Nearly all Americans use AI, though most dislike it, poll shows

Curated on Posted on by Stefaan Verhulst

Axios: “The vast majority of Americans use products that involve AI, but their views of the technology remain overwhelmingly negative, according to a Gallup-Telescope survey published Wednesday.

Why it matters: The rapid advancement of generative AI threatens to have far-reaching consequences for Americans’ everyday lives, including reshaping the job market, impacting elections, and affecting the health care industry.

The big picture: An estimated 99% of Americans used at least one AI-enabled product in the past week, but nearly two-thirds didn’t realize they were doing so, according to the poll’s findings.

  • These products included navigation apps, personal virtual assistants, weather forecasting apps, streaming services, shopping websites and social media platforms.
  • Ellyn Maese, a senior research consultant at Gallup, told Axios that the disconnect is because there is “a lot of confusion when it comes to what is just a computer program versus what is truly AI and intelligent.”

Zoom in: Despite its prevalent use, Americans’ views of AI remain overwhelmingly bleak, the survey found.

  • 72% of those surveyed had a “somewhat” or “very” negative opinion of how AI would impact the spread of false information, while 64% said the same about how it affects social connections.
  • The only area where a majority of Americans (61%) had a positive view of AI’s impact was regarding how it might help medical diagnosis and treatment…

State of play: The survey found that 68% of Americans believe the government and businesses equally bear responsibility for addressing the spread of false information related to AI.

  • 63% said the same about personal data privacy violations.
  • Majorities of those surveyed felt the same about combatting the unauthorized use of individuals’ likenesses (62%) and AI’s impact on job losses (52%).
  • In fact, the only area where Americans felt differently was when it came to national security threats; 62% of those surveyed said the government bore primary responsibility for reducing such threats…(More).”

Governing artificial intelligence means governing data: (re)setting the agenda for data justice

Curated on Posted on by Stefaan Verhulst

Paper by Linnet Taylor, Siddharth Peter de Souza, Aaron Martin, and Joan López Solano: “The field of data justice has been evolving to take into account the role of data in powering the field of artificial intelligence (AI). In this paper we review the main conceptual bases for governing data and AI: the market-based approach, the personal–non-personal data distinction and strategic sovereignty. We then analyse how these are being operationalised into practical models for governance, including public data trusts, data cooperatives, personal data sovereignty, data collaboratives, data commons approaches and indigenous data sovereignty. We interrogate these models’ potential for just governance based on four benchmarks which we propose as a reformulation of the Data Justice governance agenda identified by Taylor in her 2017 framework. Re-situating data justice at the intersection of data and AI, these benchmarks focus on preserving and strengthening public infrastructures and public goods; inclusiveness; contestability and accountability; and global responsibility. We demonstrate how they can be used to test whether a governance approach will succeed in redistributing power, engaging with public concerns and creating a plural politics of AI…(More)”.

Artificial Intelligence Narratives

Curated on Posted on by Stefaan Verhulst

A Global Voices Report: “…Framing AI systems as intelligent is further complicated and intertwined with neighboring narratives. In the US, AI narratives often revolve around opposing themes such as hope and fear, often bridging two strong emotions: existential fears and economic aspirations. In either case, they propose that the technology is powerful. These narratives contribute to the hype surrounding AI tools and their potential impact on society. Some examples include:

Many of these framings often present AI as an unstoppable and accelerating force. While this narrative can generate excitement and investment in AI research, it can also contribute to a sense of technological determinism and a lack of critical engagement with the consequences of widespread AI adoption. Counter-narratives are many and expand on the motifs of surveillance, erosions of trust, bias, job impacts, exploitation of labor, high-risk uses, the concentration of power, and environmental impacts, among others.

These narrative frames, combined with the metaphorical language and imagery used to describe AI, contribute to the confusion and lack of public knowledge about the technology. By positioning AI as a transformative, inevitable, and necessary tool for national success, these narratives can shape public opinion and policy decisions, often in ways that prioritize rapid adoption and commercialization…(More)”

AI for Social Good

Curated on Posted on by Stefaan Verhulst

Essay by Iqbal Dhaliwal: “Artificial intelligence (AI) has the potential to transform our lives. Like the internet, it’s a general-purpose technology that spans sectors, is widely accessible, has a low marginal cost of adding users, and is constantly improving. Tech companies are rapidly deploying more capable AI models that are seeping into our personal lives and work.

AI is also swiftly penetrating the social sector. Governments, social enterprises, and NGOs are infusing AI into programs, while public treasuries and donors are working hard to understand where to invest. For example, AI is being deployed to improve health diagnostics, map flood-prone areas for better relief targeting, grade students’ essays to free up teachers’ time for student interaction, assist governments in detecting tax fraud, and enable agricultural extension workers to customize advice.

But the social sector is also rife with examples over the past two decades of technologies touted as silver bullets that fell short of expectations, including One Laptop Per Child, SMS reminders to take medication, and smokeless stoves to reduce indoor air pollution. To avoid a similar fate, AI-infused programs must incorporate insights from years of evidence generated by rigorous impact evaluations and be scaled in an informed way through concurrent evaluations.

Specifically, implementers of such programs must pay attention to three elements. First, they must use research insights on where AI is likely to have the greatest social impact. Decades of research using randomized controlled trials and other exacting empirical work provide us with insights across sectors on where and how AI can play the most effective role in social programs.

Second, they must incorporate research lessons on how to effectively infuse AI into existing social programs. We have decades of research on when and why technologies succeed or fail in the social sector that can help guide AI adopters (governments, social enterprises, NGOs), tech companies, and donors to avoid pitfalls and design effective programs that work in the field.

Third, we must promote the rigorous evaluation of AI in the social sector so that we disseminate trustworthy information about what works and what does not. We must motivate adopters, tech companies, and donors to conduct independent, rigorous, concurrent impact evaluations of promising AI applications across social sectors (including impact on workers themselves); draw insights emerging across multiple studies; and disseminate those insights widely so that the benefits of AI can be maximized and its harms understood and minimized. Taking these steps can also help build trust in AI among social sector players and program participants more broadly…(More)”.

Which Health Facilities Have Been Impacted by L.A.-Area Fires? AI May Paint a Clearer Picture

Curated on Posted on by Stefaan Verhulst

Article by Andrew Schroeder: “One of the most important factors for humanitarian responders in these types of large-scale disaster situations is to understand the effects on the formal health system, upon which most people — and vulnerable communities in particular — rely upon in their neighborhoods. Evaluation of the impact of disasters on individual structures, including critical infrastructure such as health facilities, is traditionally a relatively slow and manually arduous process, involving extensive ground truth visitation by teams of assessment professionals.

Speeding up this process without losing accuracy, while potentially improving the safety and efficiency of assessment teams, is among the more important analytical efforts Direct Relief can undertake for response and recovery efforts. Manual assessments can now be effectively paired with AI-based analysis of satellite imagery to do just that…

With the advent of geospatial AI models trained on disaster damage impacts, ground assessment is not the only tool available to response agencies and others seeking to understand how much damage has occurred and the degree to which that damage may affect essential services for communities. The work of the Oregon State University team of experts in remote sensing-based post-disaster damage detection, led by Jamon Van Den Hoek and Corey Scher, was featured in the Financial Times on January 9.

Their modeling, based on Sentinel-1 satellite imagery, identified 21,757 structures overall, of which 11,124 were determined to have some level of damage. The Oregon State model does not distinguish between different levels of damage, and therefore cannot respond to certain types of questions that the manual inspections can respond to, but nevertheless the coverage area and the speed of detection have been much greater…(More)”.

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)”.