Ground Truths Are Human Constructions

Curated on Posted on by Stefaan Verhulst

Article by Florian Jaton: “Artificial intelligence algorithms are human-made, cultural constructs, something I saw first-hand as a scholar and technician embedded with AI teams for 30 months. Among the many concrete practices and materials these algorithms need in order to come into existence are sets of numerical values that enable machine learning. These referential repositories are often called “ground truths,” and when computer scientists construct or use these datasets to design new algorithms and attest to their efficiency, the process is called “ground-truthing.”

Understanding how ground-truthing works can reveal inherent limitations of algorithms—how they enable the spread of false information, pass biased judgments, or otherwise erode society’s agency—and this could also catalyze more thoughtful regulation. As long as ground-truthing remains clouded and abstract, society will struggle to prevent algorithms from causing harm and to optimize algorithms for the greater good.

Ground-truth datasets define AI algorithms’ fundamental goal of reliably predicting and generating a specific output—say, an image with requested specifications that resembles other input, such as web-crawled images. In other words, ground-truth datasets are deliberately constructed. As such, they, along with their resultant algorithms, are limited and arbitrary and bear the sociocultural fingerprints of the teams that made them…(More)”.

Generative AI for economic research: Use cases and implications for economists  

Curated on Posted on by Stefaan Verhulst

Paper by Anton Korinek: “…This article describes use cases of modern generative AI to interested economic researchers based on the author’s exploration of the space. The main emphasis is on LLMs, which are the type of generative AI that is currently most useful for research. I have categorized their use cases into six areas: ideation and feedback, writing, background research, data analysis, coding, and mathematical derivations. I provide general instructions for how to take advantage of each of these capabilities and demonstrate them using specific examples. Moreover, I classify the capabilities of the most commonly used LLMs from experimental to highly useful to provide an overview. My hope is that this paper will be a useful guide both for researchers starting to use generative AI and for expert users who are interested in new use cases beyond what they already have experience with to take advantage of the rapidly growing capabilities of LLMs. The online resources associated with this paper are available at the journal website and will provide semi-annual updates on the capabilities and use cases of the most advanced generative AI tools for economic research. In addition, they offer a guide on “How do I start?” as well as a page with “Useful Resources on Generative AI for Economists.”…(More)”

The Branding Dilemma of AI: Steering Towards Efficient Regulation

Curated on Posted on by Stefaan Verhulst

Blog by Zeynep Engin: “…Undoubtedly, the term ‘Artificial Intelligence’ has captured the public imagination, proving to be an excellent choice from a marketing standpoint (particularly serving the marketing goals of big AI tech companies). However, this has not been without its drawbacks. The field has experienced several ‘AI winters’ when lofty promises failed to translate into real-world outcomes. More critically, this term has anthropomorphized what are, at their core, high-dimensional statistical optimization processes. Such representation has obscured their true nature and the extent of their potential. Moreover, as computing capacities have expanded exponentially, the ability of these systems to process large datasets quickly and precisely, identifying patterns autonomously, has often been misinterpreted as evidence of human-like or even superhuman intelligence. Consequently, AI systems have been elevated to almost mystical status, perceived as incomprehensible to humans and, thus, uncontrollable by humans…

A profound shift in the discourse surrounding AI is urgently necessary. The quest to replicate or surpass human intelligence, while technologically fascinating, does not fully encapsulate the field’s true essence and progress. Indeed, AI has seen significant advances, uncovering a vast array of functionalities. However, its core strength still lies in computational speed and precision — a mechanical prowess. The ‘magic’ of AI truly unfolds when this computational capacity intersects with the wealth of real-world data generated by human activities and the environment, transforming human directives into computational actions. Essentially, we are now outsourcing complex processing tasks to machines, moving beyond crafting bespoke solutions for each problem in favour of leveraging vast computational resources we have. This transition does not yield an ‘artificial intelligence’, but poses a new challenge to human intelligence in the knowledge creation cycle: the responsibility to formulate the ‘right’ questions and vigilantly monitor the outcomes of such intricate processing, ensuring the mitigation of any potential adverse impacts…(More)”.

