Hacker News Reader

Comment by seanmcdirmid | Hacker News Reader

It would be ironic if the fourth AI winter came about because China over invested in AI. China's real estate boom has already hit peak Japan.

View on HN

I'm certain that AI business will plateau and there will be yet another AI winter without new big breakthroughs in fundamental research.

We are currently living in a era where the results from Hinton & Canadian Mafia are still begin adopted and refined. It's just alternating layers of affine transformations and nonlinearity with lots of tricks and improved routing.

There needs to be fundamental theoretical breakthrough in every decade or so, and not just refinements and new applications. Hinton made his "What is wrong with convolutional neural nets" speech years ago "https://www.youtube.com/watch?v=rTawFwUvnLE I believe that Hinton's capsule networks type attempts are what we need more. They may not deliver immediately impressive results because fundamental research involves lots of setbacks.

I think we have a lot to learn from the human brain still. I am continuosly fascinated by our ability to learn quickly and use previous experience to bootstrap learning. I think the next breakthrough has a high chance of being a collaboration between the neuroscience community and CS. What is really interesting about Hinton is his background, experimental pyschology.

No one can be certain, unless you are one of those that keep saying a company will go down eventually, one day you will be right.

We can avoid AI investment boom and bust cycle only if:

1) hype and expectations <= reality constantly

and

2) R&D time horizon <= investor time horizon constantly

Has that ever been the case? Technology advances but there are always periods of over-investment and lost fortunes. Dot-com bubble didn't burst because internet was a fad, it bursted because expectations and reality did not sync. It took over a decade for Amazon stock to return it's previous level.

There has already been two AI winters one in early 70's and another in late 80's early 90's.

With all the funding into AI meaning we have way more human capital involved and with real tangible AI benefits being demonstrated, it is unlikely AI winter will arrive anytime soon, I say revisit again in 5 years

> It's just alternating layers of affine transformations and nonlinearity with lots of tricks and improved routing.

and a computation is just 0s and 1s, with lots of if/then statements.

Not necessarily. Lambda calculus is Turing-complete under Church-Turing thesis, and it has nothing to do with 0s and 1s. That’s just von-Neumann architecture encoding.

youve missed the point.

the point is: anything complex can be dismissed as "just x,y,z" if you dont appreciate the massive body of work behind it.

i made that point because OP observed that "ml is just affine transformations" or something to that effect. yes, that's one way to frame it - if youre okay overlooking roughly 30 years of research.

I have to disagree - it could be the case that the algorithms are sufficient and all we need is more data and more compute.

I'd strongly recommend checking out Peter Norvig's talk on "The Unreasonable Effectiveness of Big Data"

There will be a shakeout - but the Deepnet boom is useful because there are several other significant technologies that have been produced since the late 80's which are ready for mainstream adoption: answer sets, tractable bayesian reasoners (MCMC), SAT solvers.. and one of the technologies that enabled deep nets has enabled these massively : big data. And now streaming systems are mainstream they are enabling for another raft of applications using AI.

If we think about possible real world applications of DL and related techniques, currently at most 5-10% of possible applications are introduced. We have many years of work ahead until the current technologies will achieve their benefits in real world applications.

Reply on news.ycombinator.com