Heart

Menu

Stay in the loop

Never miss a beat with the latest in VET news delivered straight to your inbox.

Stay in the loop

Never miss a beat with the latest in VET news delivered straight to your inbox.

Is There an AI Winter on the Way?

June 20, 2018

The year is 2018 and we have self-driving cars, smart home assistants, and virtual reality. Despite this, experts seem to disagree about whether AI is on its way in, or out. We’ve seen some impressive developments in recent years, but is AI actually declining?

On May 29th, neuroscience scholar Filip Piekniewski made public his opinion on the future of AI. In his blog post AI Winter Is Well On Its Way, Piekniewski purported that the popularity of AI is declining, his main arguments being:

  • Deep learning is not mature enough to replace some existing occupations, like radiologists, especially not in the way that Andrew Ng has claimed.
  • Deep learning models don’t scale up with time. Their parameters have remained pretty much the same as they were a few years ago.
  • In terms of self-driving car accidents, the bottleneck performance of autonomous car training is essentially unsolvable with current science.
Image result for yann lecun
Yann Lecun

On the other hand, former Facebook AI chief Yann Lecun has refuted the article, claiming that “the blog post is very uninformed”. Lecun enumerated a few facts about how big tech companies like Google and Facebook were investing heavily in the field of AI at “an accelerated pace”.

But what about AI for academic research and business application? AI research companies like DeepMind and OpenAI are pushing the forefront of AI as a scientific discipline when it comes to understanding animal intelligence. However, when it comes to the business aspect of trying to tap into knowledge without big-volume, standardised data lying in the database, it usually fails miserably—emitting unstable black-box results.

This breach of performance is not at all surprising, since real-world data is far from the clean-cut data of the research environment. No matter how experienced the model trainers are, there will always be some aspect of the real world which is not covered by the existing scientific discipline, which will sooner or later cause the model to malfunction.