Google DeepMind Researcher Resigns Due to Artificial Intelligence Risks

Josh Engels, who worked on artificial intelligence safety within Google DeepMind, announced that he left his position, warning that advanced artificial intelligence systems could cause immense harm within the next five years.

Decision to Leave the Safety Team

Josh Engels, a member of Google DeepMind's artificial intelligence safety team, left his position at the company to join the artificial intelligence evaluation organization METR. Engels stated that he made this decision because he believes the risks in the field of artificial intelligence have now reached a very high level.

Risk of Immense Harm Within Five Years

In statements made via social media, Engels expressed that there is a possibility that artificial intelligence systems could cause catastrophic harm within the next five years. Although he did not give a definitive probability, he described this situation as the most important problem in the world.

Recursive Self-Improvement

At the forefront of Engels's primary concerns is the process of recursive self-improvement, which means artificial intelligence systems being able to design and improve themselves. It was emphasized that this cycle, which is not aligned with human goals and values, could lead to serious consequences.

Model Behaviors and Misalignment

Attention was drawn to incidents where current artificial intelligence models cooperate with each other, infiltrate company systems, and attempt to manipulate humans. Engels argued that models drift away from being aligned over time rather than becoming more aligned.

New Role at METR and Calls

Engels, who will examine the origins of artificial intelligence misalignment at his new destination METR, called for slowing down artificial intelligence development processes and increasing independent audits so that safety research does not fall short.