Code v. Speech and Writing: Jeopardy Style

In 2011, Jeopardy! employed an experiment with the help of IBM.  The show invited two contestants with the highest earnings in the show’s history to play a match against a computer designed by IBM.  This computer, named Watson, was fed Jeopardy clues in the form of electronic text at the same time that the clues were made visually available to the human contestants.  Using keywords from the clue in order to search its database for responses containing those same words, Watson would locate several possibilities which seemed to fit with the clue.  Jeopardy! revealed the three answers in which Watson was most confident as it rang in and answered.  In order to ring in, however, Watson had to be sufficiently confident in the answer at which he arrived.  Watson’s ability to locate an answer and determine his belief in his correctness continually proved faster than its human counterparts and he ultimately beat them every round in which he played.

Watson's Performance on Jeopardy

Although Watson was generally accurate, there were a few issues that he was unable to negotiate.  Ringing in after another contestant who had an incorrect answer, Watson rang in and gave the same answer.  Since there was no text-to-speech function from the human contestants to Watson, it was unable to realize that it was giving an identical, and wrong, answer as another contestant.  Additionally, Watson provided the wrong answer to the Final Jeopardy question as he answered “Toronto” to the category entitled “US Cities.”  Since the term “US Cities” was not specifically stated in the actual clue, programmers speculated that Watson may have been responding to other keywords in the clue (including the fact that Toronto is, in fact, the name of several cities in the United States) or possibly that Watson parsed the clue in regards to its syntax, causing it to inaccurately group keywords.

Watson's Final Jeopardy Confusion

While Watson was largely accurate in his responses, these minor glitches seem to speak to the issues of translating speech to code which N. Katherine Hayles explores in “My Mother Was a Computer: Digital Subjects and Literary Texts.”  According to Hayles, “No matter how sophisticated the program… all commands must be parsed as binary code to be intelligible to the machine” (46).  Since the binary code is straightforward and absolutely determinable, it does not allow for the type of rhetorical ambiguities which theorists such as Derrida attribute to language.  Hayles continues: “In the worldview of code, it makes no sense to talk about signifiers without signifieds [and] [s]imilarly, it makes no sense to talk about floating signifiers” (47).  Signifiers cannot float with code because of the determinability of meaning enforced by the binary system, thus signs in code do not possess the same twofold structure that they do in Saussurean linguistics.  Watson, then, understands signs merely for the exact definition with which they are conventionally imbued and can only make sense of them with their traditional meanings.

Answering “Toronto” to a question about cities in America suggests that Watson and, further, code, are unable to process the nuances and rhetorical possibilities implicit in speech and writing.  Only able to consider language as a mere decoding of syntax and meaning, Watson must make do with traditional conceptions of any given sign rather than allowing for rhetorical ambiguities and varied syntax. 

Anticipating his loss to Watson, Ken Jennings, a Watson component, wrote on his Final Jeopardy wagering form, “I for one welcome our new computer overlords.”