[This post has already been read 651 times!]
Facing down another ‘AI winter’
Lucubrate Magazine, Issue 32, July 13th, 2018
By Janosch Delcker
The prospect of another “AI winter” is haunting the artificial intelligence community. Some are convinced it’s coming. But before you rush to get your winter coat, keep reading.
Artificial Intelligence (AI) is any task performed by a program or a machine that, if a human carried out the same activity, we would say the human had to apply intelligence to accomplish the task. AI systems will typically demonstrate at least some of the following behaviors associated with human intelligence: planning, learning, reasoning, problem-solving, knowledge representation, perception, motion, and manipulation and, to a lesser extent, social intelligence and creativity.
Wait, what? An “AI winter”? If the term invokes the Cold War, that’s probably because it was coined back then.
Research on AI started in the late 1950s, and some ambitious researchers predicted soon after that machines would “be capable within 20 years of doing any work a man can do.” In the years that followed, however, technology failed to deliver on those promises and the bubble burst in the late 1970s, leading to a major hiatus and the public losing interest. It’s what’s now known as the first “AI winter.” A similar crisis followed in the 1990s. And now we’re back in an era of hype, which makes some experts believe the next winter is looming. But is it?
Not in the next couple of years, the University of Delft’s Virginia Dignum, who is a member of the EU’s high-level expert group on AI, told me. Yes, a buzz surrounding the field is back, she said — but this time around, there are enough successful real-life AI applications that the bubble is less likely to burst.
That’s the great difference to previous “AI winters.” Thanks to recent advances in technology, particularly in computer power and cloud storage, AI developers are now able to turn more cutting-edge theory into real-life applications.
“If we keep focusing only on ‘deep learning,’ I can see that at some point, people will be disappointed because there are problems you can’t solve with it,”
Deep Learning is an algorithm which has no theoretical limitations of what it can learn; the more data you give and the more computational time you provide, the better it is (Geoffrey Hinton)
Deep Learning with AI
But that doesn’t mean we won’t experience some sort of bitter awakening soon. At the core of much of today’s AI sits a machine-learning technique called “deep learning,” which essentially works by classifying data. It finds patterns in a pile of information, which reveal new correlations. Problem is, however, that those correlations explain a lot but not everything: A cock may crow at dawn every morning, but that doesn’t mean it’s the cock who causes the sun to rise.
Think Beyond Deep Learning
What then are the alternatives AI researchers should also embrace? One of them could be working with models based on causality rather than correlation, she told me. Another one could be models that try to minimize the dependency on existing data.
No one knows for sure whether or not another AI winter is looming. But the vast majority of AI experts I talk to don’t think so. What they agree on, however, is that it’s time to start thinking beyond deep learning.
Original source: POLITICO
Published on 3 July 2018