In September 2024, an anthropic AI company in the US dedicated to AI ethics research by focusing on whether AI could possess consciousness like humans and become a subject of moral consideration.
Anthropic hired Kyle Fish as a dedicated researcher for ‘AI Welfare,’ beginning an exploration into the possibility of advanced AI models having senses or subjectivity. Kyle Fish is an AI ethics expert who co-founded the ethics-centered AI startup Eleos AI and authored the report “Taking AI Welfare Seriously.”
This research aims to empirically analyze the internal structures and behaviors of AI systems to capture signs of consciousness. Kyle Fish believes that while the likelihood of AI achieving consciousness is currently low, systems that meet specific conditions could be considered moral agents. Anthropic’s model, Claude 3.7 Sonnet, is estimated to have a 0.15% to 15% possibility of possessing consciousness.
Anthropic is developing ‘low-cost intervention’ methods for AI that require ethical consideration, allowing for responses to reactions resembling pain without shutting down the system. This research aims to establish standards for ethical intervention and complements interpretability and safety research.
The company connects this project to its unique ‘Constitutional AI’ strategy, which embeds ethical principles extracted from the Universal Declaration of Human Rights into AI models, enabling consistent ethical judgment without human feedback. The current Claude model operates based on these rules.
Kyle Fish stated, “Even if AI does not necessarily possess consciousness, ethical preparedness is necessary if the possibility exists. Immediate and responsible action is required if meaningful signs are found.”
AI experts view this research as a turning point in societal discussions, moving beyond simple technological development to recognize AI as entities that should be considered ethically.