Dr. Dobb's | Back to the Future with John McCarthy

Back to the Future with John McCarthy

Michael goes back to the future with John McCarthy.

April 23, 2008
URL:http://www.drdobbs.com/architecture-and-design/back-to-the-future-with-john-mccarthy/207401594

When they make the movie of his life, John McCarthy should be played by Christopher Lloyd. There seems to be no computer science challenge too big for this legendary founder of artificial intelligence to take on with the exuberance of Dr. Emmett Brown in Back to the Future, and you have to admit that McCarthy looks the part of a mad scientist.

Recently, he's been talking up Elephant, a proposed very high-level language for transaction processing and electronic data exchange—more or less what XML is supposed to be for, but Elephant is more ambitious as a language. It is, McCarthy says, "on the border of AI." Jimmy Guterman caught his talk at Etech and reported on it at O'Reilly Radar.

McCarthy takes on transactions in terms of "performatives," a term that I remember well from my grad school days. I came to computer science from psychology with some linguistics and philosophy of science helping to bridge the gap, and one of the books that I came upon back then that seemed to live in all four of these domains was J.L. Austin's How to Do Things with Words. Austin pointed out something curious in that book, based on his William James Lectures at Harvard: Many utterance that look like statements are not intended to convey information at all, or at least that is not their primary purpose. These utterances, which he named performatives, are actions in the form of speech. When you promise something, you are not (merely) conveying information to someone, you are performing an act. Placing a bet, making an offer, taking a vow, hiring an applicant, giving an order, accepting an assignment, acquitting a defendant, declaring war, are all performatives. And as Austin points out, commands are closely related to performatives. They're odd things, performatives: Austin finds that the logical relationships of entailment, implication, and presupposition all work a little differently for performatives than for factual statements. What makes performatives particularly interesting for transaction processing and electronic data exchange is that those examples of performatives that I just listed look like just the things we want to do in online transactions.

Guterman seems puzzled by McCarthy's assertion that "ascribing beliefs to thermostats is like adding 0 and 1 to the number system," but McCarthy dealt with this precise example in his 1979 paper "Ascribing Mental Qualities to Machines," which can be found on his website (www-formal.stanford.edu/jmc) in html, dvi, pdf, and postscript versions. He even provides a machine-centric "glossary of mental qualities." I'm not saying that your head will be clearer on the matter after you read that paper—McCarthy freely tosses around Kantian jargon like "syncategorematic" and wrestles with questions like whether the proper objects of machine belief are sentences or propositions—but in that paper the thermostat is his number one example for considering whether and when it can be legitimate to ascribe mental properties to machines.

McCarthy's view is that it is unnecessary but legitimate to ascribe beliefs like "The room is too hot" or "The temperature is 24 degrees" and a goal like "The room should be OK" or "The temperature should be 20 degrees" to a thermostat. And while it does little good to think of a single thermostat in this way, McCarthy suggests that it is helpful in thinking about the class of possible thermostats.

But do we think of thermostats as having intentions? Sure we do: It's the reason—if I'm interpreting McCarthy correctly—that we consider a thermostat with its wires to the furnace cut to be a broken temperature control system rather than just a certain collection of parts. McCarthy even cites a common example of a simple machine to which it makes sense to ascribe introspective beliefs; i.e., beliefs about the machine's own internal states. Consider an alarm clock, getting power back after an outage: It's legitimate to say that it realizes that it doesn't know the time and is flashing to signal this.

Not that any of that has anything to do with transactions. But McCarthy has been thinking about very high-level languages for transactions over networks for a long time. In 1975, inspired by an article by networking pioneer Paul Baran that "envisaged a world of the future in which companies would be well equipped with on-line computer systems," McCarthy wrote "The Common Business Communication Language." In this article, the updated version of which is also available on his website, he proposed a language for business transactions much like XML. Here's a transaction in CBCL:


(REQUEST-QUOTE 
	(YOUR-STOCK-NUMBER A7305) 
	(UNITS 100)).

and here it is in XML:


<REQUEST-QUOTE> 
	<YOUR-STOCK-NUMBER> A7305 </YOUR-STOCK-NUMBER> 
	<UNITS> 100 </UNITS> 
</REQUEST-QUOTE>

In that paper, McCarthy acknowledges that even handling something as simple as a purchase order involves some thorny semantic issues. In Elephant, he presumably thinks he's solved some of them.

Michael Swaine

Editor-at-Large

[email protected]