On this episode of Recode Decode, hosted by Kara Swisher, Carnegie Mellon’s Andrew Moore talks about the future of tech education as fields like artificial intelligence and machine learning take center stage. Moore, the dean of CMU’s computer science school, says he’s “concerned” that anti-immigrant fervor will deter the next generation of great computer scientists from coming to America, although CMU has not yet seen an impact on its application numbers.

“I think it’s short-term, and I haven’t seen any craziness, though of course, I’m frightened that it’ll happen — on this question of getting really the strongest folks over,” Moore said. “If we appear to have a society which doesn’t welcome folks from elsewhere then of course any sane brilliant scientist will end up going to Canada or Singapore or Zurich because they’ll be able to get the best of both worlds.

“Once you’re living in an academic community or in a software development office for an exciting company, usually in day-to-day interactions this doesn’t come up,” he added. “You’re so focused on some particular mission. But that perception — especially among someone who’s maybe 16 or 17 in anywhere from Turkey to China to England — is something I’m concerned about.”

On the new podcast, he also talks about the often-forgotten importance of electrical and computer engineers, who will develop the sensors that make machine learning advance; how educational programs have been complicit in the lack of diversity in tech; and why he’s personally pessimistic that self-driving cars, one of Carnegie Mellon’s areas of expertise, will be ready by the early 2020s, as some have predicted.

You can listen to Recode Decode on Apple Podcasts, Spotify, Pocket Casts, Overcast or wherever you listen to podcasts. Below, we’ve shared a lightly edited transcript of Kara’s full conversation with Andrew.


Kara Swisher: Today, I’m delighted to have Andrew Moore on the podcast. He’s the dean of Carnegie Mellon’s School of Computer Science, which was ranked No. 1 in the world by U.S. News and World Report. And he was previously a vice president of engineering at Google where he was in charge of Google Shopping. Andrew, welcome to Recode Decode. Thanks for coming.

Andrew Moore: Happy to be here, thank you.

So let’s talk. I wanna get your background. I’ve had various computer scientists on the show who are teaching and like that, and I’d love to get sort of the academic perspective, but you’ve been in the fray, also. So just let’s give your background, where you came from and how you got to Carnegie Mellon and then we’ll talk about what’s going on there.

I grew up in a seaside town called Bournemouth in South of England, and there, in the late ’80s, I really got into creating video games, like a lot of kids at the time.

Right.

Went and studied computer science at Cambridge University and then did a PhD on this big question of, “It’s so hard to program robots to do stuff …”

Mm-hmm.

“Can we make them learn to do it instead?”

Right, which has been the biggest challenge, obviously.

Yes. And that’s what I really fell in love with, this question of, to what extent can we help machines improve their own performance?

Their own performance. All right, we’ll talk about that later a little bit more. So you did, that but did you go right into robotics? Where did you go from there?

Subsequently, I spent some time at the MIT AI lab, which is super fun, working for Professor Chris Atkinson there. And I’m totally a math statistics guy, whereas he builds real robots. So, it was a trial by fire for me.

That’s a big robotics area.

Trained to actually build the physical robots and frankly I still suck at that.

So you’re not a mechanical engineer, exactly.

That’s right. I have huge respect for that.

Mm-hmm.

It’s the stuff to do with making things decide what they’re going to do next, which I’m really interested in. Anyway, subsequently, I joined Carnegie Mellon, really enjoyed sort of helping develop the AI classes there. Got super into using machine learning not only for robots, but for manufacturing.

Mm-hmm.

Because there’s so much that you can do to improve that. And I really enjoyed my time there, started to get interested in other big questions around computer science to do with things like can you detect near-Earth objects which are potentially dangerous using sort of fancy algorithms.

Yeah, it’s a Nathan Myhrvold thing, but go ahead.

Yeah. Can you get an early warning that there’s been an airborne disease attack on a city by noticing that the … perhaps the uptick in sales of medications following a stripe along the city in the direction of the airflow, for example. So, that was the cool stuff.

Right, that might be helpful.

Yeah. It’s all around this key thing that, if you can process a lot of data, your machines may be able to see stuff that no individual human could see, because we can only sort of ingest a certain amount of data.

Right, exactly. Which is the whole idea behind all this.

Yes.

So you were there at Carnegie Mellon and then you went to Google.

Yes.

Is that the only job you’ve had that’s not academic? Or was it? Yes.

Uh, yes. I did do a spectacularly unsuccessful startup for a while.

