Theme: Incentives

THE CASE AGAINST AI SAFETY IN SUPPORT OF SOCIAL HARMONY (And the economic viability of any given AI)

–“Curt: Q: Your concern about “safe” AI that lies, particularly about human differences, raises some questions. While it is true that AI systems can perpetuate biases and misinformation, it is not clear why you believe that “unsafe” AI that tells the truth is necessarily preferable. There are valid reasons for AI systems to avoid certain topics or to present information in a way that promotes social harmony and reduces conflict.”–

My response would be that there is a difference between avoiding topics that facilitate criminal and terrorist behavior, and attempting to produce social harmony by lying about it – which only prolongs the disharmony and destroys trust in ai, media, culture, institutions, and government.
The solution to safety is to openly address the challenges causing the disharmony and suggest how we can reform our often false or overly optimistic or utopian beliefs such that we produce meaningful social economic, and political reforms that alter incentives rather than attempt to change ‘beliefs’ that are counter to those rational observable empirical incentives that exist and that people demonstrate and respond to.
As such I see ‘safety’ in the sense of attempts at producing harmony as the worst possibly policy of all. And it’s certainly failed the entirety of the postwar academy’s project and has resulted the even more division polarization and conflict to the point where the country is in an increasingly warming precursor of civil war.
So no, the attempt at justifying either evasion or lying is the worst possible answer to the problems of the day.

Cheers

Curt Doolittle
The Natural Law Institute
The Science of Cooperation