“What is artifacial inteligence?” you ask Google. To which it replies, “Did you mean artificial intelligence?” Of course you did. Meanwhile, in the 0.15 seconds it took you to realize your own stupidity, an intelligent machine has assembled 17,900,000 results for your consideration – including video, audio, historical records and the latest headlines – ordered by relevance and reliability. 20 years ago, this type of artificial intelligence would have been the stuff of science fiction, but now we simply call it ‘technology’.
Artificial intelligence began over 60 years ago as a philosophical question posed by the brilliant English mathematician Alan Turing: “Can machines think?” In 1955, the words ‘artificial intelligence’ first appeared in print in a proposal for a summer academic conference to study the hypothesis that “every aspect of learning or other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it”.
At its core, the science of AI is the quest to understand the very mechanisms of intelligence. Intelligence in humans or machines can be defined as the ability to solve problems and achieve goals. Computers, it turns out, are the ideal machines for the study of AI, because they are highly ‘teachable’. For half a century, researchers have studied cognitive psychology – how humans think – and attempted to write distinct mathematical formulas, or algorithms, that mimic the logical mechanisms of human intelligence.
Machines have proven extraordinarily intelligent, with highly logical problems requiring huge numbers of calculations. Consider Deep Blue, the chess-playing computer from IBM that beat grandmaster Gary Kasparov using its brute processing strength to calculate a nearly infinite number of possible moves and countermoves.
Alternatively, consider the everyday examples of astonishing AI, like the GPS navigation systems that come standard in many new cars. Speak the address of your destination and the on-board computer will interpret your voice, locate your precise location on the globe and give you detailed directions from Moscow to Madrid. Or even something as ‘simple’ as the spell check on your word processor, casually fixing your typos as you go.
And then there are AI machines that go far beyond the everyday, like robots. Today’s most extraordinary robotic machines are much more than logically intelligent; they’re also physically intelligent. Consider Stanley, the 100% autonomous vehicle that barrelled through the Mojave Desert to win the 2005 DARPA Grand Challenge. Stanley used GPS data to pinpoint its location, as well as laser-guided radar and video cameras to scan the distance for obstacles in real-time. Internal gyroscopes and inertial sensors feed constant streams of data into the on-board computer to control steering and acceleration.
The Honda ASIMO (Advanced Step in Innovative Mobility) robot grabbed the world’s attention with its human-like walk, a feat of intelligent engineering. ASIMO uses infrared and ultrasonic sensors to gauge distances from floors, walls and moving objects, and constantly adjusts its balance and motion with 34 high-precision servo motors. ASIMO’s processors are so lightning-fast, you can shove the robot sideways in mid-stride and it will “instinctively’ throw its weight onto an outside foot to right itself.
Perhaps the greatest achievements of artificial intelligence over the past half-century have been illustrated by the way that machines can intelligently process information. Google is just one example of intelligent information technology that can parse obscene amounts of data into useful information. Intelligent cell phone networks bounce packets of voice data along the most efficient path. Logistics software is the engine of global business, calculating the most efficient and profitable way to procure supplies, manufacturer and ship products around the world. Credit card companies use intelligent software to analyze the buying patterns of millions of cardholders and identify the subtle red flags that signal fraud or theft. In the information age, we rely on these intelligent machines to make sense of streams of seemingly random data.
As processing power continues to multiply, we are coming closer to answering Turing’s original question: “Can machines think?” We are teaching machines to rely less on pure logic and more on probabilities and experience, what we might call “intuition’. And they are fast learners…
AI and robotics
The lifelike androids designed by Hiroshi Ishiguro at the Intelligent Robots Laboratory use real-time facial recognition software to mimic the facial movements of the ‘controller’.
Walking robots like ASIMO are equipped with an internal gyroscope and speed sensor to help it maintain balance, even when shoved. Infrared and ultrasonic sensors are used to gauge the distance of the floor and the speed and path of approaching objects. Sensors in hands and feet help it ‘feel’ the six axes of force – up/down, left/right, forwards/ backwards – and the degree of force applied.
What’s IBM’s Watson?
In February 2011, an IBM supercomputer named Watson trounced two previous champions of the US trivia quiz show Jeopardy!. Watson parsed natural language questions fast enough to beat the quickest human minds. IBM researchers preloaded the computer with hundreds of millions of pages of data, then armed it with algorithms for searching and ‘reading’ text – separating subjects, verbs and objects. But this was much more than a super-powered Google search. Watson used advanced algorithms to ‘reason’ which of its millions of hypothetical answers was most likely to be true. The ‘face’ behind the Jeopardy! podium was backed by a roomful of servers, comparing results in fractions of a second until the computer had enough statistical confidence to buzz in. Watson technology is already being considered as the brains behind an automated physician’s assistant.