Jeremy Howard: Will Artificial Intelligence Be The Last Human Invention?

8 hours ago
Originally published on April 21, 2017 1:27 pm

Part 4 of the TED Radio Hour episode The Digital Industrial Revolution.

About Jeremy Howard's TED Talk

Data Scientist Jeremy Howard has studied machine learning for 25 years. He says artificial intelligence can help achieve amazing things. But he warns the impact on jobs may cause a great deal of social instability.

About Jeremy Howard

Jeremy Howard is a data scientist and the founding researcher at — a company dedicated to making deep learning accessible.

Previously, Jeremy was the CEO of Enlitic, an advanced machine learning company. He was also the president and chief scientist at Kaggle, a community and competition platform of over 200,000 data scientists.

In addition to his research, Jeremy is a faculty member at Singularity University and a researcher in residence at the University of San Francisco.

Copyright 2017 NPR. To see more, visit


On the show today, ideas about the digital industrial revolution and how we as humans will fit into all of it.

MAURICE CONTI: So we're going to be, you know, Iron Man plus Spock plus, you know, name your series of superheroes.

RAZ: This is Maurice Conti.

CONTI: Yeah, so my name is Maurice Conti. I currently head up Applied Research and Innovation at a software company called Autodesk.

RAZ: And Maurice is talking about how maybe someday we could all be these sort of superhumans working together with machines in an era he calls the augmented age.

CONTI: I mean, you know, we've had this partnership with our own technology that has been allowing us to achieve greater things, which, I guess by definition, is augmentation. I think the difference this time is the speed with which these technologies are coming online And the speed with which we are going to adopt them, which, I think, will result in something that feels more like augmentation rather than improvement. It's like a superpower.


RAZ: OK. So what does it actually mean, I mean, in practical terms? Because right now - right now we have access to infinite knowledge, right? We can just pull out our phones and find out really pretty much whatever we want to find out.

CONTI: Yes, sort of. In fact, you know, I argue that we are already augmented because all of us probably listening have access to a smartphone, which is, in turn, connected to the Internet, which, in turn, holds vast amounts of information. I wouldn't necessarily say it's knowledge. In fact, that's maybe the next step. The part that's missing is, yes, I have access to a great body of information, but I still need to turn that into knowledge with my own point of view, my own ability to think synthetically and connect dots.

RAZ: Yeah.

CONTI: And I think, you know, that's what advanced computation is going to help us with. It's really processing that information, gleaning insight from it, making intuitive leaps and so forth.

RAZ: And Maurice says that soon, a lot of our machines won't just answer questions. They'll actually start thinking for themselves, coming up with their own ideas and even feeling the world around them. Here's how Maurice explained it from the TED stage.


CONTI: Tools are making this leap from being passive to being generative. Generative design tools use a computer and algorithms to synthesize geometry to come up with new designs all by themselves. All it needs are your goals and your constraints. I'll give you an example. In the case of this aerial drone chassis, all you would need to do is tell it something like it has four propellers, you want it to be as lightweight as possible and you need it to be aerodynamically efficient. And then what the computer does is it explores the entire solution space, every single possibility that solves and meets your criteria, millions of them. But it comes back to us with designs that we by ourselves never could have imagined. And the computer is coming up with this stuff all by itself. No one ever drew anything, and it started completely from scratch.


RAZ: OK. If the future is going to be a future of artificial intelligence, do you think that we're going to figure out a way to work together or, I mean (laughter), or not?

CONTI: Well, you know, part of me says that has to be true. The AI was built and is operating in service of some goal that I was trying to achieve, and if I'm achieving that goal, then I think, by definition, I mean, I'm in partnership with that AI.

RAZ: So, I mean...

CONTI: Feel free to call [expletive] on that.

RAZ: No, no, no, I just - I just - and I was saying this to Erik Brynjolfsson earlier in the show - right? - that I cannot imagine any single profession, including professions that we think require a whole lot of brain power like psychiatry, OK, I mean, I cannot think of a single profession that won't be - that couldn't be displaced by a machine-learning machine that just becomes smarter and smarter and smarter.

CONTI: Yeah. Certainly, it's easy to imagine how just about every profession can be affected by these technologies. You know, where it gets interesting is if we start to debate, will these professionals be displaced? And certainly, in some cases they will. But I think in many cases, they won't. Part of the reason is there's this graph that I like to draw, and it has two curves on it. The first curve is an upward exponential curve that represents our capabilities as a species based on the development of technology. So perhaps at some point, when we reach a very high point on that curve, we'll be done. There'll be nothing else to do. What I think a lot of people overlook is there's another curve right next to it that happens to be in lockstep, and that is the curve of opportunity.


CONTI: And it's actually easy to see, you know, whenever some company comes out with the latest smartphone, the next day people are like, great. What's next? I want more. And I think it's the fact that we have this other curve, which is our capacity to imagine and desire better things, better lives, better relationships, that, you know, will keep the demands on us increasing. And if that's true, then the AIs that would displace us are actually just what we need in order to keep up.


CONTI: So as computers are going to augment our ability to imagine and design new stuff, robotic systems are going to help us build and make things that we've never been able to make before. But what about our ability to sense and control these things? What about a nervous system for the things that we make? Our nervous system, the human nervous system, tells us everything that's going on around us. But the nervous system of the things we make is rudimentary at best.

For instance, a car doesn't tell the city's public works department that it just hit a pothole at the corner of Broadway and Morrison. A building doesn't tell its designers whether or not the people inside like being there. And if the designers had known what was really happening in the real world with their designs, they could've used that knowledge to create an experience that was better for the user. Now, what's missing is a nervous system connecting us to all of the things that we design, make and use.


RAZ: I mean, a nervous system, a human nervous system, is such a complex thing that it's hard for me to get my head around the idea that there could be a comparable digital system.

CONTI: I think when a lot of people think about AI and, you know, things like Skynet, one of the things they assume is that the AI is aware. And I think awareness is a prerequisite of powerful, useful AIs. And so giving AIs a nervous system, the ability to perceive their environments in order for them to carry out their jobs, is critical. And that's already happening. This isn't science fiction at all. It's just a question of getting sensor data from the real world into the AI. It's happening in your car every day. If you have a car that has any kind of autopilot or fancy cruise control, it's looking around it. It's perceiving its environment and then making decisions based on that. And so, really, it's just a continuation of that, more sensors, more real-time data coming in, that allow the systems to make relevant decisions that are helpful to us.

RAZ: So I mean - and, I mean, we've heard stories about this on the show already about how, you know, when humans and robots work together, everybody wins, you know. Like, I mean, is this idea of the human-AI collaboration, you know, is infinitely superior to everything else out there? I mean, is that always going to be the case, or is that just wishful thinking?

CONTI: No. I think human technology, you know, human-AI, human-robot collaboration is better than either one working on their own. I actually think that's a tautology. You know, humans are very good at some things. Synthetic systems are really good at others. All else being equal, I cannot imagine the argument where the combination of these two skill sets is not better.


CONTI: And that's where it gets interesting because the AI is currently not that great at intuition but really good at that sort of brute force computation on tons and tons of data. And it's a little bit like a dance.


CONTI: The human might lead, but the computer can do lots of fancy moves that together combine into something...


CONTI: ...That is greater than the sum of its parts.


RAZ: Maurice Conti - he's the director of Applied Research and Innovation at the 3D design and engineering software company Autodesk. You can see his full talk at On the show today, the digital industrial revolution. In a moment, a question - is it possible we're inventing the last human invention? Stay with us. I'm Guy Raz, and you're listening to the TED Radio Hour from NPR. Transcript provided by NPR, Copyright NPR.