The Data Revolution and the Study of Social Inequality: Promise and Perils

Curated on Posted on by Stefaan Verhulst

Paper by Mario L. Small: “The social sciences are in the midst of a revolution in access to data, as governments and private companies have accumulated vast digital records of rapidly multiplying aspects of our lives and made those records available to researchers. The accessibility and comprehensiveness of the data are unprecedented. How will the data revolution affect the study of social inequality? I argue that the speed, breadth, and low cost with which large-scale data can be acquired promise a dramatic transformation in the questions we can answer, but this promise can be undercut by size-induced blindness, the tendency to ignore important limitations amidst a source with billions of data points. The likely consequences for what we know about the social world remain unclear…(More)”.

In shaping AI policy, stories about social impacts are just as important as expert information

Curated on Posted on by Stefaan Verhulst

Blog by Daniel S. Schiff and Kaylyn Jackson Schiff: “Will artificial intelligence (AI) save the world or destroy it? Will it lead to the end of manual labor and an era of leisure and luxury, or to more surveillance and job insecurity? Is it the start of a revolution in innovation that will transform the economy for the better? Or does it represent a novel threat to human rights?

Irrespective of what turns out to be the truth, what our key policymakers believe about these questions matters. It will shape how they think about the underlying problems that AI policy is aiming to address, and which solutions are appropriate to do so. …In late 2021, we ran a study to better understand the impact of policy narratives on the behavior of policymakers. We focused on US state legislators,…

In our analysis, we found something surprising. We measured whether legislators were more likely to engage with a message featuring a narrative or featuring expert information, which we assessed by seeing if they clicked on a given fact sheet/story or clicked to register for or attended the webinar.

Despite the importance attached to technical expertise in AI circles, we found that narratives were at least as persuasive as expert information. Receiving a narrative emphasizing, say, growing competition between the US and China, or the faulty arrest of Robert Williams due to facial recognition, led to a 30 percent increase in legislator engagement compared to legislators who only received basic information about the civil society organization. These narratives were just as effective as more neutral, fact-based information about AI with accompanying fact sheets…(More)”

The New Digital Dark Age

Curated on Posted on by Stefaan Verhulst

Article by Gina Neff: “For researchers, social media has always represented greater access to data, more democratic involvement in knowledge production, and great transparency about social behavior. Getting a sense of what was happening—especially during political crises, major media events, or natural disasters—was as easy as looking around a platform like Twitter or Facebook. In 2024, however, that will no longer be possible.

In 2024, we will face a grim digital dark age, as social media platforms transition away from the logic of Web 2.0 and toward one dictated by AI-generated content. Companies have rushed to incorporate large language models (LLMs) into online services, complete with hallucinations (inaccurate, unjustified responses) and mistakes, which have further fractured our trust in online information.

Another aspect of this new digital dark age comes from not being able to see what others are doing. Twitter once pulsed with publicly readable sentiment of its users. Social researchers loved Twitter data, relying on it because it provided a ready, reasonable approximation of how a significant slice of internet users behaved. However, Elon Musk has now priced researchers out of Twitter data after recently announcing that it was ending free access to the platform’s API. This made it difficult, if not impossible, to obtain data needed for research on topics such as public health, natural disaster response, political campaigning, and economic activity. It was a harsh reminder that the modern internet has never been free or democratic, but instead walled and controlled.

Closer cooperation with platform companies is not the answer. X, for instance, has filed a suit against independent researchers who pointed out the rise in hate speech on the platform. Recently, it has also been revealed that researchers who used Facebook and Instagram’s data to study the platforms’ role in the US 2020 elections had been granted “independence by permission” by Meta. This means that the company chooses which projects to share its data with and, while the research may be independent, Meta also controls what types of questions are asked and who asks them…(More)”.

