Will Artificial Intelligence Replace Us or Empower Us?

Curated on Posted on by Stefaan Verhulst

Article by Peter Coy: “…But A.I. could also be designed to empower people rather than replace them, as I wrote a year ago in a newsletter about the M.I.T. Shaping the Future of Work Initiative.

Which of those A.I. futures will be realized was a big topic at the San Francisco conference, which was the annual meeting of the American Economic Association, the American Finance Association and 65 smaller groups in the Allied Social Science Associations.

Erik Brynjolfsson of Stanford was one of the busiest economists at the conference, dashing from one panel to another to talk about his hopes for a human-centric A.I. and his warnings about what he has called the “Turing Trap.”

Alan Turing, the English mathematician and World War II code breaker, proposed in 1950 to evaluate the intelligence of computers by whether they could fool someone into thinking they were human. His “imitation game” led the field in an unfortunate direction, Brynjolfsson argues — toward creating machines that behaved as much like humans as possible, instead of like human helpers.

Henry Ford didn’t set out to build a car that could mimic a person’s walk, so why should A.I. experts try to build systems that mimic a person’s mental abilities? Brynjolfsson asked at one session I attended.

Other economists have made similar points: Daron Acemoglu of M.I.T. and Pascual Restrepo of Boston University use the term “so-so technologies” for systems that replace human beings without meaningfully increasing productivity, such as self-checkout kiosks in supermarkets.

People will need a lot more education and training to take full advantage of A.I.’s immense power, so that they aren’t just elbowed aside by it. “In fact, for each dollar spent on machine learning technology, companies may need to spend nine dollars on intangible human capital,” Brynjolfsson wrote in 2022, citing research by him and others…(More)”.

AI Is Bad News for the Global South

Curated on Posted on by Stefaan Verhulst

Article by Rachel Adams: “…AI’s adoption in developing regions is also limited by its design. AI designed in Silicon Valley on largely English-language data is not often fit for purpose outside of wealthy Western contexts. The productive use of AI requires stable internet access or smartphone technology; in sub-Saharan Africa, only 25 percent of people have reliable internet access, and it is estimated that African women are 32 percent less likely to use mobile internet than their male counterparts.

Generative AI technologies are also predominantly developed using the English language, meaning that the outputs they produce for non-Western users and contexts are oftentimes useless, inaccurate, and biased. Innovators in the global south have to put in at least twice the effort to make their AI applications work for local contexts, often by retraining models on localized datasets and through extensive trial and error practices.

Where AI is designed to generate profit and entertainment only for the already privileged, it will not be effective in addressing the conditions of poverty and in changing the lives of groups that are marginalized from the consumer markets of AI. Without a high level of saturation across major industries, and without the infrastructure in place to enable meaningful access to AI by all people, global south nations are unlikely to see major economic benefits from the technology.

As AI is adopted across industries, human labor is changing. For poorer countries, this is engendering a new race to the bottom where machines are cheaper than humans and the cheap labor that was once offshored to their lands is now being onshored back to wealthy nations. The people most impacted are those with lower education levels and fewer skills, whose jobs can be more easily automated. In short, much of the population in lower- and middle-income countries may be affected, severely impacting the lives of millions of people and threatening the capacity of poorer nations to prosper…(More)”.

The AI tool that can interpret any spreadsheet instantly

Curated on Posted on by Stefaan Verhulst

Article by Duncan C. McElfresh: “Say you run a hospital and you want to estimate which patients have the highest risk of deterioration so that your staff can prioritize their care1. You create a spreadsheet in which there is a row for each patient, and columns for relevant attributes, such as age or blood-oxygen level. The final column records whether the person deteriorated during their stay. You can then fit a mathematical model to these data to estimate an incoming patient’s deterioration risk. This is a classic example of tabular machine learning, a technique that uses tables of data to make inferences. This usually involves developing — and training — a bespoke model for each task. Writing in Nature, Hollmann et al.report a model that can perform tabular machine learning on any data set without being trained specifically to do so.

