When I say “artificial intelligence is right around the corner”, people frequently take issue with that statement. Responses frequently include: What do you mean by artificial intelligence? How can you know? How can we create something smarter than ourselves?
What do you mean by artificial intelligence?
In general, what I generally am referring to is a computer intelligence that is roughly as intelligent as a human being. There are different ways to compare human brain power with computer power. At the low end, the brain is thought to be roughly equivalent to 100 million MIPS. (MIPS = millions of instructions per second.) At the upper end, even the most brute force simulation of the individual neurons of the brain would require 2.8 trillion MIPS. (100 billion neurons X 7000 connections per neuron X 4000 firings/second.)
How do you know when this will occur?
We know how powerful computer processors are, and how quickly their power has been growing historically. From this, we can predict when they will be powerful enough to simulate a human brain.
The Intel Core i7 Extreme 965EE processor is rated at 76,383 MIPS. Looking at the exponential growth rate in processing power from 1996 through 2009, we find out that MIPS are growing at an increase of 1.51x per year. (From 541 MIPS in 1996 to 76,383 MIPS in 2009.)
Projecting this out into the future, we can see when computers will be powerful enough to simulate a human brain. Assuming the lower bound (100 million MIPS), a single computer will be powerful enough to simulate a human brain in 2025. A network of 100 computers will have that computing power by 2015, and a network of 10,000 computers (well within the reach of modern internet companies, for example Google has an estimated 700,000 servers) could simulate a human brain in 2010.
In the worst case scenario (requiring 2.8 trillion MIPS), a single computer could do it in 2050, a 100 computers in 2040, and 10,000 in 2030. A middle ground scenario requiring 280 billion MIPS would require human equivalent intelligence 5 years earlier, in 2045, 2035, and 2025, respectively.
Ray Kurzweil is one of the foremost thinkers on artificial intelligence. He has estimated than a single personal computer could simulate a human brain by as early as 2020.
How can we create something as smart, or smarter than ourselves?
If you look at past innovations, many of them can be predicted based solely on technological capability. For example, peer-to-peer music file sharing occurred when internet bandwidth was sufficient to make it feasible to share music. Video on demand occurred when internet bandwidth was sufficient to stream video. It didn’t require smarter than normal people to make these innovations, just the combination of the idea and the technological capability to support that idea.
Similarly, when computer processors are more powerful than the human brain, I don’t think we’ll be dependent on one particularly intelligence human being to create a program that will become an artificial intelligence. Rather, it will be an inevitable consequence of the available processing power. Around the world, millions of computer programmers with an interest in machine learning and artificial intelligence will be working on software to create artificial intelligences. Sooner or later, people will stumble on innovations to make that a reality. Those innovations, some individually, and many building on each other, will together enable programmers to create those artificial intelligences. And it is likely that they’ll emerge in many places within a short period of time, with a head start to those environments that have massive clusters of computers.
In the meantime, artificial intelligence is slowly emerging all around us. For example, the recommendation algorithms at web sites like Netflix and Amazon are excellent examples of machine intelligence that can recommend movies and products to us with all the skill of a friend that knows us well. Ten years ago that would have seemed impossible, and now it is an every day experience for most of us. Google Search is uncannily like the computer on Star Trek’s Enterprise in The Next Generation. It can tell us where people are, what they are doing, and locate arbitrary information for us. That too is a form of artificial intelligence. While these recommendation algorithms don’t “have a mind of their own”, for a person unfamiliar with computers, they might easily appear to be the output of a person assigned to a task.
What happens after an artificial intelligence is created?
While we progress at evolutionary speeds, computer processors will continue to get more and more powerful, at the rate of 51% faster each year. If the first artificial intelligence is created in 2015, for example, by just two years later, it will be able to think twice as fast as a human. By five years later, it will be eight times faster than a human, and by ten years later, it will be more than fifty times faster than a human. It’s hard to say exactly how intelligence compares to speed of thought, but this artificial intelligence will be able to consider outcomes to actions, alternate ideas, alternate actions, and evaluate possibilities of all kinds fifty times faster than a human. This will manifest as an intelligence that we will not be able to keep up with.
Ray Kurzweil calls this the technology singularity. It’s a singularity in the sense that it is an event horizon beyond which we cannot see. We do not have the intelligence to foresee what might happen, and the pace of events will be so rapid the we cannot anticipate the pace or scope of those changes. For example, if the first artificial intelligence occurs in 2020, by 2030 there could be billions of artificial intelligences each a hundred times more intelligence that people. By 2040, there could be hundreds of billions of artificial intelligences each thousands of times more intelligence than humans. What further advances in technology, science, healthcare, business, and politics might result from such a huge mass of brain power?