AI Hype, or, To MIT and Beyond!

Here in Fredericton, NB, hype about IT is part of the city plan.  It reverberates in the local campus, is amplified in the media, and so it goes.  Fuzzy neural net agents.  Evolving fuzzy neural net agents.

But seriously, what about M.I.T. ?   Engineering practical robots is still not enough.  The new claims seem to reflect a change in plan - late in life.  I say that because some of the staunchest advocates may now realize that they will not live to see mechanical minds.

The strategy now covers both integration of systems and strange admissions.  While there is no good reason to believe that evolution of smart organisms such as ants, bees, cuttlefish and octopus result from anything like integrating Java modules with Python modules except at the level of rhetorical tropes, the funding flows.

Consider M.I.T. itself.  If it is true that MIT abandoned LISP for JAVA as the programming language of choice - not multi-paradigm Oz or proven engineering languages such as FORTRAN, ADA or SMALLTALK .... not R or APL ... but betting on what future?

Here is a symptom: programming for artists at MIT.  At the institution with MediaLab, LOGO, CLU in its hstory, we get processing.org and the Java-bound Processing language.

What was a language for artists?  Quite possibly ICON or UNICON - both are languages not wedded to TRUE-FALSE logic but success and failure - what works and does not work.  Both are not hard-core AI languages: back-tracking behavior in ICON is as restricted as it is in Erlang.

Currently, an "artist" would download a 60+ MB zip of Processing replete with a complete Java JDK environent.  I recently installed Processing 1.2.1 only to be confronted with egregious bugs.  Their bug list is impressive.

What is worse, Processing, the language, was never needed: MIT had already produced Curl, the language (www.curl.com).

Now we see an effort by James Resig to salvage Processing in JavaScript.  Not in Self or Self/Avocado or even Io, but in processingjs at processingjs.org.

The fate of Processing in JavaScript will be that the one innovative idea in Processing will be lost.

Why are we not teaching artists a Smalltalk such as Pharo or introducing them to a project such as Cobalt?

Now this iis just computing for artists, not computing for business and he military.  Fact: the two robotc arms of the ISS are programmed in ADA.  But artists working on simulations for ISS training would start in what ... AcionScript?

As for innovation, the legendary internet is plagued by two truly bad ideas: Perl and PHP.  HTML5 does not resolve that.  Apple may have abandoned Java, but they are using a Smalltalk-cum-C++ in Objective-C which show no sign of evolving.  Will the Steve Jobs replacement embrace Scala.

Consider Scala.  Scala tries to take a lesson from business and from OOP:  embrace VM's and Traits.  But Traits some from Self, one of those languages ignored at MIT.  Traits emerged because some dicta of programming gurus were wrong and cannot be solved by AOP.

So what impedes the evolution of smart languages?  Tenure?  Take the most recent release of Squeak Smalltalk as 4.2   The Squeak folks embrace the idea that when a program starts it can be begin by consuming a script which can modify most any aspect of both the environment of the program and the program itself.  Any of us who worked in demanding IT industry projects can tell you that this is a compliance and governance non-starter.  What leaves the project leads - the academics - so out of touch that Squeak could 'evolve' in a way which imperils its survival in any real world niche?

Now to weave back to the hype at M.I.T.   One of the most important AI innovations of the 70's was not embraced at MIT.  The French company was absorbed by a large French corporation.  The gist of those ideas lives on in AI in Europe and in 1.4 Distributed Oz of the oz-mozart project.  In the first decade of the new millenium, major projects at large US corporations showed no glimmer of those ideas.  the result is dumb software tackling medical pricing in hospitals and dumb software tackling risk management in banks, insurers, governments ... and MIT?

After years of pursuing internal representations (a notion long debunked by mere philosophers and linguists) MIT AI has embraced 'leaving much of the information out there in the environment".  This idea of "invariants in the ecology" was championed decades ago by JJ Gibson.  It was old hat by 1964 in the work of French philosopher Maurice Merleau-Ponty.  MIT AI is slowly embracing "embodied intelligent behavior" - again old hat in the work of M M-P from his critique of the psychology of his day and his lessons from the early days of the neurology of brain traumas and military amputations.

 Much of what passes for the "programme" of AI at MIT is just rhetoric and most often relies on a variety of fallacies.  When in 1960 an infallible pope declared as dogma that Mary had been taken up physically into heaven as both body and soul, the faithful did not ask for clarification as to whether she had first died at the end of her days.  The dogma trades on words.  The terms of the trade at MIT are 'agent',' 'action', 'learn', 'adjust', 'adapt', 'integrate', 'system'.

The promise of neural net began with the mollusc slug.  What reason would a skeptic have to believe that such a project will evolve to build something on the order of a cuttlefish or octopus.  Just take the one example of the obsession with duplicating the hominid hand.  It ha taken years for someone to think outside of that box the 'gripper' as inflatable/deflatable bean-bag.  A mollusc might have evolved one of those (and might yet do so if the seas survive us.)  But.  What reason to think that such a cephalopod  would resemble anything like the proposed MIT AI strategy?

Take IBM's Watson strategy.  Not a robot and not "intlelligent" (expect no witticisms, sarcasm, tantrums, devious resistance, passive aggression ... ) - but needing to show an AI result, IBM abandons the one big strategy for a more evolutionary notion of let all try and go with the most promising in any one round.  Unlike warfare on the shifting battlefield along long battle lines, Jeopardy is polite repetition.

Dolphins evolved something akin to language and even a cross-species lingua franca.  Cuttlefish have, in their skin, evolved something like a mental image, which we, their homunculi-by-proxy, witness and which often confounds their dolphin predator - perhaps on occasion evoking delphine amazement, delphine amusement, a delphine tingle that travels down the spine.  No robot wil tingle all over when it realizes that THAT piece of 'coral' ain't no coral nor will it feel the grit of its teeth on cuttlebone.  No robot appears to have any prospect of evolving to tolerate the ammoniac taste of the giant squid as has the sperm whale.  Integrating taste sensors with vision sensors is not what occurred there.  Adopting common messaging protocols across progamming libraries of procedural languages is not what occurred there.

I can think of no way except embarrassment to move someone past the doctrine of the physical assumption.  Perhaps seriously thinking of Yeshua the healer helps the believer move one step towards atheism.  The AI faiths are much harder nuts to crack: a "physical assumption to heaven" is "adjusting action to environment" - which just sounds so sensible.  Just as heaven was once above.  Just as necrosis was once a mystery.  The bodies of the pure do not rot.  The robot  "solves the problem".  Language will not make it so.  The kingdom does not come in this lifetime.