Those who have watched it, do you remember Richard’s speech at US Congress in Series 6, Episode 1?
Richard, awkwardly holding a big microphone box pulled out of the desk, speaks about the depressing reality of modern tyranny. Big internet tech companies track and monitor our everyday moves. They exploit collected data for profit. They are so powerful, and we the people are totally helpless under those modern kings.
Then Richard compares America before the Independence to where we are and conclude that we need a new, democratic, decentralized Internet to end this tyranny.
From a technical point of view, it may not be the issue of the Internet itself. It is mainly about AI. We could instead ask in this way: Can we build an AI platform that natively supports the fundamental values of American Constituion?
My invited talk presented at IEEE Services 2021 affirmatively answers this question (slides, paper). I began by talking about three specific requirements that such a platform must satisfy: Democracy, diversity and privacy. The key message is that there is at least one approach that can meet all of the requirements in a certain learning task. The original idea first came out in a top AI conference (IJCAI 19)*.
As Richard says in the show, the decentralized assumption changes the nature of learning algorithm in many interesting (but sometimes pathological) ways. My work is just tiny piece of a possible algorithm that meets the requirements. I hope that my talk has inspired some people for new directions in this field.
*Tsuyoshi Idé, Rudy Raymond, Dzung T. Phan, “Efficient Protocol for Collaborative Dictionary Learning in Decentralized Networks,” In Proceedings of the 28th International Joint Conference on Artificial Intelligence (IJCAI 19, August 10-16, Macao, China), pp.2585-2591.
Posted as a LinkedIn article: https://www.linkedin.com/pulse/decentralized-machine-learning-new-world-tsuyoshi-ide-san-ide/