Tech Leaders and AI Experts Call for Six-Month Moratorium on 'Uncontrolled' AI Experiments

An open letter signed by prominent tech leaders and AI researchers urges AI labs and companies to "immediately pause" their activities for a minimum of six months. Notable signatories, including Steve Wozniak and Elon Musk, emphasize the risks associated with advancing technology beyond GPT-4. They advocate for a break to allow society to adapt to existing AI systems and ensure that these technologies benefit everyone.

The call for caution highlights concerns that the development of advanced AI systems, such as chatbots, is proceeding without adequate planning and oversight. For instance, Microsoft has integrated the GPT-4 model into its revamped Bing search engine, while Google has introduced Bard, its own generative AI powered by LaMDA. The push to release cutting-edge AI technologies has intensified apprehension regarding their safety and reliability.

The letter, published by the Future of Life Institute (FLI), underscores the rapid and uncontrollable race among AI labs to create increasingly powerful digital systems that even their creators cannot fully comprehend or control. Musk, who previously donated $10 million to FLI for AI safety studies, joins other global leaders in this initiative, including Marc Rotenberg, president of the Center for AI and Digital Policy, Max Tegmark, president of FLI, and author Yuval Noah Harari. Harari has also recently co-authored an op-ed in the New York Times addressing AI risks.

The letter's message builds on findings from a 2022 survey of over 700 machine learning researchers, where nearly half expressed a 10% likelihood of an "extremely bad outcome" resulting from AI, including the potential for human extinction. Furthermore, 68% of researchers indicated that more should be done to ensure safety in AI research.

Individuals who share concerns about the rapid pace and safety of AI development are encouraged to add their names to the letter. However, note that new signatories may not be verified, raising the possibility of inaccuracies in these additions.