The Artificial Intelligence Podcast

A recent study predicts that tech companies will exhaust the supply of publicly available training data for AI language models by 2026-2032, threatening the current pace of progress in AI development. The tens of trillions of words online, used to make AI systems smarter, will soon run out, forcing companies to rely on less-reliable synthetic data or tap into sensitive information. This could lead to degraded performance, bias, and unfairness in AI systems. Researchers are exploring ways to address the issue, which raises questions about the future of human-generated content and online accessibility.

Meta, the parent company of Facebook and Instagram, plans to use public posts and images from its platforms to train its artificial intelligence tools, sparking criticism from digital rights groups who argue it's an abuse of personal data. Under the new privacy policy, users' public posts, images, and comments will be used to "develop and improve" Meta's AI products, but private messages will not be included. Critics argue that processing years' worth of user content is an abuse of personal data, while Meta claims its approach complies with privacy laws and is consistent with other big tech firms.

A recent survey by the Center for Democracy & Technology reveals that teachers continue to struggle with the responsible and safe use of generative AI in their classrooms. Despite 80% of teachers receiving formal training on AI use policies, many lack clear guidance on addressing potential misuse, leading to heavy reliance on ineffective detection tools. This can result in increased disciplinary action and distrust towards students, particularly those with disabilities. The survey highlights the need for schools to provide detailed guidance to educators and for education leaders to address the broader implications of AI use in schools.

A popular US news app, NewsBreak, has been accused of using AI to create false news stories, including a bogus report of a shooting in New Jersey and incorrect information about a food bank in Colorado. Despite presenting itself as a legitimate news source, the app has been publishing erroneous stories since 2021, often rewriting press releases and local news with AI tools. NewsBreak has also faced accusations of copyright infringement and has settled cases with at least two publications. Concerns have been raised about the app's Chinese connections, but the CEO claims compliance with US data and privacy laws.

A recent study tested seven large language models, including those behind ChatGPT, on cognitive psychology tests to gauge their reasoning capacity. The results showed that despite their advanced capabilities, these AI systems struggled with consistent reasoning and made simple errors. They provided varying responses to the same question, made basic mistakes like addition errors, and mistook consonants for vowels. The study highlighted the importance of understanding how these models think before entrusting them with decision-making tasks, and raised questions about whether their limitations and biases could lead to potential pitfalls.

Artificial intelligence is being harnessed to potentially end animal testing in the medical field. AI systems are trawling through existing animal testing results, preventing the need for new tests. Experts believe AI can extract and synthesize data more efficiently than humans, and is already being used to determine a new chemical's toxicity. While AI is not perfect, it's proving to be more accurate than animal testing in some cases. Several projects are underway to replace animal testing, with many experts backing AI as a key player in phasing out the practice.

The Federal Trade Commission and Justice Department are preparing to launch antitrust investigations into Microsoft, Open AI, and Nvidia, focusing on their conduct in the artificial intelligence industry. The FTC will lead the investigation into Microsoft and Open AI, while the DOJ will focus on Nvidia. The probes come as startups gain ground in generative AI, prompting tech giants to invest heavily in integrating AI technology. The investigations will examine investments, partnerships, and potential anti-competitive practices in the rapidly growing AI industry.

Join Michael Huth, Dean at Imperial College London and Chief Research Officer at Xayn, as he discusses his journey from a mathematics background to leading roles in both academia and a startup, highlighting the importance of strategic control in his department at Imperial and the innovative work at Xayn. He explores topics such as the concept of sovereign legal AI, the global landscape of AI legislation, ethical concerns in regulated industries, and the integration of cybersecurity with AI research and development.

Researchers at the National Institutes of Health have developed an AI tool called LORIS that predicts how well individual cancer patients will respond to immunotherapy. The tool uses machine learning to analyze clinical, pathologic, and genomic features to provide a personalized score for each patient. LORIS was developed by analyzing large datasets of patients with 18 types of solid tumors and identifies six key factors that predict a patient's response to immunotherapy. The technology has the potential to revolutionize cancer treatment by providing a more accurate way to predict patient response and ensuring patients receive the most effective treatment for their individual needs.