I think the author is attacking a strawman. The common ground to most successful technologies is that their potential was recognized early on, and many people worked on realizing them, at least after breakthrough proofs-of-concept. That was the case for airplanes, televisions, computers, mobile phones, etc.
But what people "dismiss" -- although, I would say question -- is not the technology's ultimate success, but its timing. Machine learning, and the eventual AI, was Alan Turing's dream years before the first computer was built. He talked to his friend Claude Shannon about supervised learning in 1943, and Shannon proposed letting computers absorb culture and arts, too (playing them music, in particular, an idea that Turing first found surprising). In 1946, Turing wrote about unsupervised learning, talking about equipping computers with wheels, arms and cameras, and letting them roam the countryside. Neural networks were invented circa 1942, and Turing started researching them in the late 40s. The algorithms used today for machine learning were invented in the '60s, but the theory behind them pretty much stalled in the '90s.
The question is, then, not whether AI is ultimately achievable, nor whether current machine learning is useful in some domains. The question is how far is (actual) AI or generally useful machine learning. Given that we've been working on the problem for 75 years now, no major theoretical breakthroughs have been made in the past few decades, and that most successes are due to better hardware but with uncertain future scalability, I see no rational reason to expect a breakthrough in the next 5 years (very smart people in the '40s, '50s and '60s were equally convinced that AI is around the corner). I would never dismiss the promise of AI, but I would certainly question the unbridled enthusiasm some people have for machine learning's current form.
But what people "dismiss" -- although, I would say question -- is not the technology's ultimate success, but its timing. Machine learning, and the eventual AI, was Alan Turing's dream years before the first computer was built. He talked to his friend Claude Shannon about supervised learning in 1943, and Shannon proposed letting computers absorb culture and arts, too (playing them music, in particular, an idea that Turing first found surprising). In 1946, Turing wrote about unsupervised learning, talking about equipping computers with wheels, arms and cameras, and letting them roam the countryside. Neural networks were invented circa 1942, and Turing started researching them in the late 40s. The algorithms used today for machine learning were invented in the '60s, but the theory behind them pretty much stalled in the '90s.
The question is, then, not whether AI is ultimately achievable, nor whether current machine learning is useful in some domains. The question is how far is (actual) AI or generally useful machine learning. Given that we've been working on the problem for 75 years now, no major theoretical breakthroughs have been made in the past few decades, and that most successes are due to better hardware but with uncertain future scalability, I see no rational reason to expect a breakthrough in the next 5 years (very smart people in the '40s, '50s and '60s were equally convinced that AI is around the corner). I would never dismiss the promise of AI, but I would certainly question the unbridled enthusiasm some people have for machine learning's current form.