Sandy Pentland on AI, Future Systems and Regulation
Alex "Sandy" Pentland[1] is somewhere near the top of the AI pantheon these days, here's his most recent talk, "Engineering Ecosystems with AI", given online on Friday 15 Sept., https://www.youtube.com/watch?v=0m8EsEmZPYQ at the MIT Mobility Forum. In the intro it is mentioned that this was a dry run for Pentland's upcoming keynote to the U. S. National Academy of Sciences. From the YT summary,
This talk covers how our society is having difficulties engineering heterogeneous systems of people and technologies; for instance, our systems for dealing with pandemics, climate change, inequality, or financial stress have been less than completely successful, in significant part because of unanticipated human behaviors.
He discusses recent work on models of panics, cascades and other highly nonlinear, long tailed phenomena--using aggregated census data--as an example of how AI and big data can be socially useful.
Then he takes the data further, showing that when people move between communities (both locally and internationally with migration) is when progress really happens. AI may have a similar "melting pot" effect, helping mid-level workers/earners (which he defines) in closing some of the performance/creativity/earnings gap to high-level workers. In particular, AI developed on restricted and well defined training sets may not get the headlines of ChatGPT, but will be very useful in many fields.
Near the end he quotes Xi Jinping of China to make it clear how big this social change is going to be (paraphrased):
Xi is the largest representative, loudest voice for Marxism and he recently said, 'data is a new primary means of production along with capital and labor'. And if you think about that, what he's saying is that classic Marxism is done. It's now not a battle between capital and labor. It's a battle between data, capital and labor and that sort of gives you a sense of the magnitude of this problem.
If you look at what society did with capital and labor, it took a century or more, for instance, to form labor unions to pressure companies, to establish principles to get laws enacted. ... And it's not a fixed thing, it's not like you can do it once and it's done. It evolves over time. So currently we're in a new evolutionary phase of labor and a new evolutionary phase of capital. The problem with data is that we don't have *any* institutions, we don't have *any* norms for it. It's new, so we're back in the robber baron era of capital, we're back in the early industrial age where kids were working 14 hour days. That's where we are with data. Just face it! What we have to do is develop the right institutions to be able to deal with this now-critical element of society.
I don't watch many videos, but this one was well worth the time. His actual talk is about 25 minutes, the rest of the hour is intro and many questions/discussions at the end.
[1]Here's Pentland's short CV, https://www.media.mit.edu/people/sandy/overview/
Read more of this story at SoylentNews.