AI Bibliography

Google CEO Eric Schmidt gave the opening talk, listing the recent advancements in AI that have enabled capabilities for solving previously intractable problems. He stressed that AI should forge a future that benefits “the many” instead of “the few.” Challenges in water purification, synthetic food production, logistics of distribution, and optimal management of supply/demand of energy are some of the areas where our ability to leverage natural flows of data governs our ability to meet our collective needs.

Schmidt envisions a world where research is “open,” contrasted with military-supported research, which is often classified and motivated by considerations other than the good of humanity. He discussed the importance of industry/academic collaboration, for example, by measuring advances on open real-world data sets available to the scientific community at large. He stressed the need for “platforms” for scientific advancement and the need for systems built on these platforms to be able to handle the dynamic nature of problems, and be inventive enough to be able to learn in parallel with their operation.

Facebook CTO Mike Schroepfer stressed a similar commitment to “10 year problems” and sharing data and algorithms, promoting open and peer-reviewed research. Facebook is eager to promote basic research by “installing labs next to where the scientists are.” 

The discussion among the scientists at the symposium was forward looking, projecting impacts and benefits, and humanity's longer-term future with AI. Five major questions emerged from this discussion:

• 1. Why is it different this time? The history of AI has seen several “boom–bust” cycles, where the optimism was driven by some perceived significant advance, followed by disappointment due to unrealistic expectations. We are currently in another boom. Is it different this time? If so, why?

• 2. How should we control systems that are potentially more intelligent than humans, whose working we don't fully understand? A version of this question that is more pressing is how should we control systems that are extremely complex and potentially more accurate than humans but can't directly explain their own behavior?*

• 3. Should there be an objective function for AI systems or is diversity more appropriate?

• 4. Are we likely to see ourselves replaced by robots for most tasks or augmented by machine intelligence? In the process, will AI create more jobs than it will destroy, or the other way around?

• 5. Is our current regulatory framework for governing the rights and actions of humans adequate for dealing with robots?