Shape

401 Notes to p. 305
24. The ‘‘no ambiguity’’ sign was sent to faculty and PhD students in the Computer Science and
Artificial Intelligence Laboratory (CSAIL) at MIT. According to one of the students, ‘‘I got the sign
from a guy who claimed to have ‘solved AI’ thus removing all the ambiguity in the field/world. . . .
I found the no ambiguity sign to be silly/fun and reusable at the same time, because a great chunk
of our research is about getting rid of ambiguity that crops up during sketch recognition.’’ The
sketch recognition language is described in T. Hammond and R. Davis, ‘‘LADDER: A Language to
Describe Drawing, Display, and Editing in Sketch Recognition,’’ in IJCAI–03: Proceedings of the
Eighteenth International Joint Conference on Artificial Intelligence, Acapulco, Mexico, August 9–15,
2003, ed. G. Gottlob and T. Walsh (San Francisco: Morgan Kaufmann, 2003), 461–467. The
authors mention shape grammars, albeit in their aboriginal form, as ‘‘shape description languages.’’
Scholarship can have some funny results. The gap between what you think you’re doing
and saying about it, and how others understand and use this, can be huge—and who’s to say
who’s right? It’s easy to be wrong about meaning, whether or not you’re the source. The only
thing to do is to keep on talking, and seeing and drawing, so that meaning can change. Change,
of course, is what CSAIL is about. It’s proud of its accomplishments and sure that it can do more—
‘‘[ We] have a history of daring innovation and visionary research which change the way the rest
of the world works. We think this is how it ought to be, and are organizing [the Dangerous Ideas
Seminar Series] to help stimulate people to think big.’’ Gerald Jay Sussman was one of the ‘‘instigators.’’
He enlarged on this theme in the title of his seminar on March 17, 2005: ‘‘Engineering as an
Intellectual Revolution.’’ And what he proposed was indeed ‘‘intellectual’’—
The key idea is the development of engineering ‘‘languages’’ that allow us to separate concerns in design. Such
languages provide ways of expressing modularity and isolation between modules. They provide means of composition
that allow the construction of compound systems from independently-specified and implemented
parts. They allow characterization of both structure and function, and how function is determined by and
implemented in terms of structure. They provide black-box abstractions that allow one to specify the behavior
of a composition independently of the implementation.
Computer scientists like to divide things into independent units to make problems combinatorial.
Language is vocabulary and syntax, and an engineering language uses both to fix structure
and function. Compositionality precedes design—what a compound system does depends on
its constituent modules (parts), what they do separately, and how they’re put together. The
shape grammarist agrees that this is a dangerous idea—before you know it, design is impossible.
Sussman, to be sure, has other hazards in mind. Once-novel ideas may be dangerous because
they’re revolutionary when they’re tried in new places. Today, compositionality is key in biological
engineering—MIT’s ‘‘Registry of Standard Biological Parts’’ holds an amazing future. And
compositionality works elsewhere outside of engineering—for example, there’s ‘‘the timeless way
of building’’ in Christopher Alexander’s pattern language. But what does any of this add to Franz
Reuleaux or Herbert Simon? What is it about how we’re trained to think that makes both design
and calculating combinatorial? Neither has to be. Still, there’s no telling where a big idea might
go, and Sussman’s big idea is no exception. Computers are used everywhere, so perhaps CSAIL
will ‘‘change the way the rest of the world works’’ in art and design—but only if there’s nothing
to see.
25. There are other options in computer science, for example, R. A. Brooks, ‘‘Intelligence without
Representation,’’ Artificial Intelligence 47 (1991): 139–159. Brooks shares the shape grammarist’s
distaste for representation—‘‘explicit representations and models of the world simply get in the
way. It turns out to be better to use the world as its own model.’’ This is true for shapes, and is
the reason for embedding and maximal elements. But the focus is elsewhere in AI—‘‘[To segment