Fairness and Machine Learning

Curated on Posted on by Stefaan Verhulst

Book by Solon Barocas, Moritz Hardt and Arvind Narayanan: “…introduces advanced undergraduate and graduate students to the intellectual foundations of this recently emergent field, drawing on a diverse range of disciplinary perspectives to identify the opportunities and hazards of automated decision-making. It surveys the risks in many applications of machine learning and provides a review of an emerging set of proposed solutions, showing how even well-intentioned applications may give rise to objectionable results. It covers the statistical and causal measures used to evaluate the fairness of machine learning models as well as the procedural and substantive aspects of decision-making that are core to debates about fairness, including a review of legal and philosophical perspectives on discrimination. This incisive textbook prepares students of machine learning to do quantitative work on fairness while reflecting critically on its foundations and its practical utility.

• Introduces the technical and normative foundations of fairness in automated decision-making
• Covers the formal and computational methods for characterizing and addressing problems
• Provides a critical assessment of their intellectual foundations and practical utility
• Features rich pedagogy and extensive instructor resources…(More)”

Power to the standards

Curated on Posted on by Stefaan Verhulst

Report by Gergana Baeva, Michael Puntschuh and Matthieu Binder: “Standards and norms will be of central importance when it comes to the practical implementation of legal requirements for developed and deployed AI systems.

Using expert interviews, our study “Power to the standards” documents the existing obstacles on the way to the standardization of AI. In addition to practical and technological challenges, questions of democratic policy arise. After all, requirements such as fairness or transparency are often regarded as criteria to be determined by the legislator, meaning that they are only partially susceptible to standardization.

Our study concludes that the targeted and comprehensive participation of civil society actors is particularly necessary in order to compensate for existing participation deficits within the standardization process…(More)”.

Toward a Solid Acceptance of the Decentralized Web of Personal Data: Societal and Technological Convergence

Curated on Posted on by Stefaan Verhulst

Article by Ana Pop Stefanija et al: “Citizens using common online services such as social media, health tracking, or online shopping effectively hand over control of their personal data to the service providers—often large corporations. The services using and processing personal data are also holding the data. This situation is problematic, as has been recognized for some time: competition and innovation are stifled; data is duplicated; and citizens are in a weak position to enforce legal rights such as access, rectification, or erasure. The approach to address this problem has been to ascertain that citizens can access and update, with every possible service provider, the personal data that providers hold of or about them—the foundational view taken in the European General Data Protection Regulation (GDPR).

Recently, however, various societal, technological, and regulatory efforts are taking a very different approach, turning things around. The central tenet of this complementary view is that citizens should regain control of their personal data. Once in control, citizens can decide which providers they want to share data with, and if so, exactly which part of their data. Moreover, they can revisit these decisions anytime…(More)”.

Privacy and the City: How Data Shapes City Identities

Curated on Posted on by Stefaan Verhulst

Article by Bilyana Petkova: “This article bridges comparative constitutional law to research inspired by city leadership and the opportunities that technology brings to the urban environment. It looks first to some of the causes of rapid urbanization and finds them in the pitfalls of antidiscrimination law in federations and quasi-federations such as the United States and the European Union. Short of achieving antidiscrimination based on nationality, the EU has experimented with data privacy as an identity clause that could bring social cohesion the same way purportedly freedom of speech has done in the US. In the City however, diversity replaces antidiscrimination, making cities attractive to migrants across various walks of life. The consequence for federalism is the obvious decline of top-down or vertical, state-based federalism and the rise of legal urbanism whereby cities establish loose networks of cooperation between themselves. These types of arrangements are not yet a threat to the State or the EU but might become such if cities are increasingly isolated from the political process (e.g. at the EU level) and lack legal means to assert themselves in court. City diversity and openness to different cultures in turn invites a connection to new technologies since unlike antidiscrimination that is usually strictly examined on a case-by-case level, diversity can be more readily computed. Finally, the article focuses on NYC and London initiatives to suggest a futuristic vision of city networks that instead of using social credit score like in China, deploy data trusts to populate their urban environments, shape city identities and exchange ideas for urban development…(More)”.