Peter Clark is a senior research manager at the Allen Institute for Artificial Intelligence, overseeing Project Aristo, an effort to build an AI system with the ability to answer questions from standardized exams, starting on 4th grade science tests, with the goal of advancing to higher grades in the future.
Clark was the featured speaker this week for the 2016 corporate kickoff event for Geeks Give Back, an annual philanthropic campaign presented by Bank of America in partnership with GeekWire, benefitting the Washington State Opportunity Scholarship program. Clark gave an overview of the state of artificial intelligence, and his work at AI2, before fielding questions from me and members of the audience.
Continue reading for highlights from the discussion, and stay tuned this fall for more information on how participating in this year’s Geeks Give Back campaign.
Todd Bishop: People look at artificial intelligence, and traditionally they think of a robot or a machine, but the internet is essentially the world brain. In some ways, artificial intelligence is about bringing consciousness to that brain. How has cloud computing changed what you do as an AI researcher?
Peter Clark: Cloud computing has pretty much revolutionized what people are doing these days. In the old days, you had your laptop or your big computer and did all your work on that. Now, most big organizations will run their computing problems on the cloud, a big distribution network of computers, so at AI2 we don’t have that many computers in the building.
All of our computation is done on Amazon’s cloud, and we can expand and contract the amount of computation we use at a whim. If somebody’s processing a million videos they’ll grab 10,000 processors and punch through that, so then they’ll be freed up and left for other people. That’s really revolutionized the way people think about computation. It’s less, “Do I have enough power?” It’s more, “Is it cost effective to buy the use of all these machines to process things?”
TB: When you look at Cortana, Siri, Alexa and others, is there one personal assistant that’s getting right, in your view, or that you feel is on a path to be real AI, maybe in 10 years?
Peter Clark: I don’t have a favorite, in that sense. What I would say is that all of these systems have little pockets of brilliance. Take Siri, for instance. Most things you ask it, it doesn’t really know about, but there are a couple of areas where it’s good. It knows about movies, and restaurants, and a bit about flights. The way the computer’s able to do that is somebody has very painstakingly crafted a model of what it is to eat at a restaurant. You need to make a reservation, you make a reservation at a time, time is something that exists in this continuous space, and you need to do that ahead of time, rather than before. … For a computer to have general intelligence, you would need hundreds or thousands of models like that.
I see those little pockets of brilliance expanding as time goes on. We don’t have a general way, at the moment, of having a computer acquire those things automatically. There are hints of things to come, I think, in those little pockets of functionality.
TB: Do you think that those types of assistants will always remain outside of our body, on devices that we hold, or how much potential is there for actual embedded devices in our own body? Like a little SD slot back here in our necks?
Peter Clark: Sometimes people use cell phones as if they’re a part of their body, they’re so glued to us all the time. I guess the nearest thing that comes to mind are many of these health applications where you have a watch which is providing a sensor. The new Apple watch, for instance, is kind of getting into being part of your body. It’ll vibrate, it’ll sense your body temperature, your pulse rates, it’ll alert you. In a way, that’s getting connected with your body and sending you little signals. There may be more of that. I don’t imagine we’ll be at the stage where we can plug a chip in and immediately assimilate everything on that chip, in the near future.
TB: That’s an interesting word, assimilate, for the Star Trek fans. Is there one robotics or AI movie that you feel has gotten closest to what our future reality will be?
Peter Clark: I’ve seen a lot of the science fiction movies. I’m thinking through all the movies I know. There’s a lot of poetic license in movies where they either ascribe to the machine warm, fuzzy feelings, like Wall-E or something like that, which is, I think, going to be different to the reality. Or they ascribe to the machine lots of evil intentions, like the Terminator movies, and I think that also does not represent the reality. “Her” might be something close, except the degree of intelligence embedded in that, that was almost human-level intelligence. Maybe in hundreds of years time. Still a long, long way off.
Q: Are you saying the robots haven’t won?
Peter Clark: The robots have not won. I should mention, as well, AIs have received a lot of bad press, recently. There’s been a lot of scare-mongering about the evil robots that are coming, and they’re about to take us over. I really don’t feel that’s the case. I feel there’s been a lot of hype about that. The state of technology is a lot more primitive than you might think. I think people transfer from the movies, thinking that’s reality. Really, machines really can only do fairly basic pattern recognition. There’s a lot you can do with that pattern recognition, like machine translation, but it’s still pretty primitive.
I think what these scare-mongering stories fail to do is also to recognize the huge number of advances and benefits that machines have for society. Techniques for avoiding accidents when driving, medical discoveries, personalized medicine, better access to information, these are huge benefits to humanity. I’d say at AI2 we’re very strong believers in AI for the common good. There’s a lot of great things to gain from that.