Another big chill?

Reading time: 2 mins


Writing in the MIT Technology Review in 2017 (Is AI Riding a One-Trick Pony?), James Somers argued that the recent surge in AI developments may have reached a plateau.

If so, this would be one of many technological plateaux experienced since the dawn of AI.  These periods of stasis have often preceded the so-called “winters” or “dark ages” in AI’s history.  During an AI winter, research funding begins to dry up and the media, temporarily mobilised by the expectation of daily revelations, take down their tents and withdraw their reporting teams.

Are we facing another AI winter?

The most recent surge, according to Somers, was the result of breakthroughs in machine learning during the late 1980s, most notably the refinement of backpropagation methods leading to dramatic improvements in real-world applications, such as image recognition.

Of course, high performance neural networks trained on terabytes of data are, without doubt, incredibly useful, but they remain woefully short of anything we would recognise as intelligence.  Hence, backpropagation may have taken us to a new level, but clearly nowhere near the summit.

Breakthroughs in AI often take decades to evolve, from conception through to occasional headline-grabbing demonstrations.  Consequently, the next AI surge may be 20-30 years away.  It's a sobering thought.

Those of us not involved in research must therefore focus our energies on the tools already developed, maximising their efficiency and utility for end users.  Even with our current knowledge of deep learning, there is enough work for a lifetime of product innovation and development.

Meanwhile, those who are waiting for the next great leap forward may find themselves shivering through another AI winter.  If that’s you, perhaps the best advice is to wrap up well.



Writer: PJ Moar of Moar Partnerships
Email: p.moar@moar.com
Twitter: @MoarPart