For example, automakers are shipping all sorts of fancy driver assistance systems, and a lot of that is driven by fancy image recognition and deep learning systems.

Looking beyond, there's the constantly-discussed self-driving car. It's not here yet, and it's going to take a ton of work, but at this point it looks like there's nothing fundamental standing in the way. It's no longer a question of whether it will happen, but whether it's happening in five, ten, or fifteen years.

Outside of transportation, businesses are applying AI techniques to analytics to squeeze ever more money out of their client base. There's the infamous story about how Target knew a teenage girl was pregnant before her father did. AT&T just sent me a letter saying they're giving me an extra 5GB/month bonus data as long as I keep my plan, which I assume happened because their data indicated I was a risk for downgrading or switching carriers. Improving these systems is worth a ton of money, and as far as I understand it fits really well with where current research advancements are being made.

As long as there are clear, immediate, profitable uses for AI advancements, there shouldn't be any worry of another winter. I guess the question is whether current advancements will peter out eventually, or whether we've reached a point where more research will always produce more immediate dividends. It sure feels like the latter to me, but I could easily be wrong.

Big difference? Do you seriously think that the entire field of artificial intelligence research produced nothing usable until today? For as long as there were AI algorithms there were also attempts to commercialize them (with varying success). The problem wasn't that all of those algorithms turned out to be useless, but rather the mismatch between hype and reality.

You should look into the history of expert systems for a classic example. With things being as they are, deep neural networks might very well go though the same cycle, which is not a good thing for the AI field in the long run.

Caution and skepticism are not the enemies of research and engineering. Hype is.

Is there a mismatch between the hype and the reality for the people funding this stuff? Certainly there's a mismatch for the general public, but they don't matter.

This is exactly why I said you should look into their history. They took off big time. Search Amazon for "expert systems" and observe how many books there are about the subject.

They were used by large corporations, hospitals, universities and governments. They were the subject of a lot of research. There were countless startups based around the concept.

> AT&T just sent me a letter saying they're giving me an extra 5GB/month bonus data as long as I keep my plan, which I assume happened because their data indicated I was a risk for downgrading or switching carriers.

This isn't Artificial Intelligence though. These seem like they fall under 'Reactive systems' or 'Data Analytics'.

In any case, it doesn't matter what you call it. Current research is producing results that can improve these systems, which means that research is worth a lot of money.

Hmm, I've not heard of this happening. Do you have any examples of it?

> In any case, it doesn't matter what you call it. Current research is producing results that can improve these systems, which means that research is worth a lot of money.

I don't see how this really adds to or is relevant to my comment, I was simply restating and adding to hitekker's point that AI is improperly named.

The general public does not set budgets for AI research.

AI research benefits from faster processors, more memory and more advanced algorithms. Only philosophers (the domain of AGI is philosophy, not so much engineering or maths) are uncomfortable with those details.

Please refer to AGI, if you are talking of the hypothetical strong AI. The current textbook definition is alright.

(And yeah, the hype around AI is palpable and annoying. That's why many researchers called their work "cognitive science", "machine learning", "optimization", "logic" or "applied maths" and avoided "AI". Because else, they'd have to argue semantics, or defend why they haven't build an artificial God yet...)

Large corporation definitely could do better when it comes to explaining the limits of their AI/ML technologies and putting those technologies in perspective (especially historic perspective). Scaling down on hyperbole, buzzwords and personification would help as well.

AI is not some magical ineffable thing that will someday appear out of nowhere. It's the name we give to the gradual progression of our efforts to build machines to perform cognitive tasks. Someone from the pre-computer age would have had no problem recognizing even the ENIAC as intelligent, in a limited sense; it could solve problems that were previously too hard even for the smartest human mathematicians. As someone who actually does AI research on a daily basis, I have no problem with granting intelligence to even very basic chatbots. That goes hand in hand with recognizing that there are many kinds and levels of intelligence, and plenty of opportunity to build smarter and more flexible systems.

* Viruses would probably be a more apt analogy.

Pedantically yes, you are taking inputs into a black box and recieving output.

This level of reductionist thinking actually hampers progress in AI, because you're stuck on Chinese room. I mean seriously, they already had this discussion like 20 years ago "fam".

Yeah, you are right. It's some progress.

* For the record, we don't even know if it even possible to make transistors in this hypothetical universe of AI.

I believe this was answered in GEB. Hofstadter mentions how humans (and some animals) can step out of a system and analyze it objectively. Meaning, we can take the rules of a system, analyze, and determine that we will never be able to generate the desired results. We can objectively lot at things (even ourselves) and reason about them.

A computer, on the other hand, even the most advanced AI, is still just blindly executing the commands given to it.

In my reading of GEB, Hofstadter was criticizing other authors who disagreed with his understanding of the Church-Turing thesis (that human intelligence is, or at least is not more powerful than, some kind of rule-following system). Hofstadter thinks that there is no inherent contradiction or essential difference in kind between the human who "can step out of a system" and the computer "just blindly executing the commands given to it".

(But Hofstadter didn't explain at a technical level how to make a computer that's as intelligent as a human being.)

As mentioned, humans can step out of a system and analyze it objectively. In essence, the rules can be broken and changed at any time, which means the system is constantly evolving. Logic does not always truly need to play a part, either. The system is constantly changing and adjusting.

Because of this limitation due to our inability to implement systems that are truly nondeterministic, we are forced to use the constantly improving resources (such as faster processors, memory, better algorithms) to improve the speed at which we can mimic this behavior. However, these methods still require the data at hand to function, and several of them lead to exponential growth. As you said, it is always limited to what it's given.

Anyone who does any coaching work is intensely aware of these limits (in everyone, btw, the coach included!)...