The new frontiers of AI and robotics, with Martial Hebert, Dean of Computer Science at CMU

Martial Hebert, dean of the Carnegie Mellon University School of Computer Science, during a recent visit to GeekWire’s offices in Seattle. (Photo GeekWire/Todd Bishop)

This week on the GeekWire podcast, we explore the state of the art in robotics and artificial intelligence with Martial Hebert, Dean of the Carnegie Mellon University School of Computer Science in Pittsburgh.

A seasoned computer scientist in the field of computer vision, Hébert is the former director of CMU’s prestigious Robotics Institute. Hailing from France, he also had the special honor of being our first in-person podcast guest in two years, visiting the GeekWire offices during his recent trip to the Seattle area.

As you’ll hear, our discussion doubled as a preview of a trip the GeekWire news team will soon be taking to Pittsburgh, revisiting the city that hosted our temporary GeekWire HQ2 in 2018 and reporting on the Cascadia Connect Robotics, Automation & AI conference, with coverage supported by Cascadia Capital.

Keep reading for excerpts from the conversation, edited for clarity and length.

Listen below or subscribe to GeekWire in Apple Podcasts, Google Podcasts, Spotify or wherever you listen.

Why are you here in Seattle? Can you tell us a bit more about what you’re doing on this West Coast trip?

Martial Hebert: We work with a number of partners and a number of industry partners. And so that’s the purpose of this trip: to establish these collaborations and strengthen these collaborations on various topics around AI and robotics.

GeekWire has been in Pittsburgh for four years. What has changed in the IT and tech scene?

Autonomous driving companies Aurora and Argo AI are growing rapidly and successfully. The entire network and ecosystem of robotics companies is also growing rapidly.

But in addition to the expansion, there is also a greater sense of community. This is something that has existed in the Bay Area and the Boston area for a number of years. What has changed over the past four years is that our community, through organizations like the Pittsburgh Robotics Network, has grown much stronger.

Are self-driving cars still one of the most promising applications of computer vision and autonomous systems?

It is a very visible and potentially very impactful application in people’s lives: transport, public transport, etc. But there are other applications that are not as visible and can also have a big impact.

For example, things that revolve around health and how to use health signals from various sensors – these potentially have profound implications. If you can have a small change in people’s habits, it can make a huge change in the overall health of the population and in the economy.

What are some of the cutting edge advances you see today in robotics and computer vision?

Let me give you an idea of ​​some of the themes that I find very interesting and promising.

  • One of them is not about robots or systems, but about people. And that’s the idea of ​​understanding humans – understanding their interactions, understanding their behaviors and predicting their behaviors and using that to have a more integrated interaction with AI systems. This includes computer vision.
  • Other aspects involve making the systems practical and deployable. We have made fantastic progress over the past few years using deep learning and related techniques. But a lot of that relies on the availability of very large amounts of data and organized data, supervised data. So a big part of the job is to reduce that reliance on data and have much more agile systems.

It seems that this first theme of detecting, understanding and predicting human behavior could be applicable in the classroom, in terms of systems to detect how students interact and engage. How much of this is happening in the technology we see these days?

There are two answers to this:

  1. There is a purely technological answer, that is to say how much information, how many signals can we extract from observation? And there, we have made enormous progress. And certainly, there are systems that can perform very well there.
  2. But can we use this effectively in interaction in a way that enhances, in the case of education, the learning experience? We still have some way to go to really deploy these systems, but we are making a lot of progress. At CMU in particular, with the learning sciences, we are very active in the development of these systems.

But what’s important is that it’s not just AI. It’s not just computer vision. It’s technology plus learning sciences. And it is essential that the two are combined. Anything that tries to use this type of computer vision, for example, in a naive way, can actually be disastrous. It is therefore very important that these disciplines are properly linked.

I can imagine that’s true across a variety of initiatives, in a bunch of different areas. In the past, computer scientists, roboticists, artificial intelligence people might have tried to develop things in a vacuum without the subject matter experts. And that has changed.

In fact, it is an evolution that I find very interesting and necessary. So, for example, we have a lot of activity with [CMU’s Heinz College of Information Systems and Public Policy] to understand how AI can be used in public policy. … What you really want is to extract general principles and tools for doing AI for public policy, and that, in turn, turns into a program and an educational offer at the intersection of the two .

It is important that we clarify the limits of AI. And I don’t think there are enough, actually. It is important even for those who are not experts in AI, who do not necessarily know the technical details of AI, to understand what AI can do, but also, more importantly, what it cannot do. .

[After we recorded this episode, CMU announced a new cross-disciplinary Responsible AI Initiative involving the Heinz College and the School of Computer Science.]

If you were new to computer vision and robotics, was there a particular challenge or problem you couldn’t wait to tackle in the field?

A major challenge is to have truly comprehensive and principle-based approaches to characterize the performance of AI and machine learning systems, and to evaluate that performance, to predict that performance.

When you look at a classic engineering system – whether it’s a car or an elevator or something else – behind that system there are a few hundred years of engineering practice . This means formal methods – formal mathematical methods, formal statistical methods – but also best practice in testing and assessment. We don’t have that for AI and ML, at least not to this extent.

It’s basically this idea of ​​going from system components to being able to characterize the whole system end-to-end. So it’s a very big challenge.

I thought you were gonna say, a robot that could serve you a beer while you watch the Steelers game.

This ties in with what I said earlier about limitations. We still don’t have the support to handle these components in terms of characterization. So that’s where I come from. I think it’s essential to get to the stage where you can make the beer delivery robot really reliable and trustworthy.

See Martial Hébert’s research page for more details on his work in computer vision and autonomous systems.

Edited and produced by Curt Milton, with music by Daniel LK Caldwell.

Comments are closed.