[peter303]

We went through a round of this in the 1980s. The first commercial graphics workstations happened to be LISP machines. So management confused non-numeric code with A.I. There was demand for workstation experts. Not to loang after this UNIX graphics workstations like Sun, Apollo and MicroVAX came out and the market switch to UNIX/Linux.

Second was the expert systems boom in the mid-1980s. This was fanned by Stanford professor Fegeinbaum who wrote the infamous book The 5th Generation about expert system computers being the future and Japan was building the best ones. These would either be LISP machines or an interesting French niche language called prologic. Prologic basically traversed a databse "if-then" rules (modus pons). These machines went nowhere and Japan economy tanked in the early 90s. Lot of Silicon Valley VCs lost big on this.

Prof Feigenbaum may still be correct, but 40 years early. However the new A.I. is driven by massive database matching possible in modern peta-level computers and not so much logical computing.

[mangeletti]

What you're referring to is the advent of AI bubbles, and thus AI Winters - http://en.wikipedia.org/wiki/AI_winter

Stating that a bubble cycle has emerged in the means only accentuates the importance of the end, to note that the desire for AI is so strong that futility hasn't kept people from trying.

Virtual reality is another such example.

[VLM]

That would be prolog. You'll have much better luck googling for prolog. I played with "turbo prolog" in the 80s and accomplished nothing (from the same place as turbo C or turbo pascal or turbo basic (am I forgetting any?)). A modern variant (of a logic oriented language) can be seen here:

https://github.com/clojure/core.logic/wiki/A-Core.logic-Prim...

It tends to suffer from management by scalable procedure disease. Its possible to successfully replace a human assembly line worker with a robot arm and a very small shell script, which inevitably leads overactive imaginations to think of replacing engineers or doctors with an immense set of unfortunately undefinable unscalable procedures and rulesets, so it always collapses with complexity at implementation time. Its like moths to a flame, you should be able to replace an engineer with a very long list of if/then statements, but it turns out to be impossible in practice. Meanwhile the more advanced techniques butts up against the rapidly scaling "DBA" "IT" type of traditional solutions or non-traditional big-data techniques.

Its hard to find something to logic program that isn't less verbose in a non-logic language or unwritable in any language including logic programming. Its like the Perl regex thing where you got a problem, so you write a regex, and now you got two problems. Its a very narrow although interesting niche. Finding something that fits would be pretty cool, although probably very difficult to maintain.

[peter303]

The first four generations were defined by hardware: (1) vacuum tubes, (2) transisitors, (3) integrated circuit boards, (4) microprocessor full CPUs on a chip. I would define (5) clusters and (6) mobile. Candidates for next generates include huge data engine clouds, wearables and internet of things.

[Florin_Andrei]

No, that was the wrong approach. Back then we believed that solving AI means being strong on logic and rules, with data being a secondary aspect. Nowadays we do the exact opposite: data is king, and the rules are expected to somehow emerge from it.