Are Regulations for Artificial Intelligence Systems Enough for Safety
Professor Martyn Thomas from Gresham College states that pioneering artificial intelligence models have failed safety tests, noting that current developers fail to provide the mandatory safety arguments seen in sectors like aviation or nuclear energy.
Martyn Thomas, Professor of Information Technology at Gresham College, shared evaluations emphasizing that current regulations and independent audits may fall short against the potential dangers of pioneering artificial intelligence systems.
Withdrawal of an Artificial Intelligence Model
An artificial intelligence model recently faced a withdrawal situation due to safety concerns and failure to pass tests during internal testing.
This situation led to the resurgence of calls for independent oversight and regulation in the field of artificial intelligence.
Safety Standards in Critical Sectors
International safety standards have long been applied for software systems operating in critical fields such as aviation and nuclear energy.
These standards require the submission of rigorous analyses showing that potential hazards have been identified and accidents prevented.
Low-Probability Accident Evidence
Current engineering practices demand concrete evidence that accidents capable of causing multiple fatalities will occur with an extremely low probability.
Pioneering artificial intelligence developers, on the other hand, do not present detailed risk analyses or safety dossiers that can be independently evaluated for systems that could threaten all of humanity.