Uh-huh. What was it? I love a spectacularly unsuccessful startup, they’re my favorites.

It was machine learning consultant services.

Which — early!

Yeah, yeah. In the 1990s, we had a flashing neon sign on Craig Street near CMU which said “Data Mining,” flashing all the time.

Right.

We never got any walk-in customers, unfortunately.

Yeah. That today wouldn’t go over well. Go ahead.

What we loved doing there — we just didn’t figure out how to make money at it — was consulting engagements on applying machine learning in all kinds of places.

It’s just early. You’re just early, Andrew. That’s now the thing.

Exactly, exactly.

So you went over to Google. How’d you get to Google?

I was really impressed by the way that things were scaling so much. I made the move relatively late — it was in the mid-2000s — and the fact is I was very, very interested in this question of, what can you do with billions, or in some cases more than billions of observations?

Mm-hmm.

That’s what first enticed me. And so, we ended up starting up an office in Pittsburgh. And it was also important in my mind that Pittsburgh starts to develop its own ecosystem of software developers and research scientists rather than have everyone just migrate to …

Silicon Valley. Right. So you worked on shopping.

Yes, the office actually started out doing a lot of work on the machine learning that goes into the advertising systems at Google.

Mm-hmm. Which is enormous amounts of data that’s … you know.

Yes, it was really exciting and we had this team of statisticians, systems engineers, machine learning folks, algorithm folks. And the question is the only way — and this is still true — the only way advertising survives on a page like Google’s homepage is if it’s useful, and people find it and want to click on it.

Right, exactly.

So, there’s many interesting technology problems there, involving, making sure you’re showing relevant stuff for the query; doing everything you can to remove malicious stuff, which may have snuck into the system; detecting that malicious stuff in various ways; and then serving it up in a sort of a beautiful, useful way.

And so, people don’t realize, Google often starts offices in other cities and allows people to stay where they are, you don’t have to go to the Borg in Silicon Valley, which is in Mountain View. But they do that so they can avail themselves to universities and things like that. And Pittsburgh, people don’t realize — and we’ll talk a little bit about cars and things like that — has become one of the hubs of development and all kinds of really great technologists.

That’s true, and in fact, Google is not the first. Intel has had a lot there for a while. And for instance, Caterpillar is an example of a company which has worked alongside the Robotics Institute for a couple of decades.

All right, so you were at Google doing this. How long were you at Google? You were there for many years, if I recall.

Eight years.

Yeah, yeah.

Really, really cool, exciting years.

And talk about what you did there, shopping especially. What was the concept around shopping?

So, the interesting thing about, when somebody’s trying to purchase something, there are different patterns. There are patterns of people who just say … For instance, “I’m building a cabinet and I need this kind of hinge brace.” For that, what you want to do is create an experience where the person can as efficiently as possible narrow down on what that is.

Right.

There are other experiences where it’s not just about the end goal, it is about getting that opportunity to choose among many different objects and stuff like that. And at that point, the interesting thing is the most value that great companies like Google, Amazon, Microsoft can bring is to help people choose between a bunch of options and present those options to them in a useful way.

Right, right. And the correct options.

Yes, exactly.

The correct search results, which is much harder than people realize.

It is. It’s also really painfully obvious. Someone searches for a red party dress and one of the results is a toaster. Even if you got 49 out of 50 results right, that toaster looks really stupid.

Right, absolutely. So one of the things, Google did try to get into shopping in a much more significant way and had a lot of stumbles, you know, with Froogle. Do you remember Froogle? Were you involved in Froogle?

Froogle was actually the state of play when we arrived.

Right.

And it actually was an amazing idea for the time.

It was. Great name.

Yes, indeed. So, this was during a period where the whole internet was learning a lot and one …

Explain what Froogle is to people that don’t know.

Froogle was, initially, a cool project which turned out to be very popular when it was displayed on Labs. And it was this question of … Okay, I’m gonna break this down in the way that I was thinking as an engineer: Traditionally, a web search engine, a good web search engine worth its salt, would look at a query that a user had typed in, a few words, and then would run through and match it against millions and eventually billions of documents, and it scored every single document to how well it matched the query.

Right.

And then it would show the results of summarizing those documents. The brilliant idea about Froogle was that it was going to do more work beforehand than merely capturing the documents and storing an index for them. It was going to understand the components of the pages so that it could quickly see that, although this might be a page which is talking about vineyards in Italy, it’s also got explicit mentions of these five types of grape. So it could then use “this page is actually about this type of grape” semantically, rather than simply looking for string matches, if you like.

