slide 2
AI governance is the idea that there should be a legal framework for ensuring that machine learning (ML) technologies are well researched and developed with the goal of helping humanity navigate the adoption of AI systems fairly.
slide 3
Sounds fair, and something that everyone shoudl be in favor of, and noone shoudl object to.
Notions of fairness, transparency, privacy, human safety, and explanability, are all good things, great to have, if possible. But is it feasible to acheive these qualities, Is it feasible to create AI that abides by, enforces these constraints?
Turns out operationlaizing these principles is hard. And not just becvause people are mean, it is hard due to deep techical reasons
slide 4
concrete codes of condduct: what are they fairness, transparency, privacy, human safety, confidentiality of personal records.
AI cannot analyze information in way that generates unfair or intrusive results
Human oversight.
slide 5
Defining "fairness" within a tecgnical context is hard
proxies
slide 6
equity definition is even more ambiguous and if done naively might be opposite to what you want
slide 7
Human supervision is not necessarily better. Remember humans have biases.
slide 8
The risk of state control. Is this really a cocern about AI?
slide 9
Misinformation: The risk of AI mediated culling of free -speech
slide 10
What is teh solution: democratize AI. Should not be only in the hands of teh state.
slide 11
Predictive policing example