Artificial intelligence (AI) is making huge strides in problems that have traditionally plagued the field by harnessing the power of machine learning, a statistical method that relies on a lot of data.
Machine learning superficially mimics the interconnected structure of the human brain, but beyond that, it looks nothing like how humans think or reason.
But when Tech Insider asked 26 AI researchers whether AI has to mimic the human brain to be truly intelligent, many of them flat out said no. Many of them believe that trying to recreate how humans achieve intelligence isn't fruitful, and the best bet for creating intelligent systems is with machine learning.
AI was first established in 1956 as a research field to study the nature of intelligence by building it. Early commonplace approaches to build thinking machines included attempts to encode knowledge, logic, and reasoning, according to the American Scientist. A lot of the work was done in "small, proof of concept" projects that delivered few results.
By the 1980s, AI changed course, according to the Atlantic, and shifted to tackling intelligence piecemeal, like by building programs to solve specific problems.
That's how it remains today. Because machine learning is delivering real results, companies like Google and Facebook are doubling down on implementing even more machine learning in their services.
But as the approach to intelligence has changed, so has the definition of intelligence. According to Thomas Dietterich, director of Intelligent Systems at Oregon State University, intelligence isn't a single trait — it "refers to many things." Because intelligence is measured "by how well a person or computer can perform a task," then really, it's the results that really matter, not how it's done.
"By this measure, computers are already more intelligent than humans on many tasks, including remembering things, doing arithmetic, doing calculus, trading stocks, landing aircraft, etc," Dietterich told Tech Insider.
In fact, many of the researchers Tech Insider spoke to compared aircraft and birds to illustrate the point.
"I think that's one of the few things that most of us will agree on," Oren Etzioni, the CEO of the Allen Institute of Artificial Intelligence, told Tech Insider. "The best analogy is with flight. Planes are very different from birds. Birds flap their wings and they're very light. Airplanes are heavy and they're different, the principles of aerodynamics are the same but these are very different mechanisms. I think most of us believe that's going to be the way with intelligence as well."
But the recent advances aren't really intelligent, according to Douglas Hoftadter, author of the so-called "bible of AI." He believes that in order to built machines with human-like intelligence, with all its rich nuances, you have to build machines that think like humans. The debate has gotten so contentious Hoftstadter has basically shunned the rest of the field.
"I don't want to be involved in passing off some fancy program's behavior for intelligence when I know that it has nothing to do with intelligence," Hofstadter said to the Atlantic when asked about Deep Blue, the IBM supercomputer that defeated the reigning chess champion Garry Kasparov in 1997. "I don't know why more people aren't that way."
Detractors of machine learning, like Hofstadter, say the approach misses the point of AI entirely, and will never truly achieve human-like intelligence.
Scott Phoenix, the cofounder of Vicarious, is trying to build the world's first human-level AI. He told Tech Insider that we have to look to examples of intelligence that occur in nature if we truly want to understand intelligence. That doesn't necessarily mean completely copying the biological structures of the brains of intelligent beings, that would be as absurd as planes with feathers and beaks, Phoenix said.
"It's probably going to be a lot more difficult to ignore neuroscience and ignore the brain when you're trying to build something that works like a brain," Phoenix told Tech Insider. "At the same time I don't think it's strictly necessary that you duplicate all of the biological functions of the brain."
But Bart Selman, a computer scientist at Cornell, said building an intelligent machine based on human intelligence would be impossible because we simply haven't figure out how humans do it. The best hope we have is trying to "get to a performance at a human level without getting the details of the human brain all figured out."
"Speech recognition is a good example," Selman told Tech Insider. "We don't quite know how the brain does it, the brain is probably more complicated than the way we're doing it right now."
For their part, machine learning advocates admit their method doesn't try to emulate human intelligence. The IBM supercomputer Watson, which beat two Jeopardy champions — relies on statistical methods and a lot of data. But even Dave Ferrucci, the team leader on Watson, admits that Watson doesn't have anything to do with human intelligence.
"Did we sit down when we built Watson and try to model human cognition?" Ferruci told the Atlantic. "Absolutely not. We just tried to create a machine that could win at Jeopardy."
Ferrucci told the Atlantic that despite what detractors say, there's no reason why machine learning would have to act anything like human intelligence.
"It's artificial intelligence, right? Which is almost to say not-human intelligence. " Ferrucci said. "Why would you expect the science of artificial intelligence to produce human intelligence."
Join the conversation about this story »
NOW WATCH: An exercise scientist told us how long it takes before working out gets easier