, thus, how we should pursue human-level artificial intelligence…(More)”.

Non-human humanitarianism: when ‘AI for good’ can be harmful

Curated on Posted on by Stefaan Verhulst

Paper by Mirca Madianou: “Artificial intelligence (AI) applications have been introduced in humanitarian operations in order to help with the significant challenges the sector is facing. This article focuses on chatbots which have been proposed as an efficient method to improve communication with, and accountability to affected communities. Chatbots, together with other humanitarian AI applications such as biometrics, satellite imaging, predictive modelling and data visualisations, are often understood as part of the wider phenomenon of ‘AI for social good’. The article develops a decolonial critique of humanitarianism and critical algorithm studies which focuses on the power asymmetries underpinning both humanitarianism and AI. The article asks whether chatbots, as exemplars of ‘AI for good’, reproduce inequalities in the global context. Drawing on a mixed methods study that includes interviews with seven groups of stakeholders, the analysis observes that humanitarian chatbots do not fulfil claims such as ‘intelligence’. Yet AI applications still have powerful consequences. Apart from the risks associated with misinformation and data safeguarding, chatbots reduce communication to its barest instrumental forms which creates disconnects between affected communities and aid agencies. This disconnect is compounded by the extraction of value from data and experimentation with untested technologies. By reflecting the values of their designers and by asserting Eurocentric values in their programmed interactions, chatbots reproduce the coloniality of power. The article concludes that ‘AI for good’ is an ‘enchantment of technology’ that reworks the colonial legacies of humanitarianism whilst also occluding the power dynamics at play…(More)”.

Why You Need an AI Ethics Committee

Curated on Posted on by Stefaan Verhulst

Article by Reid Blackman: “…There are a lot of well-documented and highly publicized ethical risks associated with AI; unintended bias and invasions of privacy are just two of the most notable kinds. In many instances the risks are specific to particular uses, like the possibility that self-driving cars will run over pedestrians or that AI-generated social media newsfeeds will sow distrust of public institutions. In some cases they’re major reputational, regulatory, financial, and legal threats. Because AI is built to operate at scale, when a problem occurs, it affects all the people the technology engages with—for instance, everyone who responds to a job listing or applies for a mortgage at a bank. If companies don’t carefully address ethical issues in planning and executing AI projects, they can waste a lot of time and money developing software that is ultimately too risky to use or sell, as many have already learned.

Your organization’s AI strategy needs to take into account several questions: How might the AI we design, procure, and deploy pose ethical risks that cannot be avoided? How do we systematically and comprehensively identify and mitigate them? If we ignore them, how much time and labor would it take us to respond to a regulatory investigation? How large a fine might we pay if found guilty, let alone negligent, of violating regulations or laws? How much would we need to spend to rebuild consumer and public trust, provided that money could solve the problem?

The answers to those questions will underscore how much your organization needs an AI ethical risk program. It must start at the executive level and permeate your company’s ranks—and, ultimately, the technology itself. In this article I’ll focus on one crucial element of such a program—an AI ethical risk committee—and explain why it’s critical that it include ethicists, lawyers, technologists, business strategists, and bias scouts. Then I’ll explore what that committee requires to be effective at a large enterprise.

But first, to provide a sense of why such a committee is so important, I’ll take a deep dive into the issue of discriminatory AI. Keep in mind that this is just one of the risks AI presents; there are many others that also need to be investigated in a systematic way…(More)”.

AI Can Predict Potential Nutrient Deficiencies from Space

Curated on Posted on by Stefaan Verhulst

Article by Rachel Berkowitz: “Micronutrient deficiencies afflict more than two billion people worldwide, including 340 million children. This lack of vitamins and minerals can have serious health consequences. But diagnosing deficiencies early enough for effective treatment requires expensive, time-consuming blood draws and laboratory tests.

New research provides a more efficient approach. Computer scientist Elizabeth Bondi and her colleagues at Harvard University used publicly available satellite data and artificial intelligence to reliably pinpoint geographical areas where populations are at high risk of micronutrient deficiencies. This analysis could potentially pave the way for early public health interventions.

Existing AI systems can use satellite data to predict localized food security issues, but they typically rely on directly observable features. For example, agricultural productivity can be estimated from views of vegetation. Micronutrient availability is harder to calculate. After seeing research showing that areas near forests tend to have better dietary diversity, Bondi and her colleagues were inspired to identify lesser-known markers for potential malnourishment. Their work shows that combining data such as vegetation cover, weather and water presence can suggest where populations will lack iron, vitamin B12 or vitamin A.

The team examined raw satellite measurements and consulted with local public health officials, then used AI to sift through the data and pinpoint key features. For instance, a food market, inferred based on roads and buildings visible, was vital for predicting a community’s risk level. The researchers then linked these features to specific nutrients lacking in four regions’ populations across Madagascar. They used real-world biomarker data (blood samples tested in labs) to train and test their AI program….(More)”.