Tabular machine learning shares a rich history with statistics and data science. Its methods are foundational to modern artificial intelligence (AI) systems, including large language models (LLMs), and its influence cannot be overstated. Indeed, many online experiences are shaped by tabular machine-learning models, which recommend products, generate advertisements and moderate social-media content3. Essential industries such as healthcare and finance are also steadily, if cautiously, moving towards increasing their use of AI.

Despite the field’s maturity, Hollmann and colleagues’ advance could be revolutionary. The authors’ contribution is known as a foundation model, which is a general-purpose model that can be used in a range of settings. You might already have encountered foundation models, perhaps unknowingly, through AI tools, such as ChatGPT and Stable Diffusion. These models enable a single tool to offer varied capabilities, including text translation and image generation. So what does a foundation model for tabular machine learning look like?

Let’s return to the hospital example. With spreadsheet in hand, you choose a machine-learning model (such as a neural network) and train the model with your data, using an algorithm that adjusts the model’s parameters to optimize its predictive performance (Fig. 1a). Typically, you would train several such models before selecting one to use — a labour-intensive process that requires considerable time and expertise. And of course, this process must be repeated for each unique task.

Figure 1 | A foundation model for tabular machine learning. a, Conventional machine-learning models are trained on individual data sets using mathematical optimization algorithms. A different model needs to be developed and trained for each task, and for each data set. This practice takes years to learn and requires extensive time and computing resources. b, By contrast, a ‘foundation’ model could be used for any machine-learning task and is pre-trained on the types of data used to train conventional models. This type of model simply reads a data set and can immediately produce inferences about new data points. Hollmann et al. developed a foundation model for tabular machine learning, in which inferences are made on the basis of tables of data. Tabular machine learning is used for tasks as varied as social-media moderation and hospital decision-making, so the authors’ advance is expected to have a profound effect in many areas…(More)”

Comparative perspectives on the regulation of large language models

Curated on Posted on by Stefaan Verhulst

Editorial to Special Issue by Cristina Poncibò and Martin Ebers: “Large language models (LLMs) represent one of the most significant technological advancements in recent decades, offering transformative capabilities in natural language processing and content generation. Their development has far-reaching implications across technological, economic and societal domains, simultaneously creating opportunities for innovation and posing profound challenges for governance and regulation. As LLMs become integral to various sectors, from education to healthcare to entertainment, regulators are scrambling to establish frameworks that ensure their safe and ethical use.

Our issue primarily examines the private ordering, regulatory responses and normative frameworks for LLMs from a comparative law perspective, with a particular focus on the European Union (EU), the United States (US) and China. An introductory part preliminarily explores the technical principles that underpin LLMs as well as their epistemological foundations. It also addresses key sector-specific legal challenges posed by LLMs, including their implications for criminal law, data protection and copyright law…(More)”.

SciAgents: Automating Scientific Discovery Through Bioinspired Multi-Agent Intelligent Graph Reasoning

Curated on Posted on by Stefaan Verhulst

Paper by Alireza Ghafarollahi, and Markus J. Buehler: “A key challenge in artificial intelligence (AI) is the creation of systems capable of autonomously advancing scientific understanding by exploring novel domains, identifying complex patterns, and uncovering previously unseen connections in vast scientific data. In this work, SciAgents, an approach that leverages three core concepts is presented: (1) large-scale ontological knowledge graphs to organize and interconnect diverse scientific concepts, (2) a suite of large language models (LLMs) and data retrieval tools, and (3) multi-agent systems with in-situ learning capabilities. Applied to biologically inspired materials, SciAgents reveals hidden interdisciplinary relationships that were previously considered unrelated, achieving a scale, precision, and exploratory power that surpasses human research methods. The framework autonomously generates and refines research hypotheses, elucidating underlying mechanisms, design principles, and unexpected material properties. By integrating these capabilities in a modular fashion, the system yields material discoveries, critiques and improves existing hypotheses, retrieves up-to-date data about existing research, and highlights strengths and limitations. This is achieved by harnessing a “swarm of intelligence” similar to biological systems, providing new avenues for discovery. How this model accelerates the development of advanced materials by unlocking Nature’s design principles, resulting in a new biocomposite with enhanced mechanical properties and improved sustainability through energy-efficient production is shown…(More)”.

