Will computers overtake humans in intellectual ability?
In 1997, world chess champion Garry Kasparov played a series of six games in a rematch against the supercomputer Deep Blue. The year before, Kasparov defeated Deep Blue, winning three games to one (with two draws) against the supercomputer. During the rematch, Kasparov won the first game but then began to struggle. When it was all over, Deep Blue claimed the victory with two wins against one with three draws.
In 2011, IBM's computer Watson competed against Ken Jennings and Brad Rutter on the "Jeopardy!" game show. Backed by more than 2,800 processor cores and a library's worth of information, the computer defeated the two human opponents. Watson showed that with the right processing power and programming, a computer could learn to interpret natural language and respond appropriately. It marked a leap in the field of artificial intelligence. Today, IBM is turning Watson's power to solving problems in the medical field and beyond.
Were these losses signs that computers had become smarter than people? It's true that computers can perform calculations at a blistering pace. The Sequoia supercomputer can perform 16.32 quadrillion floating operations (or petaflops) every second [source: Top 500]. How does that compare to the gray matter that's in our heads?
Measuring how fast humans can think isn't easy. Creative estimations are the best we can manage. Using visual processing as a starting point, robotics expert Hans Moravec of Carnegie Mellon institute estimated that humans can process about 100 trillion instructions per second (or teraflops) [source: Moravec]. But Chris Westbury, associate professor at the University of Alberta, estimates the brain may be capable of 20 million billion calculations per second, or around 20 petaflops [source: Westbury]. Westbury bases this estimation on the number of neurons in an average brain and how quickly they can send signals to one another. What's clear is that computer processing speed is at least approaching, if not outpacing, human thought. But are computers smarter?
Currently, computers fall short of possessing intelligence. But will that always be the case?
Computers and the Human Brain
There's more to intelligence than processing speed. While a supercomputer like the Sequoia can analyze problems and reach a solution faster than humans, it can't adapt and learn the way humans can. Our brains are capable of analyzing new and unfamiliar situations in a way that computers can't. We can draw upon our past experiences and make inferences about the new situation. We can experiment with different approaches until we find the best way to move forward. Computers aren't capable of doing that -- you have to tell a computer what to do.
Humans are also very good at recognizing patterns. While we're making progress in machine pattern recognition, it's mostly on a superficial level. For example, some digital cameras can recognize specific faces and automatically tag photos of those people as you take pictures. But humans can recognize complex patterns and adapt to them -- computers still have trouble doing that.
Could computer scientists build a machine that simulates the way humans think? It's not as easy as it sounds. The human brain is incredibly complex. We still don't have a full understanding of how the brain works. Without this understanding, it's challenging to create a meaningful simulation of the brain.
One of the features computers would need to be more intelligent than humans is the ability to draw conclusions from observations. In a study published in 2009, computer engineers at Cornell University designed a program that could do this on a limited scale. The program gave the computer a basic set of tools it could use to observe and analyze the movements of a pendulum. Using this foundation, the software was able to extrapolate basic laws of physics from the pendulum's motions. It took about a day for the computer to arrive at the same conclusions it took humans thousands of years to grasp [source: Steele].
While the Cornell project was a remarkable achievement in computer engineering, we're still years away from computers that can make conclusions from general observations. The Cornell software gave the computer the tools it needed to draw conclusions -- the computer was unable to create or refine these tools for itself.
As long as computers rely on sets of pre-installed instructions to perform tasks, they can't be said to be more intelligent than humans. Even IBM's Watson can only respond to input -- it can't spontaneously pull up information or think in the way we humans do. Only when computers can adapt and perform tasks outside their initial programming will they be truly intelligent. Until that time, computers are just very sophisticated calculators.
Hundreds of computer scientists are trying to solve this crucial problem. Some are attempting to design computers that can mimic human thought -- a tricky situation considering we still don't have a complete understanding of how we think. Other computer scientists prefer to design systems that don't use the brain as a model. Futurists like Dr. Ray Kurzweil predict that it's just a matter of time before we develop a computer system capable of being self-aware. After that, we may see computers capable of recursive self-improvement. That means computers will be able to analyze their own capabilities and make adjustments to improve performance.
But creating a self-aware computer system is beyond our capabilities right now. It may not even be possible. While we continue to gain understanding in the fields of biology and computer science, we may encounter a fundamental obstacle before we can ever create a self-aware machine. Alternatively, we may come to a point where human and machine intelligence merge, leaving the entire question moot.
Learn more about artificial intelligence and computer science by following the links on the next page.
Author's Note
I'm torn on the subject of artificial intelligence. On the one hand, we live in an age of technological innovation and we're seeing enormous leaps in computational ability every year. On the other, thinking is hard. I suspect we'll have the hardware capable of supporting thought before we crack the software side of the problem. It's probably only a matter of time before machines are actually thinking in a way that's analogous to our own processes. But it's still hard for me to imagine.