About Creating AIs with Human-Compatible Values; Mark Stefik with BayCHI Program

July 9, 2024
baychi.org/cal...
Recent advances in generative artificial intelligence (AI) have led to wide public discourse about AI risks, guardrails, and regulations. An underlying concern is that AIs lack human-compatible values and goals. Achieving the intended purposes of the AI guardrails and regulations would require technology that does not yet exist.
Addressing difficult challenges requires asking the right questions. Multi-disciplinary perspectives and insights helps in finding and sharpening such questions.
What are values and human-compatible values and what is their nature?
What are the origins and utilities of values?
Why do different groups of people have different values?
How do human children and adults (and members of other social species) acquire values?
How could AIs be created that maintain alignment with human-compatible values?
Mark will consider how common misconceptions about values and their nature undermines current efforts to create human-compatible AIs. It is part of a vision of a bio-inspired research and developmental path where AIs and people work together with high collaboration skills and values to improve and sustain their common good. It is drawn from the second of four papers recently posted on arXiv: www.markstefik...