To Whom Does the World Belong?

Curated on Posted on by Stefaan Verhulst

Essay by Alexander Hartley: “For an idea of the scale of the prize, it’s worth remembering that 90 percent of recent U.S. economic growth, and 65 percent of the value of its largest 500 companies, is already accounted for by intellectual property. By any estimate, AI will vastly increase the speed and scale at which new intellectual products can be minted. The provision of AI services themselves is estimated to become a trillion-dollar market by 2032, but the value of the intellectual property created by those services—all the drug and technology patents; all the images, films, stories, virtual personalities—will eclipse that sum. It is possible that the products of AI may, within my lifetime, come to represent a substantial portion of all the world’s financial value.

In this light, the question of ownership takes on its true scale, revealing itself as a version of Bertolt Brecht’s famous query: To whom does the world belong?

Questions of AI authorship and ownership can be divided into two broad types. One concerns the vast troves of human-authored material fed into AI models as part of their “training” (the process by which their algorithms “learn” from data). The other concerns ownership of what AIs produce. Call these, respectively, the input and output problems.

So far, attention—and lawsuits—have clustered around the input problem. The basic business model for LLMs relies on the mass appropriation of human-written text, and there simply isn’t anywhere near enough in the public domain. OpenAI hasn’t been very forthcoming about its training data, but GPT-4 was reportedly trained on around thirteen trillion “tokens,” roughly the equivalent of ten trillion words. This text is drawn in large part from online repositories known as “crawls,” which scrape the internet for troves of text from news sites, forums, and other sources. Fully aware that vast data scraping is legally untested—to say the least—developers charged ahead anyway, resigning themselves to litigating the issue in retrospect. Lawyer Peter Schoppert has called the training of LLMs without permission the industry’s “original sin”—to be added, we might say, to the technology’s mind-boggling consumption of energy and water in an overheating planet. (In September, Bloomberg reported that plans for new gas-fired power plants have exploded as energy companies are “racing to meet a surge in demand from power-hungry AI data centers.”)…(More)”.

Collaborative Intelligence

Curated on Posted on by Stefaan Verhulst

Book edited by Mira Lane and Arathi Sethumadhavan: “…The book delves deeply into the dynamic interplay between theory and practice, shedding light on the transformative potential and complexities of AI. For practitioners deeply immersed in the world of AI, Lane and Sethumadhavan offer firsthand accounts and insights from technologists, academics, and thought leaders, as well as a series of compelling case studies, ranging from AI’s impact on artistry to its role in addressing societal challenges like modern slavery and wildlife conservation.

As the global AI market burgeons, this book enables collaboration, knowledge sharing, and interdisciplinary dialogue. It caters not only to the practitioners shaping the AI landscape but also to policymakers striving to navigate the intricate relationship between humans and machines, as well as academics. Divided into two parts, the first half of the book offers readers a comprehensive understanding of AI’s historical context, its influence on power dynamics, human-AI interaction, and the critical role of audits in governing AI systems. The second half unfolds a series of eight case studies, unraveling AI’s impact on fields as varied as healthcare, vehicular safety, conservation, human rights, and the metaverse. Each chapter in this book paints a vivid picture of AI’s triumphs and challenges, providing a panoramic view of how it is reshaping our world…(More)”

Beyond checking a box: how a social licence can help communities benefit from data reuse and AI

Curated on Posted on by Stefaan Verhulst

Article by Stefaan Verhulst and Peter Addo: “In theory, consent offers a mechanism to reduce power imbalances. In reality, existing consent mechanisms are limited and, in many respects, archaic, based on binary distinctions – typically presented in check-the-box forms that most websites use to ask you to register for marketing e-mails – that fail to appreciate the nuance and context-sensitive nature of data reuse. Consent today generally means individual consent, a notion that overlooks the broader needs of communities and groups.

