The Alignment Problem: Machine Learning and Human Values

Brian Christian (UC Berkeley)
simons.berkele...
Over the last five years, the field of machine learning has experienced incredibly rapid growth, and with that growth have come increasing concerns about potential harms, both present and near-future. High-profile successes have alternated with equally high-profile failures and controversies, and questions of ethics and safety have shifted from being somewhat taboo to comprising one of the central topics of the field. Out of this has come something of a movement: a vibrant, multifaceted, interdisciplinary effort producing some of the most exciting research happening today. Brian Christian, the Simons Institute's Scientific Communicator in Residence and author of the acclaimed bestsellers The Most Human Human and Algorithms to Live By, will survey this landscape of recent progress and the frontier of open questions that remain.

[@davedouglass438]

Autonomous artificial intelligences were legalized in the SCotUS' "Santa Clara County v. Southern Pacific Railroad Company, 118 U.S. 394 (1886)" ruling, and have rapidly increased in both power and stupidity (while suppressing the intelligence, mental health, and morality of living humans).
They are called "corporations too big to fail."
This is the underlying, and underdiscussed, crisis reflected superficially in such debates as that over "Citizens United."

[@alexanderkurz3621]

I think this is a very interesting observation. Not enough people are talking about it. The whole economy is an AI trained by a reinforcement learning algorithm that maximizes the objective function profit. This also means that there is another way to read the book. I dont know whether the author is aware of it. Namely, one can read the book as being about the problem of how to realign the economy.

[@iMeta4]

I will have to comb through this information once more if not check out the associated literature. Otherwise, I am grateful for the lecture on material I unfortunately did not have time or the focus to study further. Thank you for the lecture. I will revisit this another time.

[@finnk1289]

Superintelligent AI has to be stopped. We have little to gain, and everything to lose.