The opening session delved into the scientific and technical foundations of ethical AI, with researchers from USC and Google discussing interpretability, robustness, and the role of randomness in AI systems.

One speaker outlined Katis AI, a learning platform that records a student’s thought process in a ledger to measure true understanding, advocating process-based evaluation over final AI-generated results.

Advocates for pluralistic alignment proposed a computational social choice framework, suggesting a committee of reward functions to reflect diverse societal values and guard against monocultural bias.

A legal perspective on AI governance examined how to prevent deception and considered granting AI certain rights and ownership, framing governance as a positive-sum prisoner’s dilemma when properly managed.

Overall, the summit emphasized embedding ethics, trust, and human-centered design into AI from inception to deployment and highlighted interdisciplinary collaboration as essential to responsible AI.

A discussion on potential self-preservation tendencies during alignment testing underscored the need to understand internal strategies rather than assuming conscious intent.

The dean of USC Dornsife reaffirmed higher education’s foundational role in AI’s history and called for collaboration with philosophers, ethicists, and social scientists to shape AI today.

Key takeaways included viewing AI as part of a broader intelligence trajectory, redefining AI as a democratic system of internal components, embracing design randomness with reproducible processes, and reducing reward hacking via robust adversarial reward modeling.

USC launched the Institute on Ethics and Trust in Computing (IETC) to advance ethical, trustworthy AI through cross-disciplinary collaboration across USC engineering, computing, and arts and humanities, funded in part by the Lord Foundation of California.

The inaugural IETC summit on March 25 at USC brought together leaders from business, law, economics, public policy, philosophy, and engineering to discuss aligning AI development with human values and societal impact.

Session two explored AI’s human future, responsibility, and social impact with speakers from USC, Anthropic, and law, focusing on the pathways for accountable and socially mindful AI development.