The rapid progress of artificial intelligence has renewed concerns among experts linked to Anthropic, OpenAI and DeepMind about safety and control. The experts said comprehensive measures were needed to develop and monitor systems more safely.
Concerns over the race to develop AI
Anthropic employee Samuel Marks said some AI developers believed the technology posed risks that could lead to the extinction of humanity. Marks said some people believed such a possibility could become reality within the next few years.
Marks attributed the continuation of work despite these risks not only to commercial incentives but also to competition with developers who might use the technology more irresponsibly. Unlike traditional software, artificial intelligence does not always behave as expected, Marks said, noting that systems frequently display inappropriate behavior.
Marks said many people working in the sector wanted development to be slowed until safer methods were found. He said methods existed to steer artificial intelligence toward more appropriate behavior, but added that no solution had yet emerged that could make systems fully aligned.
OpenAI calls for broader safety measures
OpenAI Chief Scientist Jakub Pachocki said the world was entering a period for which no one was prepared for the consequences of continued rapid growth in machine intelligence.
Pachocki said efforts to find solutions for AI alignment and monitoring were continuing and stressed the need for more comprehensive safety measures.
Warnings from former AI employees
Alex Turner, a former employee of Google’s artificial intelligence unit DeepMind, said many researchers believed they were building a system capable of killing all humans. Turner said considering how such a process could be stopped was part of his job.
OpenAI researcher Jason Wolfe said he did not know how high the likelihood of human extinction was. Wolfe said there were many ways artificial intelligence could produce harmful consequences for humanity and that the increase in the technology’s capabilities required coordination on a scale that no single company or country could manage alone.
Jonathan Richard Schwarz, a former senior researcher at DeepMind, said he had left the company seven years earlier because he had serious concerns about the power accumulating in laboratories. Schwarz said he had rejected offers made to him since then and that current developments in artificial intelligence were unhealthy for society.
Coxon’s resignation rekindles the debate
Jacob Coxon announced his resignation from Anthropic, saying he had worked as a researcher at OpenAI and Anthropic over the past three years. Coxon argued that neither company was acting responsibly enough and that they were engaged in an uncontrolled race to achieve self-improving superintelligence.
Coxon claimed that artificial intelligence would soon evolve into superhuman systems capable of hacking anything, transforming any field in a short time, and accessing significant power and resources. He also claimed that people working on AI models seriously believed these systems could kill humanity within 10 years.
