Is “Artificial Intelligence” Dead? Long Live Deep Learning?!?
Has Deep Learning become synonymous with Artificial Intelligence? Read a discussion on the topic fuelled by the opinions of 7 participating experts, and gain some additional insight into the future of research and technology.
Our next quote on the subject comes from University of Sheffield professor of machine learning Neil Lawrence. Neil leads the ML@SITraN group, is involved in research in probabilistic models with applications in computational biology and personalized health, and blogs regularly at inverseprobability.com.
Neil presents a few points particularly relevant to why conflation of these terms is so commonplace, even among technologists. The ever-evolving concept of "artificial intelligence" is central.
There are a few different aspects to the question, in particular. What does AI mean to different sectors of society, the public, technologists and researchers. I think it's definitely true that the terms are being conflated a little for the public and technologists, but this is likely to be a temporary phenomenon. For researchers, there'd be necessarily more sensitivities to what these terms do mean.
One thing that is clear is that the term Artificial Intelligence means different things to different people. I've said in the past that many debates on AI would be more coherent if the A stood for Anthropomorphic intelligence, because that seems to be the sense in which many people understand AI, emulation of human intelligence. This means that there'll naturally be a conflation when a new technology has a breakthrough in emulation of human abilities, just as deep learning has. However, just as people came to understand that Deep Blue wasn't equivalent to anthropomorphic intelligence, they'll realise the same for deep learning.
Another sense in which the term is used relates to artificial intelligence technologies, the group of techniques that people are using to solve challenges in language, speech, computer vision. Again, deep neural networks have made significant inroads here, so there is currently a conflation from the technologists perspective. I believe this will also be temporary, once these techniques are more commonplace and their limitations are better understood, the domain of what is considered an AI technology will evolve and new classes of techniques will be needed.
Next up is Xavier Amatriain. Xavier leads Engineering at Quora, where he builds machine learning algorithms, and formerly held the position of Director of Netflix Recommendation Algorithms. Xavier's insight reminds us that complexity is an important factor, and that its artificial increase is never a worthy goal.
Deep learning has accomplished some impressive feats lately. However, deep learning is not the only, and not even the most effective, approach to machine learning and therefore AI. There are many other approaches to learning from data. As a matter of fact, most other approaches are usually simpler, easier to understand, and more efficient in terms of computing cycles. By defaulting to deep learning whenever thinking about machine learning or AI, we run the risk of over-complicating our current solution and compromising future innovation and improvements. That said, if you are working with images, speech, text, or any similar high-dimensional signal, you want to keep deep learning in your toolkit.
Xavier is correct that deep learning can be needlessly complex for a given scenario. He is also correct in stating that it is a great addition to your machine learning toolkit, when ready to be dispatched in appropriate situations.
Gregory Piatetsky-Shapiro, President of KDnuggets, contributed the following, which nicely summarizes the comparison between machine learning, deep learning, and artificial intelligence, and identifies another dimension to the confusion between AI and deep learning.
AI is a much larger field than its subfield of Machine Learning, and it includes other topics like Knowledge Representation and Reasoning, Robotics, and Intelligent Agents.
Machine Learning, in turn, has much larger scope than its subfield of Deep (Neural Networks) Learning, since it also includes other approaches such as Support Vector Machines, Bayesian networks, and Reinforcement Learning.
However, as sometimes happens in life, the child has exceeded the parent and was in turn exceeded by grandchild.
Deep Learning recent and amazing successes - especially in image understanding, speech recognition, and game playing - where DL reached and sometimes exceeded the best human ability, have captured the imagination of the popular media and venture capitalists, so much so, than Deep Learning has become identified with AI.
If AI can be compared to a car, then Machine Learning is its engine, and Deep Learning is just one type of engine. Many new components need to be added, including probably multiple engines, before general purpose AI will be reached.
If there had been disagreement between those with expertise, perhaps further debate would have been warranted. However, while the specifics and the emphasis between the experts may differ, the overwhelming message is clear: deep learning is not synonymous with artificial intelligence. While deep learning can (and is) employed in artificial intelligence pursuits, AI also encompasses a number of other technologies (reinforcement learning, Monte Carlo tree search, genetic algorithms, etc.). Likewise, deep learning is not employed solely in the pursuit of a conservative definition of artificial intelligence. The fact that these terms can mean somewhat different things to different people does not negate these ground truths.
Finally, we will end on the insight provided by data scientist and machine learning consultant Charles Martin, who puts the deep learning tipping point into context, and reminds us to stay excited about current and future breakthroughs.
Deep Learning is at a tipping point, where internet technology was in the late 90s. More than help enterprises sift through their data, I think we are going to see whole new products and industries sprout up in the next few years that can take advantage of this amazing technology. It is exciting to be here, to be part of it as it is happening.
I would like to thank everyone who has taken the time to contribute to this article.
- The Data Science Puzzle, Explained
- Common Sense in Artificial Intelligence… by 2026?
- 3 Thoughts on Why Deep Learning Works So Well