Artificial intelligence ranks as one of the most common tropes in science fiction, but a relatively new approach involving computer systems designed to mimic the functions of the human brain is having success solving problems that older brute-force computing methods haven’t been able to crack.
On Financial Sense, Matthew Bey, science and technology analyst at Stratfor, discusses the changing paradigm and how AI is advancing in interesting new ways.
“In the last month we’ve seen AlphaGo, which is Google’s platform to play the game Go, play Lee Sedol, one of the biggest and most successful Go players in the world, and beat him four games out of five,” Bey said.
Go masters rely more on intuitive play, Bey noted, unlike Chess masters, who often think anywhere from 15 to 25 moves in advance. Go is an ancient Chinese game that is more complicated than Chess, and features so many possible outcomes that it’s effectively impossible to solve … until now.
Google wrote AlphaGo using advanced neural networks as a way of unraveling the computing challenges Go poses, Bey noted. The program’s approach is based more on human-like perception as opposed to computational brute force to come to an answer.
Google is at the forefront of this kind of technology, and has also recently developed robots that use similar technologies for perceiving objects in a room and moving those objects out of the way, Bey said.
“This is something that’s pretty revolutionary when you talk about the potential impacts for artificial intelligence in the work environment and other places,” he added.
Brain-based AI
Unlike traditional computer architecture emphasizing a large central processing node, advanced neural networks are modeled on the human brain, Bey said. With potentially thousands of processors networked together, it’s possible to do a lot of computations at once, and it also allows for communication between these nodes so they can do a lot of parallel computing, he added.
These networks aren’t going to be good at solving every problem, Bey noted, but they will be useful for applications that benefit from a more “bottom-up” computing approach, as opposed to traditional top-down models. Natural speech recognition is one such area where ANNs may be extremely useful.
“When we read sentences or words, if you take away the vowels, humans can still recognize what that’s saying without having all the letters in a row,” Bey said. “I think it’s that same principle of pattern recognition rather than brute-forcing … a probabilistic interpretation is definitely the way to look at it.”
Must-read: The Neural Network We Call the Market
These computing algorithms are also very good for learning, he said. By taking feedback over time from a lot of examples, these networks are better at adjusting output and behavior to solve problems and learn patterns more quickly.
Ultimately, this means ANNs can effectively program themselves, Bey said. These networks will likely be large, but that means some companies are better suited to take advantage.
“When we talk about speech recognition, I can’t necessarily have that same kind of setup on my phone,” Bey said. “But I can actually have it being processed somewhere around the globe ... This has served as a big incentive to build these kinds of computers because you can now use it on a much bigger platform. And that’s something Google and a few other companies are well positioned to actually use.”
Another company at the forefront of ANN research is Baidu in China, Bey said. They have developed software to predict riots and other forms of social protest, he said, which raises questions about how countries such as China might deploy this kind of technology.
Potential Market Applications
With the ability to model complex systems, ANNs may one day influence how we understand markets.
“When we look at something as complex as the modern economy, and modern trading platforms … we’re not to the point now, and we probably never get to the point, where we can fully understand (them),” Bey said.
However, a bottom-up approach using a neural network may allow much more sophisticated modeling, he added. This could even extend to the regulatory environment, with ANNs being used to detect trades made using insider information, for example.
Though companies like Google in the U.S. and Baidu in China are leading the way, other companies are likely to get on board when they see the potential applications, Bey said.
“The advantage that a Google, or a Baidu, or even maybe an Amazon has, is the volume of data they can process,” he noted. “They can use that data to achieve whatever end application they want to.”
From an investor standpoint, Bey feels it will make the most sense to focus on Silicon Valley-type tech companies. But the future for AI research may have wide-ranging impacts and could change market conditions considerably going forward.
Listen to this full 20-minute podcast interview with Stratfor's Matthew Bey by logging in and clicking here. Not a subscriber? Click here.