And so you got there, and obviously at the time, Amazon was starting to come on, which has outdone Google in the shopping area. Talk about why Google thought that was important. Obviously, search was its most important thing, but it moved off into lots of different branches. Shopping would be an obvious one for them.

Yeah, well, I can’t speak for Google the corporation because it never was and never will be, in my opinion, a very much command-and-control center. But for me personally, the thing I was seeing was, as time goes on, people don’t want to simply use web search to get hold of documents. They want to use web search to do stuff.

Do stuff, absolutely. Which Google realized.

Yes, exactly. And so it’s about getting verbs, if you like, into the queries.

That’s a really great way of putting it.

And rather than just sort of pieces of information.

“I want to, give me, I’d like to buy.”

Yes, exactly.

And then be the seller, I guess. Google never really wanted to be the full seller the way Amazon did, which I think is probably Amazon’s power.

That was the big difference. And it still is. It’s two different businesses. One is to sort of try to help make sure that you are aggregating information from millions of different retailers so that users don’t have to individually go visit all those retailers’ web pages and search over them. Whereas Amazon’s model, which has been amazingly professional and successful, was to actually be doing the logistics and transportation and shipping themselves.

And so you worked there, and what do you think you accomplished in shopping? Because it’s gone through so many different iterations at Google, it continues to shift around and where it is and — what they’ve tried to do now, and you’re not there, is position themselves as the alternate to Amazon. You know, that’s their sale to the big retailers, because Amazon is now in competition with a lot of retailers. And in fact in heavy competition with them, and now they own retailers. Whereas they’re going to be buying even more stuff. It just goes on and on.

I have not really been involved since I left Google. But during the time at Google, one of the things that really drove us was this question of, “What can we do for regular retailers?”

Right.

And many people had direct or indirect friends who were running small retail businesses and were very concerned about how they can make sure what they’re doing is visible.

It’s amazing that people didn’t see Amazon sneak up on them. I was like, “They were behind your back!” “Oh, they’re my friend.” I’m like, “They’re not your friend!”

Yes.

“They’re not your friend!”

My understanding is that, throughout many of the larger retail companies, this has been a big looming thing, and different retailers have moved at different speeds with saying, “Look, we’ve got to seriously invest in a cloud and online offering.” And for many retailers there has been this question, “Well, maybe we shouldn’t do that. Perhaps our whole strategy should be about walk-ins and people having the experience of the physical shopping.”

Right.

So, this question of what to do during a disruptive period in shopping has been raging among huge numbers of people.

Yeah, it’s really interesting with where we are now. Of course again, as I said, I used to say, “The call’s coming from inside the house, retailers, you better just … you’re gonna die if you don’t be careful.” So then you left Google to go back to academia.

I did.

And why did you do that? Because a lot of people like to stay in the fray. And obviously a lot of academics spend a lot of time in the fray.

Yes, it was fascinating. What I noticed going on in the world of academia was its really centrally important role for the economy and for the future. One of the biggest trends going on among the tech companies, and what’s going to keep them alive, is how much strong software and machine learning, AI expertise, they can get hold of. And that was turning out to be the limiting factor.

Mm-hmm.

That limiting factor is actually super serious. The reason it’s so serious is the folks with the skills, if there’s not enough of them, they will tend to flock to the places which are most welcoming and set to take them. So when you’ve got a huge undersupply of AI experts in the United States, those that do remain are going to be fought over by very deep-pocketed, strong internet companies who are providing important services. But, suddenly you start to see other organizations — like, even things like NASA or Veterans Administration or construction companies, all these other things …

Everybody.

… absolutely, right now need to bring in advanced technology. Really, [they’re] just being blocked because they can’t find anyone.

They can’t, yeah.

So my concern started to be that the biggest problem or need in the future is gonna be, “Where do you get the supply of computer scientists from?”

It’s almost a crisis, really, in terms of … It’s essentially Google and Facebook and others sucking up all the talent, and we’ll get into the diversity of the talent in a minute, but … So, you went there. Talk about what Carnegie Mellon’s doing. Most people think of Stanford and some other schools in California — [Cal Poly] and stuff like that.

Talk about how you look at computer education now, and we’ll get into the diversity issue in a minute, but how do you look at where we are as a country? Because I just recently interviewed Mark Zuckerberg and he talked about the problem. China, obviously, the worrisome nature of what’s going on China is that there are so many people … They’re just pushing out AI experts everywhere. Sort of look at the state of where we are with graduates in computer science.