While we understand the need to safeguard information about an individual such as, say, their health status, this information can help address or even prevent societal health crises. Individualised notions of consent fail to consider the potential public good of reusing individual data responsibly. This makes them particularly problematic in societies that have more collective orientations, where prioritising individual choices could disrupt the social fabric.

The notion of a social licence, which has its roots in the 1990s within the extractive industries, refers to the collective acceptance of an activity, such as data reuse, based on its perceived alignment with community values and interests. Social licences go beyond the priorities of individuals and help balance the risks of data misuse and missed use (for example, the risks of violating privacy vs. neglecting to use private data for public good). Social licences permit a broader notion of consent that is dynamic, multifaceted and context-sensitive.

Policymakers, citizens, health providers, think tanks, interest groups and private industry must accept the concept of a social licence before it can be established. The goal for all stakeholders is to establish widespread consensus on community norms and an acceptable balance of social risk and opportunity.

Community engagement can create a consensus-based foundation for preferences and expectations concerning data reuse. Engagement could take place via dedicated “data assemblies” or community deliberations about data reuse for particular purposes under particular conditions. The process would need to involve voices as representative as possible of the different parties involved, and include those that are traditionally marginalised or silenced…(More)”.

Harnessing AI: How to develop and integrate automated prediction systems for humanitarian anticipatory action

Curated on Posted on by Stefaan Verhulst

CEPR Report: “Despite unprecedented access to data, resources, and wealth, the world faces an escalating wave of humanitarian crises. Armed conflict, climate-induced disasters, and political instability are displacing millions and devastating communities. Nearly one in every five children are living in or fleeing conflict zones (OCHA, 2024). Often the impacts of conflict and climatic hazards – such as droughts and flood – exacerbate each other, leading to even greater suffering. As crises unfold and escalate, the need for timely and effective humanitarian action becomes paramount.

Sophisticated systems for forecasting and monitoring natural and man-made hazards have emerged as critical tools to help inform and prompt action. The full potential for the use of such automated forecasting systems to inform anticipatory action (AA) is immense but is still to be realised. By providing early warnings and predictive insights, these systems could help organisations allocate resources more efficiently, plan interventions more effectively, and ultimately save lives and prevent or reduce humanitarian impact.


This Policy Insight provides an account of the significant technical, ethical, and organisational difficulties involved in such systems, and the current solutions in place…(More)”.

Harvard Is Releasing a Massive Free AI Training Dataset Funded by OpenAI and Microsoft

Curated on Posted on by Stefaan Verhulst

Article by Kate Knibbs: “Harvard University announced Thursday it’s releasing a high-quality dataset of nearly 1 million public-domain books that could be used by anyone to train large language models and other AI tools. The dataset was created by Harvard’s newly formed Institutional Data Initiative with funding from both Microsoft and OpenAI. It contains books scanned as part of the Google Books project that are no longer protected by copyright.

Around five times the size of the notorious Books3 dataset that was used to train AI models like Meta’s Llama, the Institutional Data Initiative’s database spans genres, decades, and languages, with classics from Shakespeare, Charles Dickens, and Dante included alongside obscure Czech math textbooks and Welsh pocket dictionaries. Greg Leppert, executive director of the Institutional Data Initiative, says the project is an attempt to “level the playing field” by giving the general public, including small players in the AI industry and individual researchers, access to the sort of highly-refined and curated content repositories that normally only established tech giants have the resources to assemble. “It’s gone through rigorous review,” he says…(More)”.