Semantic Communication: Uncertainty in the Digital Age
Photo: Jason Dorfman
CSAIL Professor Madhu Sudan is thinking about communication. His project posits that communication is possible between beings with no common bond of language or shared history. Its direct substantiating case examines a theoretical instance of the third kind, placing extraterrestrials and their instruments in contact with humans and their computers in a quest for understanding. The scientific background of the work that Sudan and Brendan Juba have done is fascinating enough. But when the exceedingly far-reaching implications are taken into consideration, it raises exciting questions about the meaning of communication itself in the digital age and beyond.
The project has its beginnings in a 1948 text entitled “A Mathematical Theory of Communication.” Authored by Claude Shannon of MIT and Bell Laboratories, the paper dealt with the problem of mathematically encoding data for transmission. Shannon, widely considered to be the father of information theory, laid careful groundwork for the parameters of mathematical communication and the limits to which it could successfully be taken. Turning the clock ahead to 2009, many researchers feel that Shannon’s limits are being approached, or maybe even have already been reached. But the fact that there are still errors in computational communication necessitates a different way of examining the system and where it might go wrong.
Says Sudan, “the new model of error is not what happens in the wire that’s connecting me and you – it’s in me and you.” The idea of flawed communicators leading to flawed communications is both very old and very new. Legions of subdivisions of the humanities are dedicated, in one way or another, to adequately and exactly expressing an idea despite inherent limitation. But whether the medium is a learned language or a style of painting, the imperfections at the ends of the wires impose their own boundaries on what’s possible. As he discusses the near-certainty of flaws, professor Sudan explains that computing systems have attempted to circumvent this issue by imposing standards, which work to an extent and for a while.
“But we’re all adaptive entities; our computers are constantly adapting their software. And when they’re adapting their software, the meaning of the bits that are being sent to them gets misunderstood. We really had to figure out: what is a notion of misunderstanding, and how can you make it precise enough that a computer can deal with it and start fixing it?”
No small task: pinpoint and clarify the idea of obfuscation. In order to leave as little room as possible for misunderstanding in the explanation itself, the researchers needed to look at communication using methods more sparse than those employed by theorists in the past. When communication was viewed simply, as a means to an end rather than an end in and of itself, it was far easier to judge where flaws occurred and whether or not the goals of the communicative effort were being met.
One thing the team noticed immediately was that flawed communications began to occur due to the fact that computers were checking their own conversational flow in order to bring in human input based on their fundamental design antecedents: computers that did not have the communicative power to speak directly to each other. Since modern computers have evolved to the point when they are reaching and even exceeding their conversational limits, it may be time to stop bringing the human to the machine and start, instead, to bring the human into the machine.
The key idea is a proposed separation of computer communication into a system that functions less like a mechanical circuit and more like the human nervous system, in which the division between automatic and voluntary actions is more heavily weighted towards the automated. The implication is that over time, human intervention would become necessary on only the problems where their approval is expressly needed, imbuing the machine with more logical and reasoning capability, resulting in an ultimately more powerful computing system – one with a degree of agency.
What to do with these newly empowered machines? Use them to speak to aliens, of course – or at least externally test the limits of communication. This is done by setting up a controlled experiment wherein the (newly clarified) idea of misunderstanding is a necessary function of the task at hand. This is a twist on the theory of interactive proof, which deals with mistrust as a fundamental part of communication, in which information is selectively revealed to the other party so that the necessary data is established without compromising a certain level of security. By changing mistrust to misunderstanding, one is able to posit a new way of understanding the exchange of knowledge – and in so doing, take a surprising look at the fundamentals of communication itself.
Adwoa Gyimah-Brempong, CSAIL