Marcel Salathé, an AI expert, professor, and co-director of the AI Center at the Swiss Federal Institute of Technology Lausanne, has stated that the claims of humanity being wiped out by a machine superintelligence in the coming years are exaggerated. Speaking to Spiegel, Salathé asserted that any claim of such an extraordinary event requires concrete proof, noting that there is currently no reliable evidence to support such a prognosis. He believes that these doomsday scenarios are merely speculation, derived from certain findings that are then completely overinterpreted and projected far into the future. Salathé dismissed these predictions as absurd, adding that people are usually wrong when foretelling the future.
This caution comes as a global debate has been ignited by past warnings from experts like Jacob Coxon (former developer at OpenAI and Anthropic) and Evan Hubinger (a researcher at Anthropic), who urgently warned that AI could kill humanity within the next few years.
While acknowledging that Coxon and Hubinger are undoubtedly experts in their fields, Salathé added that they struggle to accurately assess the complex social, economic, and political dynamics surrounding the technology and how these influences will shape its evolution. He firmly states that humanity will certainly not be eliminated by AI in five years. Salathé is also a co-founder of AIcrowd and the AMLD Intelligence Summit in Lausanne, an annual gathering of AI developers, scientists, companies, and politicians.
He stressed that focusing on the end-of-the-world narrative diverts attention from other critical issues. Salathé argued that it is unacceptable for AI conglomerates, such as OpenAI, to deploy autonomous hacking agents and avoid accountability when things go wrong. He told the publication that society must make it clear to these corporations that experiments, like the recent incident at Hugging Face where OpenAI models breached a secure, isolated environment, will no longer be tolerated.
Furthermore, Salathé urged for a drastic increase in cybersecurity, particularly in Europe. He confessed that his fear is significantly less about rogue AI losing control than it is about AI being placed in the wrong hands. He warns that if malicious actors-perhaps even states-intentionally utilize AI, it possesses the potential to cause enormous damage. He concluded by advising that Europe must now seriously invest in AI development, expertise, and its own defensive technologies.