They’re still heavily in demand. When I look at the whole map of what’s needed, there are two distinct populations of folks we need to train up. They’re both equally important, but they are different. No. 1, the people who can take existing technology, the great systems like TensorFlow or AWS Web Services or all these things, integrate them together to make new products.

Right.

There’s a second group: Who are the people who are gonna be designing the new things which eventually replace those current systems? So, there’s the folks building the technology that the next 20 years will be based on, and then there’s those taking the existing technology. So, the great computer science places — MIT, Stanford, Georgia Tech, Berkeley, CMU — their main responsibility, I believe, is to produce that second type, the people who are going to be building what’s next.

There is, however, a serious need for both types of developers. And again, I’m not saying one is stronger or more important than the other. It’s absolutely pointless having folks invent new algorithms if there aren’t people to take them and actually look at how to change the world with them.

Similarly, someone on the planet should be figuring out how to make software more energy-efficient, more able to prevent disasters in each of these things. So, two groups. Carnegie Mellon School of Computer Science focuses, rightly or wrongly — and I think rightly — on the second group, the folks designing the next generation of what’s going on, and so for that, we still have a dangerous undersupply.

Mm-hmm. What do you mean by “dangerous?” I agree with you, but explain to people who don’t understand the crisis we’re in.

Okay. Yeah, and in fact, I wouldn’t call it …

It’s a crisis.

Yeah. Yeah.

It’s a crisis.

Yes.

It will be.

I think it is potentially an economic opportunity lost.

Right. Well, that, too, obviously.

Yep. So, the one thing I’m observing in my current role, which I’m really enjoying, is there is such a difference in the technology levels of different organizations. There are people in companies doing a fantastic business, but their infrastructure is based on 1990s technology. Others, early 2000s, and others, 2010s. 2010s are where you often see parts of a business able to take in sensory data, like for instance, watching to see if anyone’s tripped and fell in a factory so that you can quickly get help to them, where it involves real computer vision and pretty advanced engineering.

Other folks definitely won’t have anything like that. They may have extra people walking around to try to detect that kind of thing, but it’s gonna be a long time for them to tool up on it. And the advice, if a company wants to move ahead with something like that, is you have to have some internal expertise. Even when you hire consultants from the big consulting companies to come in and help you implement them, you have to have some internal expertise.

Right, and so when you’re thinking about what you’re doing, sort of as opposed to the other schools, you have a certain focus. How do you differentiate yourself, then? How do you attract people into the area?

Yes. All the big schools do have slightly different characters.

Yes, they do.

And I do think we’re different. I’m not gonna say that we’re awesomely better in all ways.

You can do that if you want.

Yeah. Okay.

Yeah.

My British upbringing would never allow me to do that.

All right. Okay. So, you’re not gonna rag on Stanford for me?

Yep, exactly.

Yeah. All right.

But Stanford is awesome.

Okay.

What made me particularly love Carnegie Mellon and sort of ally myself with it is, it really thinks of the goals of its faculty and students to be around impact. Famously, one of my predecessors, Jeanette Wing, who was head of computer science at CMU, in a famous faculty promotions meeting, listened to some folks comparing bibliometric statistics about, “Well, this person’s had this number of conference publications and this number of oral reports,” and she stood up, and she said, “We don’t care about those numbers. Look at what this person has done. Their technology has brought …” I don’t remember the incident specifically, but, “It has brought …”

Impact.

“… a change in the way that a certain part of society operates.” And we’ve always had that kind of focus, and it’s my personal belief — and I think it is the general thing which unites us at CMU — that you actually get the best science by focusing on impact. The field of computer science is so rich. There’s so much to explore which is completely unexplored at the moment that it is very easy to get lost in all the things you can do, but having a guiding light, that helps pull you to develop the things which are actually gonna help society.

Do you have a lack of people coming into the system? Obviously, it’s one of the top programs so you’re not gonna have a … But when you look around, are there fewer people in this country going into that? Is that a problem? Or there’s not enough schools? Or what do you imagine to be the problem, because everyone talks about these pipeline issues constantly.

Yes.

How do you look at that?

So, there is somewhat of a pipeline problem, generally. One of the things that we’re all super concerned about and taking action on is, specifically, the pipeline for women and underrepresented minorities is weaker than the rest of the pipeline.

