Evaluating lectures: AI and why people should be afraid

Evaluating lectures

Professor Stuart Russell, the Center for Human-Compatible Artificial Intelligence at the University of California, Berkeley, gave this year’s lecture for Wright. Four of his lectures, Life with Artificial Intelligence. Discuss the existential threat posed by machines more potent than humans – and point the way forward.

Last month, he spoke with News technology correspondent Rory Selan-Jones about what to expect.

How do you organize lectures?

The first draft I submitted was too sharp, too focused on the intellectual roots of AI and the different definitions of rationality and how they fit into history and the like. So I tuned in – and we had a talk that presented AI and perspectives on the future, both good and bad. And then we talk about guns, and we talk about jobs.

And then the fourth: “Okay, this is how we will avoid losing control of artificial intelligence systems in the future. Do you have a formula, definition of what artificial intelligence is? Yes, these are machines that feel and act, and we hope they choose actions that achieve their goals.

Everything else you read, like in-depth training, etc., are just exceptional cases. This is a continuum. The thermostat senses and acts and, in a certain way, has a little rule that says, “If the temperature is below, turn on the heater.

“If the temperature is above that, turn off the stove. So it’s a trivial program, and it’s a program written entirely by one person, so no training is included. On the other hand – you have a self-driving car where the decision-making is much more complicated.

Where it takes a lot of learning to achieve that quality of decision making.

We can’t say that nothing below counts as AI and everything above counts. And can it be said that there has been tremendous progress, especially in the last ten years? For example, in object detection we’ve been experimenting with since the 1960s. We’ve gone from being entirely pathetic for being superhumans of a specific size.

And in machine translation, we went from very pathetic to very good again. If you look at what the founders of the field said, their goal was general-purpose AI. Which is not a very good program in the game Go or an excellent program in machine translation, but something that almost any human can do. And probably more because machines have a huge advantage over humans in terms of bandwidth and memory.

Robot trucks, construction robots, construction management software will figure out how to build one, how to get a permit, how to talk to school districts and principals to find the right design for the school, and so on – and a week later, you’ll be in school.

Ella: