184 | Gary Marcus on Artificial Intelligence and Common Sense

Sean Carroll's Mindscape: Science, Society, Philosophy, Culture, Arts, and Ideas

Sean Carroll | Wondery

Society & Culture, Physics, Philosophy, Science, Ideas, Society

4.8 • 4.4K Ratings

🗓️ 14 February 2022

⏱️ 84 minutes

🔗️ Recording | iTunes | RSS

🧾️ Download transcript

Summary

Artificial intelligence is everywhere around us. Deep-learning algorithms are used to classify images, suggest songs to us, and even to drive cars. But the quest to build truly “human” artificial intelligence is still coming up short. Gary Marcus argues that this is not an accident: the features that make neural networks so powerful also prevent them from developing a robust common-sense view of the world. He advocates combining these techniques with a more symbolic approach to constructing AI algorithms.

Support Mindscape on Patreon.

Gary Marcus received his Ph.D. in cognitive science from MIT. He is founder and CEO of Robust.AI, and was formerly a professor of psychology at NYU as well as founder of Geometric Intelligence. Among his books are Rebooting AI: Building Machines We Can Trust (with Ernest Davis).

See Privacy Policy at https://art19.com/privacy and California Privacy Notice at https://art19.com/privacy#do-not-sell-my-info.

Transcript

Click on a timestamp to play from that location

0:00.0	Hello everyone and welcome to the Mindscape podcast. I'm your host Sean Carroll.
0:03.9	If you've been paying attention to advances in technology, science, or just news in the world,
0:09.7	it's hard not to be impressed with recent progress in artificial intelligence, mostly driven by
0:15.8	neural networks and deep learning, machine learning kinds of techniques. We've really been able
0:20.6	to do things with AI that when I was your age we just couldn't do. For one thing, artificial
0:26.5	intelligence programs are easily able to kick the butts of human beings when it comes to games
0:32.1	like Go and Chess, which was considered very far away not too long ago. Another example is GPT3,
0:40.6	which you may have heard of, which is one of these language processing things where you can ask
0:44.6	it a question or you can give it a prompt in some sense and it will respond or will continue on
0:50.0	in the vein of the words that you gave it on the basis of the fact that it has read a lot of things
0:56.0	and it looks for correlations between them. Finally, and most importantly for everyday lives,
1:02.7	AI is everywhere around us. In vision recognition, recognizing the images that are in front of us
1:10.1	may be even self-driving cars or something like that someday. But certainly recommendations,
1:15.5	what music to listen to, what movies to watch, etc. AI is at work. So on the one hand is very
1:21.4	impressive. On the other hand, none of these are going to be confused for human beings. None of
1:26.3	these versions of AI are going to pass the touring test in some very advanced way. You can sort of
1:31.9	jigger up versions of the touring test that are passable by modern AI, but it's not a full,
1:37.0	blown general intelligence, right? AGI, artificial general intelligence, the kind of AI that would
1:43.8	really fool you into thinking it might as well be human. So today's guest Gary Marcus thinks that
1:50.7	he knows why. We're not able to do that. More importantly, he thinks that we're moving in the
1:55.9	wrong direction or focusing on the wrong things to make progress in this particular direction.
2:01.9	The idea is that there are certain kinds of things that neural networks deep learning is good at
	...

Please login to see the full transcript.

Previous episode | Next episode

Disclaimer: The podcast and artwork embedded on this page are from Sean Carroll | Wondery, and are the property of its owner and not affiliated with or endorsed by Tapesearch.

Generated transcripts are the property of Sean Carroll | Wondery and are distributed freely under the Fair Use doctrine. Transcripts generated by Tapesearch are not guaranteed to be accurate.