Senior editor for AI at MIT Technology Review, Will Knight wrote, “For all the remarkable progress being made in artificial intelligence, and warnings about the upheaval this might bring, the smartest computer would still struggle to make it through the eighth grade.”
Knight relates how programmers competed in an Allen Institute for Artificial Intelligence (AI2) contest. The programmers were challenged to write computer programs that could take a science test that was eighth-grade level.
During the annual Association for the Advancement of Artificial Intelligence (AAAI) meeting, the winner was announced.
Israel based programmer Chaim Linhart came in first place with the computer scoring 59% correct answers. The computer database consisted of “hundreds of thousands of questions paired with correct answers so that it could learn to come up with the right answer.”
The score is considered to be a remarkable one for any proud computer programmer. However, Knight points out that the test was “simplified” with only multi-choice answers in an effort to “make it practical for computers to attempt.”
Computer Learning Algorithms
The leaps and bounds that current computers have made are ones that use visual or audio processing. These machines have the ability to learn algorithms. Knight points to Google’s program that plays the game, “Go” as a successful use of algorithm for AI.
Knight quickly adds that while many people took this as a sign that AI (2) is on the brink of making the breakthrough needed, it’s still a ways off. He points out that AI researcher Oren Etzioni that much still needs to be developed before “basic competency at more complex tasks” can be achieved.
The Turing Test, developed in 1950 by Alan Turing (3), is the main gauge used to measure AI progress, but Knight reminds that the Turing Test is easy enough to “rig using simple tricks.”
In 1950, Alan Turing created a test that measures a machine’s ability to exhibit intelligent behavior equivalent to, or indistinguishable from, that of a human.
According to Wired, AI is already here, but just not in the form depicted in movies. AI is found in programs, such as translating languages, speech recognition, navigation and control systems for vehicles and games (4).
A prime example of this type of AI is the chess program that defeated chess player Gary Kasparov. Wired explains how the program was able to evaluate thousands of moves in its library to give it the advantage.
This type of learning algorithm gives a machine a specific skillset or skillsets to complete a task or several tasks. While it is a more sophisticated program than those seen decades past, Wired states that the machines perform by brute force with a kind of insight that “makes them look smart.”
But, Wired warns that what is commonly perceived as intelligence (usually through human conditioning to expect a machine to perform as AI) is simply implementing algorithms. Yet, the flipside is Ambient Intelligence and Augmented Intelligence thanks to “the Internet of Things and the Web of narrow components of Artificial Intelligence — programs that are crafted to a particular task and niche.”
More contests like the challenge AI2 presented to programmers are being introduced as a way to encourage and inspire programmers to create new approaches to achieving true AI. If and when this breakthrough happens remains to be seen.
References & Image Credits:
(1) Technology Review
(2) TSW: Stanford Conducts Century Long Study on Artificial Intelligence
(3) Turing Test