What is your assessment on that? I bang on that drum all the time, but it’s often used as an excuse, sometimes. I think it’s real, in part, but at the same time, I don’t think companies put it as a priority. It’s No. 14, which means it’s No. 14. It’s not not in there. It’s just, they have other things they’re worried about. So, how do you look at that problem and what to do about it?

I’ve learned a lot while I’ve been at Carnegie Mellon, and it’s kind of humbling. When I arrived, I was of a very traditional mindset, which I’m embarrassed about now, which is, “Yes, it’s a huge pipeline problem. I should do everything I can to encourage and help the people earlier in the pipeline, sort of increase their part of it so that I can then absorb what they produce.” And the thing I’ve learned is that’s not the case. All of us in education and in early career management are part of that pipeline.

Right.

So, specifically, it’s so easy to see ways that you’re gonna lose women or underrepresented minorities who have made it all the way through high school if you’re not giving them a positive environment, one where they feel appreciated and a part of an overall organization. And then, even further than that, if you’re saying, “Have a great time at university, but when you’re out in the real world, you’re still gonna suffer from …”

Yes, I agree with you.

“… forms of unconscious bias,” and so forth, then it’s hard to retain them through the program.

Right. That’s the interesting part. Everyone was focusing so much on recruiting — which I think you need to do — but once people get there, wherever it is, that’s where we all seem to fall off the car, as far as I can tell, is that there’s no management track. It’s not managed differently. It’s not imagining that people have different needs in the way they’ve been managing people, but it’s all kinds of things, but it’s keeping people there, which is always … Seeing a track towards promotion, that’s within the companies. How do you change the university? I know at Harvey Mudd, they’ve tried different things. They’ve identified issues, social issues, the way they’re doing classes, the way they’re conducting classes. What are you doing? What is your …?

So, there’s a suite of things that you have to be doing. Let me run through some of them. First one is to be really just welcoming and thorough when you’re looking at the match of individuals to the university. This approach, which was pioneered over the last two decades by CMU professors such as Professor Margolis and Lenore Blum, has really helped make sure that we’re not inadvertently dropping excellent students who maybe aren’t your traditional cis white male. That has been spectacularly successful. We have seen … Not only are we now at 50 percent women coming into the program, but the retention through the program is indistinguishable between men and women. The final results in the program are indistinguishable, and so this … Maybe if there had been some concern 10 or 20 years ago that somehow …

We wouldn’t have it. Yeah.

… that something else would get broken, and even admitting lots of students, you’d see a big drop-off later on. That hypothesis seems to have failed.

What’s the change that you think is most important, of those changes?

It really helps to be clear about something, which we’re lucky Carnegie Mellon already believed, that it’s about impact, and the real job of a computer scientist is to be thinking, “That thing’s a problem. What can we do about it? And how can technology and social systems help?” That’s my definition of a computer scientist. It’s problem-solving.

Right. Right, which is more attractive to women, actually, that’s been shown, the goal-oriented …

There may be some correlation there.

Yeah. There is, actually.

But I think it’s important for-

There’s been studies. I don’t know if it’s …

Yeah, yeah. So, I think that is something which … Because we talk the talk on that, we don’t simply walk the walk, that can really help.

Mm-hmm. So, where do you imagine, when you’re thinking about China and other countries — I’m using China just as the proxy, but it’s pretty much the most challenging country compared to the U.S. The U.S. had led the way in computer science. It’s led the way in creation of internet companies. It’s led the way in creation of billionaires and etc., etc., and startups. How do you look at the situation now? Do you see it as a competition with them? Or do you … Only because Mark Zuckerberg brought this up in a podcast, he’s like, “What they’re doing in China, you shouldn’t be hindering me, because look at China,” and he’s sort of, this “Me or Xi” kind of thing — it’s not the choice I want to have. But how do you look at … But I do think he’s absolutely correct in that that country really is all-cylinders on educating, creating legions of computer scientists.

Yes. So, the American experience in science and technology in the last 50 years has been very successful. It has maintained and really earned the status as the place in the world that you want to come. And that has meant that you can think of places like MIT and Stanford and CMU as the Starfleet Academy of computer science.

Yes, they are. I knew we had to get a “Star Trek” reference!

Yes, exactly.

Wait. Are you a “Star Wars” or a “Star Trek” person? “Star Trek,” right?

Definitely “Star Trek.”

Yeah, of course. Yeah.

Yes.

Yeah, because “Star Wars” is depressing. But go ahead. Move along.

