Business Insider pitches in to the discussion of artificial intelligence, what it will be capable of, and whether we’ll know it when we see it, by discussing futurist Ray Kurzweil’s Theory of Accelerating Returns.
The long form of Kurzweil’s theory states that “fundamental measures of information technology follow predictable and exponential trajectories.” To put it another way, computers keep growing more powerful, on an upward trend line with no visible peak, and they’re getting better faster.
In most other fields of technological endeavor, the pace of improvement slows down to become a series of minor refinements, until some massive breakthrough kickstarts the development cycle again. Not so with computer technology – it steadily improves at a pace that continues to accelerate with each passing generation. Breakthroughs come at us in a blur, occasionally slowed by manufacturers’ difficulty in finding a commercially viable way for consumers to take advantage of dramatically improved technology… a difficulty they always manage to overcome.
It’s interesting to watch consumers absorb these changes, transitioning from the old notion of “future shock” to cheerfully accepting miracles. One of the easiest ways to see what Kurzweil means is to dig up a cell phone from only two or three years ago, and compare it to your current model. Even hot phones from 2012 seem like primitive ancestors of today’s units. The capabilities of a device that has not changed significantly in size – perhaps a bigger screen with a thinner body – have increased substantially.
There are three major ramifications of these accelerating returns: machines of roughly constant size become more powerful, while machines of roughly constant power become smaller, and all of those machines get cheaper… which means there are more of them. A graph reproduced by Business Insider compares the amount of computing power you can buy for $1,000, compared to the power of a human brain:
To summarize the mind-blowing future projection of that graph, you’ll be able to buy a human brain’s worth of computing power for a thousand bucks in 2020 (which means just about everyone will have it.) You’ll be able to buy a machine with the computing power of all human minds combined for a thousand bucks by 2040.
What will a thousand dollars buy in 2060, or 2100? A demigod?
Granted, there are aspects of consciousness that don’t correlate directly with raw computing power, but the interesting thing about the state of A.I. research is that scientists are starting to get an idea of how much processing speed, parallel processing and memory capacity are needed to duplicate what a human intelligence does on a routine basis. More accurately, there have been solid theories about the kind of electronic chassis that could hold an artificial sentience for some years now, and the estimates grow more exact as the power of computers increases. Light can be seen at the end of this tunnel.
The A.I. project will no doubt be aided by computer programs – the electronic mind will be born of both man and machine, working together to construct it. Expert systems that are almost aware will help create the program that finally, ultimately is. That’s a part of the puzzle that could be conceived in the early days of computing, but not fully appreciated until programming became the powerful machine-assisted art form it is today.
There will soon be many places where an A.I. could live, many systems powerful enough to host its code plus useful applications. By way of comparison, the operating system on whatever computer you’re using to read this article would have utterly overwhelmed computers of previous generations, but now these powerful, user-friendly, feature-packed operating systems consume only modest portions of the processing power and memory capacity boasted by inexpensive computers. In a similar fashion, artificially intelligent programs that now appear just beyond the capability of the most powerful and expensive computers will one day be able to occupy consumer-priced machines, with plenty of processing and memory capacity left over for other programs it can interface with. And then that machine will shrink down to fit in your pocket, and on your wrist, and there will be millions of them…
Another of Kurzweil’s theories is the “Singularity,” sometimes described as the moment when machine intelligence surpasses humans. More precisely, the reason he chose the name Singularity is that he predicts there will be a moment when the line between human and machine intelligence is blurred, and then erased. They won’t beat us, as the doomsayers of A.I. fear; they’ll join us, and we’ll become them.
Is that day really so hard to imagine? When was the last time you heard someone say they couldn’t get along without their smartphone – a single device they rely upon for communications, memory, calculations, entertainment, information retrieval, navigation…?
If a person with a fully-functional smartphone was hidden inside a shuttered booth and presented to an audience as an all-knowing genius, they would be hard-pressed to disagree. Why, that man or woman speaking unseen from the booth can answer just about every question in a matter of moments! The illusion would only break if the necessity of fiddling with the phone’s interface made it take too long to answer questions from the crowd. But what if the resources of that Internet-capable high-speed device were directly integrated with the human mind, so that the user need only think about some passage from a Shakespeare play to have it instantly at hand, as swiftly and accurately as a scholar who had devoted his entire life to studying the literature? The illusion of super-intelligence would be complete. In fact, it would arguably no longer be an illusion.
As clever and user-friendly as our computers have become, the business of wrestling with physical interfaces – keyboards, mice, touch-screens – remains the bottleneck to unleashing their full power. To get the rest of the way there, we need the Singularity: an A.I. that can smoothly integrate with our minds, and help us refine half-formed desires. We’ll probably just talk to them at first, and they’ll be like Siri on steroids – they’ll help us figure out what we really want, suggest answers to questions we weren’t certain how to ask, and learn us well enough to know what we really mean when we give them instructions. And then we’ll decide that’s too cumbersome.
I wouldn’t bet against a large number of consumers being happy to take those next steps, especially when it can be done for less than a thousand dollars. And by “consumers,” I mean your grandchildren, if it even takes that long to get there.