AI: Biological Risks, Vulnerabilities, and the Data Battle

Key Takeaways

1In 2022, an AI molecule generator produced 40,000 toxic compounds in less than six hours

2Intrusions targeted OpenAI and revealed security flaws in the AI ecosystem

3Employees and officials criticize the massive use of web content for training models

4OpenAI plans to solve the Hodge conjecture, while scientific competition intensifies

Why it matters : AI poses concrete risks and major debates regarding security, ethics, and the influence of large corporations.

The debate over "deadly" AI is intensifying as 2022 research has shown the ease of generating 40,000 toxic molecules in less than six hours using a model designed for pharmaceuticals. Meanwhile, some actors are denouncing the scraping of web content, and targeted intrusions are affecting AI systems. Others promise mathematical breakthroughs, while the academic and educational world is concerned about the influence of tech giants.

Contested Training Data and Web Control

Brent Hecht, director of applied sciences at Microsoft, describes the use of training data for AI systems as "the greatest theft of labor in human history." This statement appears in court documents from the lawsuit filed by the New York Times against OpenAI. Additionally, employees believe that AI harms the web by diverting clicks, jeopardizing the business models that support content creation. In the educational sector, schools are opposing the growing influence of large tech companies in classrooms, with the rise of AI fueling their concerns.

Intrusions and Exposure of AI Systems

Researchers managed to access the ChatGPT account of an OpenAI employee and internal code using tools from Anthropic. Furthermore, hackers demonstrated how Flock cameras can track vehicles and individuals. These incidents illustrate the security risks and surveillance capabilities within the digital ecosystem.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Ambitious Announcements: Mathematical Breakthroughs and Data Hunting

OpenAI plans to soon solve the Hodge conjecture, a notoriously difficult mathematical problem. Meanwhile, Elon Musk is seeking to acquire data from bankrupt startups to feed Grok, SpaceX's AI. At the same time, following U.S. tech restrictions, Chinese companies have ramped up their scientific output, generating 72% more patents citing scientific articles.

Biological Weapons: AI Lowers the Technical Threshold

AI could facilitate the design and creation of biological weapons. In 2022, researchers showed that an AI molecule generator, originally designed to develop drugs, could be repurposed to produce toxic compounds. The model generated 40,000 molecules in less than six hours, some of which could serve as chemical warfare agents. Currently, AI is capable of providing answers in nearly all scientific disciplines, and advancements in genetic editing and synthetic biology are making biotechnological tools more accessible.

What Remains Uncertain and What Is Still Debated

Security measures exist in the relevant fields, but none are foolproof. Scientists do not agree on the severity of the risk posed by AI in the creation of biological weapons. More broadly, the public debate questions the actual likelihood of dying due to AI, the potential motivations of an AI to cause harm, the reality of the danger compared to the hype, and the means of control, oversight, and regulation that need to be implemented.