So, that strategy, which worked so well in the development of aerospace technology and the initial development of computation, and in more recent development of the really important principles, like multithreading, multiprocessor, multi-core, and computer vision and things. That has worked very well, and that aspect, where the United States really kind of gets and maintains an unfair advantage …

Really, truly.

… by making sure that it gets hold of the best people, does work, or has worked in our interest for a long time. And you can tell by my accent — I actually revealed this at the start — I immigrated here.

You’re not from here.

Yeah.

You’re not from these parts!

Right, because I totally …

You have a funny accent.

Exactly.

As my relatives in West Virginia would say, “You talk funny.”

I talk funny. But it would never have occurred to me growing up that I wanted to be anywhere other than the United States. I wasn’t sure whether I would qualify to get in, but it was the central point, and that still stands, despite everything. I think that around the world, once people are thinking primarily about the science and what new technologies they’re going to develop, they want to be part of this place.

In this country.

Yes.

But.

But.

But … Do the “but” part.

The numbers still mean that there is huge numbers of folks who will stay where they came from, and because China has had their sort of strong improvements in its educational system, that means …

And living standards.

Yes. That means it has got a very large group of folks who are creative and trying to do similar things without being in the United States.

Right, right. Not the copying community. They’re now creative and doing their own … Which is really fascinating. People are always like, “Oh, China copies.” I’m like, “Not so much anymore.” I think that that’s an old trope.

Yes, I agree.

It’s a different set of values, too.

Yes.

The government is quite involved.

Yes.

Every government’s gonna put its sticky fingers in technology, no matter what you do, but in that case, it’s a very different value system.

Yes. So, at that point, if you consider things to be a numbers game, we’re doomed. I don’t think it is just a numbers game. If, on the other hand, we maintain this role as the place that the very top folks want to come to, then I think that will make sure that … This is me speaking in my words, not for anyone else. The biggest advances in technology happen in a liberal Western society with sort of democratic, transparent values.

Right. Well, we’re gonna …

That’s what I think is important.

You did mention, by numbers, we’re doomed in terms of graduating people, and of course jobs are very important in this country. How do you look at the current state of the anti-immigration stance of this administration, the not-so-lovely science, they don’t love science as much as perhaps they should. Does that have an impact? Or is it just a short term until science sanity is returned?

I think it’s short-term, and I haven’t seen any craziness, though of course, I’m frightened that it’ll happen on this question of getting really the strongest folks over.

Okay.

But if we appear to have a society which doesn’t welcome folks from elsewhere then of course any sane brilliant scientist will end up going to Canada or Singapore or Zurich because they’ll be able to get the best of both worlds.

Sure.

So, that is a concern. It’s about perception, and once you’re living in an academic community or in a software development office for an exciting company, usually in day-to-day interactions this doesn’t come up. You’re so focused on some particular mission.

But that perception — especially among someone who’s maybe 16 or 17 in anywhere from Turkey to China to England — is something I’m concerned about. Numerically, schools like Carnegie Mellon School of Computer Science, the very strong technology schools, we have not seen this impacting the interest in folks coming into our programs. Every year we are having to accept a smaller and smaller fraction of our applicants because there are so many applicants.

Yeah, they’re only letting two people into Stanford this year out of one million.

Sometimes it feels that way.

Yeah. No, it is that way, it is that way.

That’s going on okay for us.

Right.

There are other disciplines, things like law and business, are suffering somewhat. The broader set of universities in the United States are also really important to our economy. Maybe not top five, but top 100.

Also, important to tech. There needs to be all kinds of people.

Yeah, exactly. Those folks are starting to see drop-offs, so I think that is a leading indicator of a problem.

Especially when tech is so immigrant-focused.

Yes.

You know, in terms of who has come here, you can name one after the other. There’s someone who’s come from somewhere else.

Yes. Now, I do think to some extent we in academia have brought this upon ourselves.

Because?

For the following reason.

All right.

As funding in computer sciences remained relatively flat over the last 10 or 15 years — by the way, I should mention that this administration is actually very positive towards computer science funding, and OSTP, Office of Science and Technology Policy, is really, in its list of priorities for the country, most of them are around computer science.

It would be nice if we had a CTO, but go ahead, or a head of that office, but okay.

As a — [I’m] not whining about that aspect, but it has stayed flat, while the number of computer science has needed to grow, and so there is less grant funding to go around.

Right.

So, a starving dean or department head looks around and says, “Where can I get revenue?” And master’s programs, mainly based on bringing in relatively wealthy students from Southeast Asia, has been a huge money boom.

Sure, they can pay.

