22 comments

[ 3.2 ms ] story [ 61.3 ms ] thread
From the article:

"Fast forward to 1992. The world looks very different. In 1990 the Nikkei Index, which had risen strongly for an unprecedentedly long period, from the beginning of the FGCS project, was about to breach 40,000. But instead of continuing its rise, it started a decline and was down to half the peak value by 1992. Most of the Toyotas and Hondas driving around in the US were mostly made in the US. If MITI was mentioned at all, it was in studies revealing that MITI had never sponsored a successful project; that industry, far from being helped by MITI, had been hindered by its meddling."

From Wikipedia: "The Ministry of International Trade and Industry (通商産業省 Tsūshō-sangyō-shō or MITI) was one of the most powerful agencies of the Government of Japan." http://en.wikipedia.org/wiki/Ministry_of_International_Trade...

Article assumes technology is adopted top-down, from the powers that be. While that was true at the time of mainframes, for a while now, technology is often adopted bottom-up, particularly in startups that are free to use the most effective tool for the job; and in non-mission critical systems within the enterprise. For example, this is often how open source was adopted.

The picture is complicated by libraries and tool-chains that can make an established language (c, java, python) a better choice by giving an enormous (e.g. x1000) boost to productivity, obscuring the intrinsic merits of the language itself. (Although, languages that facilitate library and tool use will promote creation of their ecosystems).

"Logic programming is a cool idea, but not that useful for most things" would be a better explanation for why it's not popular today. It strikes me as a rather narrow and dogmatic AI technique; and one that you're better off reimplementing on demand (Greenspun-style), or using in a library.

> "Logic programming is a cool idea, but not that useful for most things" would be a better explanation for why it's not popular today. It strikes me as a rather narrow and dogmatic AI technique; and one that you're better off reimplementing on demand (Greenspun-style), or using in a library.

Particularly since confining logic programming to an embedded domain-specific language lets you keep it clean from any of pollution that was piled onto Prolog in an attempt to make it more of a general purpose language.

Incidentally, implementing logic programming in Scheme is the topic of an entire chapter of SICP.
"Logic programming is a cool idea, but not that useful for most things"

Except that many regard Prolog as a precursor that little used language SQL.

Sql is slightly older than Prolog. They may have come from similar roots. And sql is used as an embedded subset not a whole systems programming language.
My explanation at LtU, in response to a question from Ehud Lamm:

    I think that Prolog has lasting interest as a teaching language, and is one of 
    the handful of PLs that everyone with a broad interest in programming should  
    learn, but that the limitations of its basic computational mechanism make it 
    uncompetitive as a general PL. I find that Haskell, through the list of 
    successes paradigm, is usually a better language for writing the kind of 
    programs Prolog is good at than Prolog.

    I think the LP community should be seriously looking for a successor. There's 
    no shortage of candidates.
- http://lambda-the-ultimate.org/node/2803#comment-41684

As I said in that thread, I think the LP community is in denial about the unsuitability of Prolog as a general-purpose programming language; this article echoes the Lispers claim that Prolog was killed by its association to over-ambitious AI projects: I think the argument in the case of Prolog is much weaker than in the case of Lisp.

I don't think most of the LP community still promotes Prolog, although it's hard to delineate exactly what the LP community is. Going by papers presented at AAAI and IJCAI, the bulk of the research is on answer-set programming (http://en.wikipedia.org/wiki/Answer_set_programming), which tends to be used as a more specialized solver rather than as the top-level loop of a program. In other corners of the world, multiparadigm logic/functional languages like Mercury seem to be what people are hanging their hats on.
That's a good example. I've always thought of ASP being more influenced by SAT solvers than Prolog, but following the links from the WP article, there's a lot of Prolog culture there: Horn clauses, Herbrand interpretations, &c. So maybe LP is moving on. I should take a closer look. (I probably should have years ago: I've done a fair bit of refereeing for IJCAI over the years).

Multi-paradigm languages are not so great for the Kowalski-esque LP world, because the elements of these languages don't fit in the Kowalski/Pierce picture of knowledge representation and reasoning. Something like lambda-Prolog, which uses higher-order unification to model functional programming within a clausal language, seems to me to be a much better, more conservative step for this community.

I think ASP's a bit of a merger of the two cultures. The implementations are heavily SAT-influenced, with all existing ones based on eliminating variables to ground to a finite program that's then solved. But its semantics (stable-model semantics) originated in the early-90s flurry of attempts to come up with a solid logical semantics for LP (alongside other attempts like the well-founded semantics), to replace Prolog's operationally defined "whatever SLDNF resolution returns" semantics, and especially its displeasing-to-logicians property of the semantics depending on statement order. And the most widely used surface syntax, sometimes called "AnsProlog", is an adaptation of Prolog syntax.

This research group in particular is now pushing it heavily in the programmability/expressivity direction, which is pretty interesting: http://potassco.sourceforge.net/ . There is even some other research floating around on more software-engineeringish things like ASP debugging techniques.

A few followups:

A elaboration from van Emden, which describes part of the problem as Prolog being pushed by Lispers who had only recently become enamored with Prolog: http://vanemden.wordpress.com/2010/08/31/the-fatal-choice/

A different take that blames "technological and cognitive issues" at least in part, especially debugging: http://synthese.wordpress.com/2010/08/21/prologs-death/

A different take that blames competition from more specialized declarative-programming languages: http://www.kmjn.org/notes/prolog_lost_steam.html

Nice point on competition (the last one), that SQL (and even LINQ!) replaced some aspects of prolog. I would guess that mathematica/matlab/R also have...? BTW In academia, a related competing paradigm is http://en.wikipedia.org/wiki/Constraint_logic_programming.
Yeah, I'd say some use of constraint-programming in lp is now nearly standard, though sometimes in a better-integrated way than other times. Nearly every Prolog system now has both clp(fd) and CHR built in, for example, but they don't always interact with Prolog derivations in an ideal way (especially not with Prolog negation-as-failure).

Answer-set programming has constraint stuff more as a core feature. A common paradigm in asp is to mix "normal" lp type rules, where you specify how to derive exactly the correct solutions, with a generate-and-test style where you over-generate (with choice rules) and then filter with constraints (called integrity constraints in asp), which is sometimes easier to specify.

The recent comments by Eric Schmidt about school level CS education in the UK (even though there is no UK-wide educational system) made me look at what my 12 year old son might be studying in a few years.

Interestingly enough it appears that Prolog is alive and well and in Scottish schools - for example see this exam past paper:

http://www.sqa.org.uk/pastpapers/papers/papers/2010/AH_Compu...

Prolog is great for teaching: I don't many disagree with this, and I don't that this is what's at stake in <i>Prolog has been killed</i>.
My experience with Prolog was that of a very arrogant CS student which got very confused by the non-procedural way of prolog and the (for me) weird syntax.

So it's probably the root cause for me that we've not got an proper introduction to non-procedural languages. What left is a horrid experience of trying to solve backtracking problems with prolog.

For my personal universe, prolog got killed by my education.

It's not dead yet - it was used for NLP in Watson.