Artificial Intelligence Defeats Visual Turing Test

            Artificial intelligence is catching up to humans. A new, military-funded artificial intelligence can generate new ideas as quickly and accurately as a human can. Brenden M. Lake, Ruslan Salakhutdinov, and Joshua B. Tenenbaum presented the Bayesian Program Learning framework, or BPL for short, in a paper published by the journal Science.

BPL is said to classify objects and generate ideas using just a single data point. To put BPL to the test, they showed BPL along with several participants 20 handwritten characters from 10 different alphabets. They then are asked to match the letter to the same character written by someone else. BPL scored a 97%, about the same as the other human participants, not to mention much better than previous algorithms. What’s even more amazing is that BPL also passed a visual form of the Turing Test. BPL was asked to draw letters that most humans couldn’t distinguish from another human’s writing.

An early form of this kind of AI dates back to 1997, when Gail A. Carpenter and Stephen Grossberg published a paper on adaptive resonance theory, or ART. Their model was based on a human’s neural network and, while learning quickly, was limited to simple pattern recognition. BPL, on the other hand, uses actual reasoning to infer why the data is the way it is.

If the reports are true, this could be a huge breakthrough in military surveillance, as indicated by the numerous military donors to the project such as the Air Force Office of Scientific Research, the Office of Naval Research, DARPA (Defense Advanced Research Projects Agency), and IARPA (Intelligence Advanced Research Agency).

Pentagon leaders have been discussing the use of this kind of Artificial Intelligence in surveillance missions, where teams might fly for over 6,000 hours before deciding to strike a specific target. Of course the BPL is not in condition yet to fully replace pilots. “A lot of things will have to be done that involve expertise that none of us have” says Tenenbaum. Advances in this AI could also lead to far more capable drones, faster intelligence collection, and more accurate targeting through AI.

Erik Byrnjolfsson and Andrew McAfee predicted that artificial intelligence would soon reach our levels, and they appear to on the right track. Not long ago, AI couldn’t even pass the visual Turing Test, with the percentages ranging from 67% accuracy to 92%, more than double the error of both humans and the BPL. Now that an AI can pass the Turing Test, only time will tell what next breakthrough will be.