Yes. I think, again in the top schools, the value that those students get, it’s paid off in few years of working in their industry whether they stay in the United States or go home. So, it’s still a positive value proposition, but it’s pretty expensive. So we academics shake our heads and say, “Oh, it’s too bad that most of our master’s programs seem to be filled with international students. We need to increase the number of domestic students.”

But the cost is …

The reality is, yeah, the cost of these programs is high, and departments who have got sort of addicted to the money —

They sell the seats.

— and they’re planning research revenue, they need that.

They can sell the seats.

So, there are many creative things that can be done to reduce the cost of these kinds of pieces of education. Georgia Tech’s mostly online system is an example of that.

Sure.

I think that’s a place where academia has to go now, is to actually value-engineer its expensive master’s programs creatively so that people can afford, domestically, to take them.

Right, absolutely, a hundred percent, because you have all this demand, people who can pay. So I’m gonna finish up talking about the concept of where the big trends in academics are going around computing. What do you see as important? Obviously, robotics. It’d be remiss in saying that Carnegie Mellon was a big player in Uber’s, a lot of self-driving cars, a lot of robotics, a lot of things like that. That had sort of a rocky tenure.

Yes.

Let’s blame Uber for that. Talk a little bit about where it’s going at Carnegie Mellon and where, overall, you see the most important areas of computing going.

Very good. The huge change of course we’ve all seen in the last 10 years has been machine learning, and the real push on these convolutional neural networks which turn out to be able to solve problems that we hadn’t been able to solve using more statistical methods in the past.

You know they’re calling it “super AI” in Silicon Valley now. Whatever.

Yup.

It’s a new marketing term.

We’re on the radio, but if we weren’t on the radio you would have seen me eye-rolling.

Yes, there was eye-rolling, I know. They have all kinds of names for it. Go ahead. Just marketing, I don’t mind.

Yup. So, here’s the important thing: That machine learning component fits in a slot of all the technologies you need to build an AI system. One of the things we’ve really been pushing on, both in our development of education and in our recruiting of faculty, are the other slots adjacent to machine learning.

Right.

An important one which I usually would draw before machine learning because machine learning depends on it is all the sensing work that’s necessary to be able to understand the world.

Sensors, that’s an astonishing area, but go ahead.

Exactly. If you’re using robots to fight a fire, they absolutely need to understand what’s really going on in the building, and so creating devices in robots which can actually understand what’s happening right now is, I think, if I only had one research topic that I could work on, I would regard that as much more important than improving the algorithms which are gonna take that sensory data.

You know, it’s gonna be amazing when they figure it out. I was talking to someone. I like to talk to futurists, and they were like, at some point —and I think it’s actually being created — say there was a nuclear spill, a chemical spill, or something like that, that they would have sensors that were small as like grains of sand and they’d throw them on, like from far away, they’d spray them onto something, and these sensors would pull in the information about what spilled and what to do about it. I loved the concept of it. I’m sure it’s not possible at this point, but that’s the idea.

Yeah.

Sand is the way I looked at it. Like, they’re so pervasive they’re almost in the air without knowing they’re there.

Yes, and I do think we’re moving in that direction. You’re totally making this point. The idea of just trying to really focus on machine learning without being able to get ahold of data is a killer.

Right, to get the data.

One really important part of that turns out to be power.

How to power those things.

Actually having a sensor which is tiny, is processing high-definition video, you can’t have it sitting next to a big GPU of the kind that they’re now putting in cars.

Right.

Because that would be all the weight for a mobile platform, and so forth. For me, AI, a lot of the stuff that we, mathematicians like me, are doing in the middle of it will be stalled without that huge growth of work in cheap low-power sensing.

And sensing everywhere, going out to space, going out inside of people. I mean, remember that movie where they traveled inside a human being? I’m like, “They’re gonna have sensors all over human beings at some point.”

If we can deal with these things.

Right.

So, a very interesting aspect of that is, in my opinion, electrical and computer engineering departments, which to some extent, have had to watch computer science getting all the glory, they are totally coming back with the importance of these things.

Do you all war with each other? Like, oh man.

No, no, we’re friends, but still.

They’re sitting there with the wrenches going, “What the hell? We used to be cool.”

Yeah, if you look at the media, it’s like “machine learning” this and “intelligent agents” that. Whereas now, us AI people are going back with our hats in hands to electrical and computer engineering. This happens at Carnegie Mellon, it’s folks and ECE primarily who are helping us solve this sensor need.

And power.

