Berkman Klein Center Panel at Harvard Discusses the Limits of Artificial Intelligence
Researchers from Anthropic and Microsoft discussed artificial general intelligence, safety, and misuse risks at a panel held at Harvard.
Researchers from Anthropic and Microsoft evaluated the limits, safety, and potential risks of artificial intelligence at a panel organized by the Berkman Klein Center at Harvard University.
AI Panel Held at Harvard
Researchers from Anthropic and Microsoft laid out differing views on the limits of artificial intelligence at a panel organized by Harvard University's Berkman Klein Center.
The first session of the event series, which examines how artificial intelligence reshapes human relations, beliefs, and creativity, drew the interest of more than a hundred attendees.
Microsoft Representative's View on Artificial General Intelligence
Ken Archer, a product lead in responsible AI at Microsoft, stated that he believes artificial intelligence will never reach general intelligence.
Arguing that predictions of machine superintelligence lack an awareness of historical scientific advancements, Archer emphasized that his belief in artificial general intelligence is zero.
Anthropic Representative Points to Lack of Experience
Jackson M. Kernion, who manages the human feedback program at Anthropic, stated that the difference lies in experience rather than capability.
Kernion expressed that language models lack the continuity of experience that humans bring to every interaction.
Misuse and Safety Concerns
When asked what calls for a slowdown should be used for, Kernion stated that his biggest concern is misuse rather than out-of-control capabilities.
Kernion argued that artificial intelligence is a dual-use technology that can be used to protect systems or infiltrate them, and said that time needs to be dedicated to safeguards.
Model Training and Values Framework
Kernion noted that Anthropic published the constitution used to train its Claude model, stating that this represents an approach to the values the system should possess.
He noted that this framework does not represent the world, but rather reflects a specific San Francisco-based mindset.