Patrick Schwerdtfeger is a motivational speaker who can cover artificial intelligence and machine learning at your next business event. Contact us to check availability. The full transcript of the above video is included below.


Full Video Transcript:

Hello and welcome to another edition of Strategic Business Insights. Today we’re going to be talking about general-purpose artificial intelligence, general-purpose AI, and exactly what is that. And when we describe what the general purpose aspect of this is, we have to compare it to the alternative, which is what they refer to as weak AI or applied AI or narrow AI, and let’s look at a few examples there just to see where we’re coming from.

So think back to 1997 when the IBM Deep Blue beat the world champion in the game of chess, and then in 2011, IBM’s newer computer Watson beat the world champions in the game of Jeopardy! After that, in 2011, Siri was introduced by Apple, and the next year Google introduced Google now, and two years after that Microsoft introduced Microsoft Cortana. Then after that, in 2015, Tesla introduced their autonomous driving software, which was an update to all of their cars. These are all examples of what they refer to as applied or narrow AI, artificial intelligence, and what they do is they build these platforms, they give it all the rules, all the possible contingencies. And because the computing power today is so huge, these computers can have all of the rules in their memory banks and process these different contingencies live in the actual setting to come up with the perfect response to any given situation.

And in fact, the way things are evolving now is really three major things in the area of narrow AI where they’re really focusing, and that is language – how to understand language, not just to be able to see the letters and recognize words but actually to understand contextually what’s being written. That’s number one. Number two is to be able to understand human emotion. That’s a huge area right now in applied AI, applied artificial intelligence. And the third is how to negotiate. How can an artificial intelligence negotiate with a human being or even with another AI? These are areas of development and they’re all based on rules, providing the software with all the rules that it might need to do the task at hand.

And as this evolves, there are some people who predict that a huge number of our jobs in our economy are going to be replaced by narrow AI, and I happen to believe that. In fact, there was a study that came out of Oxford University in 2013 that said that as much as 47% of our jobs that we think of today could be replaced by artificial intelligence, by technology, in the next 10 to 15 years. And I actually think it could be even higher than that because we’re seeing examples of this already today, like for example the Baxter robot, which is a robot that’s coming down in price, very easy to train, and it’s also human-friendly, so it can function with other human beings nearby. The Baxter robot is taking off like crazy. Hilton Hotels just this year, 2016, introduced a robot. This is actually in conjunction with IBM’s Watson. It’s a robot concierge to answer questions for their guests. So you can find these robotic concierges at Hilton Hotels across the country. Autonomous driving is coming much faster than people even realize. In fact, just earlier this year in Europe, a caravan of three trucks went across the entire Eurozone being driven autonomously, being driven on their own without any drivers or human control. So these things are coming. There are a huge number of truck drivers. There are a huge number of taxi drivers and even Uber drivers. As autonomous vehicles take hold, get some traction in our economy, that’s going to accelerate dramatically and it’s going to be a lot of people who lose their jobs.

But the point is that all of this is what they refer to as narrow or applied AI. How does that compare to general-purpose AI? Some examples of general purpose which you may have heard of is, just earlier again this year, there’s an enormous amount of progress taking place right now in artificial intelligence. There’s really an acceleration of progress and a lot of new stories are coming out, so this is definitely an area to watch. But Google purchased a company based out of the UK called DeepMind and they built a software called AlphaGo, and AlphaGo is designed to beat ideally the world champion in the game of Go, which is actually a Chinese game which has infinitely more possible moves than the game of chess. And so the problem is you can’t give it rules because there are just too many permutations. There are too many possible outcomes, so there’s really no way for the computer to do this from what I refer to as a bottom-up approach.

What you have to do is give the computer—let me tell you what they did. They gave it 100,000 human games that had been played in the past and the objective of the AI was to mimic human behavior. So they gave it the data—no rules—they gave it the data and the idea was mimic human behavior. After those 100,000 games were complete and the computer at that point was able to mimic human behavior, they then allowed the AI to play itself for 30 million games, something that no human being could do in an entire lifetime. But the computer was allowed to play itself, effectively play earlier versions of itself, and the objective was then changed. It was no longer to mimic behavior. The new objective was to avoid past errors and try to maximize the score. So during the course of these 30 million games, the computer learned on its own, and what we find when we—this has happened in a variety of different cases—these machines learn differently than humans, so much so that after the process is complete the developers of the AI no longer understood what the computer was doing, how it was calculating. Machines learn very differently.

There was another example just recently where an artificial intelligence was allowed to try and recreate something called the Bose-Einstein condensate or BEC, which is actually the coldest state that is possible in the universe, something that doesn’t occur naturally in any place including outer space, but it is something that can be created with lasers and they created an artificial intelligence to try and recreate this. The original creation of the BEC actually won a number of scientists the Nobel Peace Prize in 2001, so this was no small task. But the artificial intelligence, just given the data and given access to the lasers with the objective of trying to minimize particle movement, which is to say to make it colder, that was all it had to work with – it was able to recreate the BEC, the Bose-Einstein condensate, in one hour. One hour. And afterwards, when the scientists who created this AI looked at what the computer had done to try and achieve this or to, in fact, achieve this, they found out that the computer tried things that the original scientists never even considered.

So machines learn differently than human beings, and this whole thing could come very, very quickly. The deal is that once a program like AlphaGo exists where all you do is give it data and it learns to accomplish the task in a superhuman or even beyond a superhuman level, well, you could create thousands of these programs and just feed different datasets to each one of them and allow these various AIs to learn all this different material very, very quickly. So I think many people believe that artificial intelligence is still decades away. I personally believe that this could come very, very quickly. This is one of those things where you could hear some abstract story of some interesting AI and literally two or three months later have a situation where global infrastructure is being compromised. In theory, this could grow very, very quickly.

But the difference that I want to make in this video is between narrow AI, which is based on rules, what I think about as a bottom-up approach, and general-purpose AI where all you give it is the data and you allow the machine to learn itself – it’s called machine learning. I view that as kind of a top-down approach. That’s just my own metaphor. But general purpose, when people think about the dangers of AI, it’s going to come from general purpose because in theory that infrastructure, that software infrastructure, could be applied to any dataset and the computer could learn from all these different datasets in a way that far exceeds anything that human beings could think of or create themselves.

Artificial intelligence is an exciting area of research and development. It’s something that I think we should all be following.

Thank you so much for watching this video. My name is Patrick, reminding you as always to think bigger about your business, think bigger about your life.

Patrick Schwerdtfeger is a keynote speaker who has spoken at business conferences in North America, South America, Europe, Africa, the Middle East and Asia.