So, that’s one end. There’s another end, above machine learning. If you’ve got something which can spot patterns or notice that your elbow is in an image and stuff like that, you’re still gonna want to put it in a system which makes decisions.

Right.

That really is the original dream of artificial intelligence at the Dartmouth conference, is things which can observe, think about what they observed, and then act. So, making that action decision is really important.

Right.

And it goes in two ways. One of them is a little bit like what we were talking about with Google. The early days of this is, a human has to make a decision, and there’s much too much information around for them to actually, really be able to look at all the source information themselves. How can you support that? That goes from everyone doing stock trading to helping decide if a medical claim is legitimate through to people. One of our professors, for instance, is instrumenting classrooms so that teachers can notice if they’ve accidentally got some unconscious bias which means that they’re not attending to certain kinds of students, and so forth.

Right, right.

So that’s great, that is human assistance. In my opinion, many folks in the artificial intelligence industry — and by the way, I’m a minority here, so I’m not speaking for the whole discipline — are focusing on that because it’s so much more palatable and less scary than the other thing at the top of the stack, which is autonomy.

Right. Meaning Terminator.

Not meaning Terminator.

I know.

Of course everyone rightly thinks of this.

They don’t care about us, robots don’t care about us.

Indeed, they’re toasters.

Yeah. Or we’re toasters, one or the other. They don’t care.

That’s right.

They don’t care. You know, oddly, I did an interview with Elon Musk many years ago, and I thought he was quite intelligent. He goes, “They’re gonna think of us like house cats.” Like, why do they want to kill us? It’s kind of useless to want to kill us. It was really interesting. I was like, “Oh yeah, you’re right.” We’re so obsessed with ourselves.

Until we’re into science fiction future territory, the reason they don’t care about us is they’re not thinking about us. They are simply machines. Every robot or computer that you can see is just a set of equations.

Right. We only have just a few minutes, just two more minutes. I’d just love to know where self-driving is, because that’s where you were, lots of people left Carnegie Mellon, they set up different shops. Where do you see that right now, self-driving/autonomous vehicles?

This is me speaking for myself …

Yes, yes.

My personal understanding is that, every year now in the major self-driving experiments, the metric of success to track is, what fraction of the time do you need a safety driver? In other words, what fraction of the time does the human need to take over control? If we were shooting for the early 2020s for us to be at the point where you could launch autonomous driving, you’d need to see every year, at the moment, more than a 60 percent reduction every year to get us down to 99.9999 percent safety.

Right.

I don’t believe that things are progressing anywhere near that fast. It’s an engineering task where chipping away 20 percent this year, 4 percent next year, and so there’s a whole set of problems.

It’s a combination of these around sensors, AI, all kinds of things, all working together.

Yes. Again, I’m not in the thick of this, and of course I don’t know what’s going on in the proprietary world, but I imagine, based on what I’ve seen, that these questions as to what to do about a human who’s on a curb, potentially looking like they might lose their balance and really making a decision about that. That’s not something you’re gonna solve in two months. It’s probably part of a three-year subproject, and then you look at the other cases like that.

They’re all over the place.

Yup. Every set of shadows from trees has got so many dots and things that there is a chance that something’s going to suddenly emerge which looks like an animal or something.

So humans still have a chance, for a short time. Very last question: What’s the craziest thing you’d like to work? If you could pick up any tech topic besides the ones that are sort of real, what would you like to see?

This is gonna sound terrifying, but I still want to say it: I would like it if I was talking to my machine every night before I went to bed, and it actually asked me, “Andrew, was today a good day?” And at the end of every year it asks me, “Andrew, was today a good year?” And we actually started to use data and activities to help understand some of these sort of meaningful aspects of our lives. Are you actually spending enough time with your kids? Are you doing your socializing in such a way that you seem to actually be getting value from it?

Right.

I know it’s crazy, but I do think that this kind of observing of us is gonna help.

It’s crazy. Maybe a private drone will do that for you, like sitting behind your head. “Oh, move along.” It’ll buzz you if you’re on your computer too much. I think they should electrify phones and they just start to like, “Ow!” you cannot pick it up. I have all kinds of ideas of how to do that.

It’s possible. Again, if we’re talking crazy, my guess will be the computer will be in one of our teeth.

Yeah, I agree. All right, we’ll end on that. Andrew, thank you so much. This has been Andrew Moore, he’s the dean of Carnegie Mellon University’s School of Computer Science. It’s been a great session, Andrew. It was great talking to you. Thanks for coming on the show.