Avoiding another AI winter

A while back I felt like I have been put to sleep for a decade or so and I have missed out on innovations in AI (artificial intelligence) — innovations that long-time AI enthusiasts like me have always been dreaming of. This feeling was triggered by a barrage of articles in the mainstream media — articles heralding the coming of AI with bold headlines that would make “The Soviets are coming” during the height of the Cold War fade in comparison: “Computer chip thinks like a human being“, “Computer brains that reason like humans“, and “A radical new AI that does anything you ask“. The hype was (and still is!) so extreme that it has prompted the likes of Bill Gates and Professor Stephen Hawking to speak of the potential dangers of super intelligent machines!

After some reflection, I soon realized however that I have not been put to sleep for several years — we are still going at it, and Gates and Hawking can sleep well. False alarm, folks — all is well (or not, if you do like to see real progress in AI!)

Apparently, this media frenzy was due to recent advances in pattern (speech and image) recognition and IBM Watson’s win in Jeopardy! This relative advance in state-of-the-art was in turn due to three factors: (i) the plethora and wide accessibility of data; (ii) the tremendous increase in computational power to process such (big!) data; and (iii) advances in data analytics and machine learning algorithms (algorithms that find regularities and patterns in data). But using thousands of processors and training on millions of examples for several days to (barely) recognize objects in images — something the most primitive of species do well — is hardly an indication that we are about to achieve Artificial Intelligence.

AI is not about advances in computing, processing power, or data analysis. AI is about the development of thinking machines, and we do not, as of yet, have a computer program that understands a simple English sentence that a 5-year old effortlessly comprehends. As an AI enthusiast I welcome any advance in state-of-the-art, but like many of us that lived through the AI Winter of the 1980s, which was brought about by overselling, overpromising, and overhyping simple advances, we fear that the current misguided hype is detrimental to the field of AI and might even cause another (and longer) AI Winter.

Some sane voices are starting to speak out, however, and there are good signs that this important subject will not be left to hackers and over-excited and misinformed journalists.

In this column I hope to positively contribute to this discussion, reporting on real progress in AI, the state-of-the-art, and challenges that still lie ahead.