Page 24 - AI Governance Day - From Principles to Implementation
P. 24
AI Governance Day - From Principles to Implementation
Figure 9: Rumman Chowdhury, CEO, Humane Intelligence, during the Multistakeholder
Panel
Challenges in AI governance
• One of the primary challenges in AI governance is the nascent understanding of AI
models and their implications. Balancing the benefits and risks of AI requires a deep
understanding of these technologies and their potential impacts.
• Interoperability at the international level is crucial to enable small companies to scale
across borders and participate in the global AI ecosystem. This requires detailed
workstreams and ongoing collaboration among countries.
Risks of the AI race
• The competitive race among AI companies often prioritizes market dominance over
safety, leading to the use of unlawful shortcuts. Negative consequences include the
development of AI models that can hack into potential safety limits implemented.
• Governance frameworks must shift the incentives towards safety and responsible
development to mitigate these risks. The historical failure to anticipate the externalities
of social media serves as a cautionary example.
Preparedness and science-based governance
• Effective AI governance must be rooted in scientific principles, with preparedness work
based on reliable scientific data. This includes leading in safety and capability while
ensuring models are not prematurely released to the public.
• There is a need for standardized evaluations and methodologies to assess AI models,
considering their probabilistic outputs and the challenges in creating consistent tests.
14