The hedge fund manager Michael Burry said he met his wife through the dating site Match.com with a profile that read, “I am a medical student with only one eye, an awkward social manner, and $145,000 in student loans.” Burry said she wrote back, “You’re just what I’ve been looking for.” She understood that if Burry was willing to advertise his unflattering shortcomings on his profile page, he was likely to be someone who possessed the trait she was most seeking in a partner: honesty.
If you take a step back and look at what Burry was doing, you see that the meaning of his message was not specific to his vision, social traits, or finances. It was about a hunch that prospective partners on a dating site would see honesty, a flattering attribute that’s hard to convey, in a guy who was willing to list his unflattering attributes.
Burry took a calculated risk. His hunch was that some people would be turned off by his message — perhaps many people, or even most people would be turned off — but there would be at least one person who would accept his negative traits, and that was the kind of person he was looking for.
In the 1930s Claude Shannon was thinking deeply about how people communicate with each other. It may sound like an odd infatuation for a graduate student with degrees in mathematics and engineering. But his scientific investigations into areas of communication through the 1930s and 40s led to several discoveries that continue to influence work being done today, long after his death.
Shannon began with the idea that when someone communicates with someone else, there is a chance that the speaker’s message may not have the desired effect on the listener. Despite the speaker’s best efforts, the attempt to communicate effectively may fail for various reasons. For instance, the speaker may not choose the most accurate words, an issue that Shannon described as a technical problem. Or, the listener may have a different understanding of the words that the speaker chose, which Shannon called a semantic problem. Or, the received message may not affect the listener in the desired way, which Shannon called an effectiveness problem. Regardless of whether the problem is technical, semantic, or lacks effectiveness, the communication may fail.
He wondered if there was a way to generalize all those reasons into one broad category. Shannon identified them all as forms of “noise” that interfered with the speaker’s intention of achieving a desired effect for the listener.
He described the process in the context of electronic communications. There’s an information source that gets transmitted as a signal that the listener hears acoustically. In the process, there is potentially some noise that gets introduced. So when the receiver picks up the signal, there is both the original information and the noise combined. The noise may be quiet and insignificant, or the noise may be so loud that it drowns out the information, or the noise may be any volume in between. Depending on how much noise there is, the receiver may need to block the noise or filter it or reduce it so as to be able to understand the message embedded in the signal.
I know we’re talking about two different things here — the communication of symbols (like words) and the communication of signals (like acoustic sound waves) — but Shannon thought about both in the same way. Both forms of communication have to overcome challenges for the communication to be successful. Both have to deal with noise.
At Bell Labs, the research subsidiary of American Telephone & Telegraph Company (AT&T) where Shannon went to work, the standard way to overcome noise when sending a signal over a telephone line was to simply increase the strength of that signal. The farther a signal had to travel over carrier lines, the more it relied on boosters, relays, and amplifiers. But eventually, those technologies faced limits. For instance, transmitting a signal from New York to London involved thousands of miles of deep-sea cables which were subject to punishing conditions and required massive amplification systems. The more the cables eroded, the more noise degraded the signal, and the more amplification would be required to overcome it. In other words, the solution to noise was to simply make the signal louder.
Shannon wondered if it were possible to solve the communications challenge using logic rather than engineering. It was a counterintuitive thought at the time. What does communication have to do with logic? Well, the connection wasn’t clear until Shannon did some deep thinking about the connection. And he realized there are at least three solutions when a listener couldn’t hear you beyond just shouting louder.
Let’s say you’re on a plane. There’s a lot of noise. The flight attendant asks you if you want pretzels or cookies. You say cookies. The exchange is simple. A question with 3 words: “Pretzels or cookies.” It’s easy to understand. You answer with one word: “Cookies.” That answer is easy for the flight attendant to understand. You get cookies and the attendant moves on to the next traveler.
The question has two choices, pretzels, cookies, and one qualifier, the word “or,” which signifies that you are allowed to choose one or the other. In Shannon’s logic, the question “Pretzels or cookies?” could be abstracted to the form, “P or C?” You respond, “Cookies,” which would be the equivalent of the response, “C.”
So, the whole conversation goes as follows.
Flight attendant: “P or C?”
You: “C.”
But the flight attendant could have asked a different question. She could have asked you if you wanted both pretzels and cookies. The qualifier “and” signifies that you are allowed to have both choices. In Shannon’s logic, the question can be represented as “P and C?” Now you have a slightly broader array of possible answers. You can answer, “yes” (both P and C), “no” (neither P nor C), “pretzels” (P only) or “cookies” (C only). The range of possible answers is now four instead of two, but it’s still a straightforward exchange.
You could broaden this exchange by adding new qualifiers, to the question and the answer such as “if,” “then,” “not,” and “but.” If you did this, your logic would become more complex, but all the premises would still be valid.
But remember, you’re on an airplane, and the conversation is noisy. There’s the possibility that you might not understand what the flight attendant has asked you. Instead of the question, “P or C?” you might hear, “B or Z?” and you think you are being offered something that you are not actually being offered, which could lead to confusion.
So what could the flight attendant do to address the noise challenge? In Shannon’s logic, the simplest way would be to simply repeat the question, “P or C, P or C, P or C?” This would reduce your chance of misunderstanding what you are being offered. That’s a good approach that doesn’t involve shouting louder.
Another way would be for the attendant to somehow filter out the noise in the verbal message, making the question clearer. For instance, the attendant could get very close to your ear when asking the question. This would increase the signal relative to the background noise. Looking at this mathematically, if the noise is represented logically by the letter “x”, and the unfiltered question is “Px or Cx?” then filtering out the “x” would leave the unambiguous question, “P or C?” I know, if the flight attendant got close to your ear to ask the question, she wouldn’t be eliminating the noise altogether, she would just make it sufficiently small so as to make it insignificant relative to the question. But you get the point. Filtering the noise is another good approach that doesn’t involve shouting louder.
As an aside, you may already be using that type of system if you are using noise-cancelling headphones. Those headphones receive noise in the form of various frequencies at various volumes, and then dynamically filter those frequencies at those volumes, thereby cancelling out the noise to leave you with a signal that is easier to hear.
Another way to overcome the noise would be to change the type of information that’s being communicated, such as using visuals rather than words. The attendant might do this by holding up pretzels in one hand and cookies in the other for you to choose. She might even make eye contact and raise her eyebrows as if to imply a question: “Do you want pretzels or cookies?” Then you could respond by simply pointing to the one you want. That’s another solution to the noise problem that doesn’t involve shouting.
Once Shannon began to formulate his theories about communication, signals, noise, and the various ways that noise could be processed, he began to see that the same logic could be applied to any type of information. Language, sound, pictures, movies, you name it, these could all be handled with the same signal-to-noise logic.
We’ll pick up with that topic in the next post.