The Artificial Intelligence Podcast

Artificial intelligence (AI) is increasingly posing a threat to elections worldwide by enabling the creation of convincing fake content aimed at deceiving voters. Previously, creating phony photos, videos, or audio clips required technical skills and resources, but with the advent of free or low-cost generative AI services, anyone with a smartphone and a text prompt can create high-quality "deepfakes". These deepfakes have already been used in Europe and Asia, manipulating candidates' images and eroding trust in the electoral process. However, efforts are being made to tackle this issue, such as the European Union mandating special labeling of deepfakes and tech companies pledging to prevent AI disruption in elections.

The European Union (EU) has become the first major global standard-setter in artificial intelligence (AI) after passing a historic set of regulations. The EU AI Act categorizes AI technology based on risk levels and will come into force in May 2025. Some EU member countries have previously favored self-regulation, but the EU recognizes the need to challenge the dominance of major tech players. The EU also introduced landmark competition legislation to curb anti-competitive practices by large tech companies. Concerns over the potential misuse of AI, such as deepfakes, have been growing, but some companies are taking self-regulatory measures. The EU's AI Act is seen as a potentially influential model for other countries.

Mayo Clinic researchers have developed a new class of artificial intelligence (AI) called hypothesis-driven AI, which incorporates specific research questions or hypotheses to maximize the utility of data. Unlike traditional AI models, which rely solely on data, this approach uses existing scientific knowledge to enhance knowledge discovery efforts in medicine. Hypothesis-driven AI allows for a more focused approach to answering research questions and yields more interpretable results. In the field of oncology, this new type of AI has the potential to validate hypotheses, surface new insights, and contribute to cancer gene discovery, drug response prediction, and patient stratification. However, it also presents limitations, such as the need for specialized expertise.

European Union lawmakers have given final approval to the AI Act, making it the world's first comprehensive set of AI rules. The act will come into effect later this year and will serve as a global reference for other countries seeking to regulate AI. It adopts a risk-based approach, with stricter requirements for high-risk AI applications. Some uses of AI are banned altogether, such as social scoring systems and certain types of predictive policing. The law also addresses generative AI models, requiring data transparency, compliance with copyright law, and labeling of AI-generated media. Its impact is expected to extend beyond Europe's borders, as other countries and organizations are also working on their own AI regulations. The AI Act is set to become law by May or June, with enforcement by individual EU countries and potential fines for violations. There may be additional legislation in the future to cover areas not fully addressed in the current law.

China has successfully implemented artificial intelligence (AI) to manage its high-speed railway system, the largest in the world. The AI system, based in Beijing, achieves an 89% accuracy level and uses real-time data to quickly identify and alert maintenance teams to abnormalities. Its successful implementation has led to an 80% decrease in minor track faults and a reduction in rail movement caused by strong winds, ensuring smoother and safer railway operations. The AI system has proven indispensable for timely maintenance, keeping the rail lines in better condition compared to when they were first built. China's adoption of AI systems will increasingly play a vital role in the efficient management of the high-speed rail network.

Artificial intelligence (AI) tools are now being used by doctors to improve communication with patients, such as answering messages and taking notes during exams. While proponents argue that AI saves time and prevents burnout, concerns about trust, transparency, privacy, and the future of human connection have been raised. The use of AI in doctor-patient communication raises questions about the impact on patients and their consent, as well as the accuracy and potential mistakes made by AI tools. Doctors must ensure the accuracy of AI-generated notes, as large language models can occasionally misinterpret input or fabricate inaccurate responses. Balancing the benefits of AI tools with the need for a human touch is crucial. Privacy is also a concern, with healthcare systems needing to ensure that patient information remains secure when working with AI tools.

The United States is leading efforts to pass the first-ever United Nations resolution on artificial intelligence (AI). The draft resolution aims to ensure the safety, security, and equal access to AI technology, bridging the digital divide between countries. It highlights the importance of inclusive discussions on AI and the need for global consensus on safe and trustworthy AI systems. While not legally binding, the resolution would establish principles for AI development and use and provide a framework to leverage its potential while mitigating risks. The resolution has gained consensus support from all 193 UN member states and aims to accelerate progress towards the UN's Sustainable Development Goals for 2030.

Google's AI chatbot, Gemini, will no longer provide answers to election-related questions, redirecting users to Google Search instead. The decision was made to prevent the spread of disinformation during the numerous elections taking place worldwide in 2024. The move comes after Gemini generated inaccurate historical depictions of individuals and following incidents of AI-generated fake audio recordings spreading on social media platforms. Meta, the parent company of Facebook, Instagram, and WhatsApp, is also creating a dedicated team to address disinformation and the misuse of AI. In India, companies must now seek government approval before releasing unreliable or trial-phase AI tools.

Open AI's recently released Sora text-to-video AI tool has sparked excitement and concern among researchers. The tool can generate photorealistic videos from short text prompts, allowing users to create visual scenarios. Researchers worry about potential misuse and the impact on society, particularly in terms of spreading misinformation and influencing politics. Some propose using watermarks to verify the authenticity of AI-generated videos, but doubts remain about its effectiveness. However, there are also positive implications for the technology, such as simplifying complex text and assisting with data analysis. The introduction of text-to-video AI tools like Sora raises broader societal issues, requiring new ways of evaluating content and embracing potential benefits.

Researchers at Stanford University have utilized artificial intelligence (AI) models to uncover new discoveries about biology. These models were trained on raw data about the genetic and chemical makeup of millions of cells, without any prior explanations from researchers. Surprisingly, the models were able to independently analyze the data and identify a previously unknown kidney cell called the Norn cell. This breakthrough highlights the potential of AI-powered programs in biology, as they could uncover secrets about diseases like cancer and aid in cell manipulation. The development of these AI models poses both exciting possibilities and new risks, including the potential for the misuse of the technology.