– And now to our truly
amazing keynote speaker, we’ve all watched lots
of movies about robots and computers that think
and feel and fall in love, but there’s probably no one in the world who’s thinking more profoundly about these possibilities and doing more to realize their potential than Professor Hod Lipson. Dr. Lipson, who is Columbia University’s James and Sally Scapa
Professor of Innovation, I like that, Professor of Innovation, also directs the Universities
creative machines lab. Dr. Lipson calls himself a roboticist and a mechanical engineer, but those terms hardly begin to describe the range of his work. To give you just a few examples, Dr. Lipson and his
students have been pioneers in the fields of multi material
printing and bio printing and more recently food printing which I’d love to learn more about later. Dr. Lipson created a program called “Eureka” that in a single day came up with a mathematical representation of Newton’s second law
and has since described the behavior of various chemical compounds and oxygen deprived yeast cells. Dr. Lipson is also the co-founder of four different companies
and his unforgettable 2007 Ted Talk on self aware machines, including one that teaches itself to walk, are among the most universally watched lectures ever given on
Artificial Intelligence. Obviously I could go on but fortunately for all of us, I don’t have to. Please welcome the innovation man himself, Professor Hod Lipson. (audience clapping) – Artificial Intelligence is one of these technologies that sort of they start slowly, they disappoint us, they’re under performers. But then something happens and then they exceed our expectations
when we’re not ready. And we are at this point today. So, if you look at what
Artificial Intelligence can do, you’ll hear a few smart people tell you AI is the most dangerous thing we’ve ever invented. It’s going to take over the world. That’s what Hollywood said as well. At the same time you have
a lot of smart people who will tell you AI is nothing but spreadsheets and analytics. Yes they get better with time but they’re not going to
take over the world soon and it’s just all hype. So what is it? Is it hype or is it going
to take over the world? That’s the question
we’re all struggling with and the reality is that AI doesn’t look like
either of these things. AI in reality looks like this. Like nothing. (audience laughing) Alright. You cannot see it. You cannot touch it. You cannot smell it. It is like air. It is everywhere. It worms its way into everything we do. It’s predicting the stock market as governing our economy. It’s predicting the weather. It’s predicting what song
you’re going to listen to. What product you’re going to buy. What link you’re going to click on. It’s listening to you. It’s talking to you. And this is the challenge
with Artificial Intelligence. It is everything and
nothing at the same time. It doesn’t matter if you study engineering or journalism or education, you need to understand how Artificial Intelligence will shape our future world and creates opportunities for you as well as create various risks and hazards. We’ve had Artificial
Intelligence for decades. So what has changed? So there’s really two ways in which Artificial Intelligence systems are built. The first one is what we call rules based AI and that’s the majority of AI systems out there. And that’s where you hire an expert in some topic. An expert programmer, they write rules conditional rules if
(mumbles) they explain the computer how to do something. The computer can execute that very quickly in a gazillion transactions. I’d say 99% of AI systems out there are based on these rules. But the challenge is that to make a rule based system, you have to have an expert that will tell you what the rules are. And experts as we know are slow, expensive and wrong. (audience laughing) What do we do without experts? The alternative is what we call machine learning data driven AI. AI that’s based on examples. You don’t tell the computer what to do, you show it. You don’t tell the computer what the criteria for
fraudulent transactions is, you give it examples of a few dozen. And the computer can calculate the odds of these things happening. Again, we call this
conditional probabilities. Now we humans, we don’t like probabilities. We are not, we have no intuition about statistics. We hate it. But computers love it. From supermarket to Tic Tac Toe to driving a car, can be done with this incredibly powerful single idea about learning
statistics automatically. But here’s the frustrating thing. Both of these approaches through AI existed for 50 years. So why did it take so long? 1956 IBM demonstrated a computer that can beat the world
champion in checkers. An incredible accomplishment. The New York Times writes
“The World is Over.” Machines can learn. But, the next thing that happened is amazing. Experts in the AI academia swoop in and kill the project. Why? They say this is
interesting but impractical. Where are you going to get data? How are you going to store data? Where are you going to transmit this data? This will never work. It is an infertile idea. Intelligence systems based on rules and logic and reasoning, what we like to think is intelligence. But there’s another kind of intelligence out there. Intelligence based on data and I like to… the equivalent of data driven AI is like intelligence based on intuition on experience. So a different kind of intelligence that’s maybe infrequently more powerful. And this is what is happening today is that that transition of rules better than data is beginning to change because we have more data. Up until a few years ago, most of that data with all the machine learning and all the computing power, all that data was useless. We call that unstructured data. Nobody with all the data
and all the computer power could write software that
will tell the difference between a cat and a dog. The needle is not moving on this problem of perception. If there’s one term you
have to remember in AI is deep learning. A car that can tell the difference, not just between what’s a car, but what kind of car it is. It is just because it was given examples of these things to learn from. So when you think about how your gonna innovate in this space and what the opportunities are remember that technology itself is free as a commodities universal talent. Is easily obtainable. The computing power is free. It’s what you do with this technology and what data you use to train your system. That’s where the innovation is happening. What can it track when it can see colors we cannot see and hear frequencies we cannot hear. We will not have words to describe what it can understand. We won’t have the imagination to think about what it can think about. The most powerful explanation of them all and the most overlooked and that is the Cloud. We use the Cloud all the time, but the Cloud means something else for AI. It means AI teaching other AI. It gets even more powerful when you have an adversarial relationship between AI’s. Run an AI that plays chess. You have it play against another AI that plays chess. They will generate their own fuel. And that is an incredibly powerful thing. AI moves forward in waves. Each wave brings a capability that was thought impossible in a previous wave. We’re now in the third wave in the middle. So what’s going to happen next? The first wave was rule based AI and it talked about the disappointing phase 50 years of rule based and it didn’t take us anywhere. Then, in the 90’s, we started with predictive analytics. The first generation of machine learning. The stock market uses Wall Street. That we use to predict that works with any tabulated data like stock data. We can use it to predict how many cucumbers we’re
going to sell next month. This is important stuff but it’s not the future. The third wave of AI
in which we are today, is the wave of cognitive computing. For the first time in history, we have machines that can understand images, video, text, audio. Things that we thought impossible just five years ago. But one area that is actually difficult for AI is having a conversation. When you think about the future of education, if it involves talking to people in the conversation, there’s no AI in the planet that’s
even close to doing this. Yes you can replace professors in academia cause we only talk one way, and replace us with a video, but anything early childhood education that involves having a conversation, no AI is even close to doing that. And not only is it not close, we have no clue how to do it. But we humans, we have a different kind of intelligence. We can generate new things. We can be creative. Creativity is at the crosshair of AI. Cause AI can create things, not just images like I showed you, but physical things. This is the biggest race in AI right now is who can build a AI that can create the next AI. So we lump all this big words all these big words emotions and freewill under this word sentients. Can machines be sentient? Many people think that the answer is no I think the answer is yes it will happen and it will happen when all this incredible capabilities turned inside and AI begins to think about itself. Rather than the world. So the big questions are ethical. Should we do this? How far can we keep up with this? Can ethics keep pace with this? Can regulators keep pace with this? Where are we going? A lot of people are terrified that we can make use this technology do bad things and we can and some people will. But I argue that by and far we will use this technology to do good things. So, the question is for you. How are you going to use this technology to make sure that we use it in the right way. Because we are at an
incredible transition point. A point that we will never look back and the future will be nothing short of amazing. Thank you. (audience clapping)

Leave a Reply

Your email address will not be published. Required fields are marked *