Tags
communication, deception, experiment, HCI, IBM, life, media, psychology, relationships, truth, UX, writing
In my early days at IBM Research (1970’s), we were focused on trying to develop, test, or at least conceive of ways that a larger proportion of people would be able to use computers. One of the major ways of thinking about this was to use natural language communication as a model. After all, it was reasoned, people were able to communicate with each other using natural language. This meant that it was possible, at least in principle. Moreover, most people had considerable practice communicating using natural language.
One popular way of looking at natural language (especially among engineers & computer scientists) was essentially an “Encoding – Decoding” model. I have something in my head that I wish to communicate to you. So, I “encode” my mental model, procedure, fact, etc. into language. I transmit that language to you. Then, you “decode” what I said into your internal language and — voila! — if all goes well, you construct something in your head that is much like what is in my head. Problem solved.
Of course, people who wrote about communication from this standpoint acknowledged that it didn’t always work. For instance, as speaker, I might do a bad job of “encoding” my knowledge. Or, I might do a good job of encoding, but the “transmission” was bad; e.g., static, gaps, noise, etc. might distort the signal. And, you might do a bad job of decoding. It’s an appealing model and helped engineers and computer scientists make advances in “communication theory” and helped make practical improvements in coding and so on.
As a general theory of how humans communicate, however, that notion is vastly over-simplified. I argued then that a better way of looking at human communication was as a design-interpretation process, not as an encoding-decoding process. One of the examples that pointed this out was a simple observation by Don Norman. Suppose someone comes up to you and asks, “Where is the Empire State Building?” You will normally give a quite different answer depending on whether the two of you are in Rome, Long Island, or Manhattan. In Rome, you might say, “It’s in America.” Or, you might say, “It’s in New York City.” If you are on Long Island, you might well say, “It’s in Manhattan.” If you are already in Manhattan, you might say, “Fifth Avenue, between 33rd and 34th.”
Building on Don Norman’s original example, but based on your own experience, you can easily see that it isn’t only the geographical relationships that influence your answer. If you were originally from Boston, now on your own in Rome, struggling with Italian and homesick and someone came up to you and asked that question in American English with a Boston Accent, your response might be: “Are you joking? But how did you know I was an American. My name’s … “
On the other hand, if you’re a 13-year old boy in Manhattan — one with a mean streak — and someone asks you this question in broken English and they’re looking around like they are totally lost, you might say, “Oh, no problem. Just follow 8th Avenue, all the way north up to 133rd. It’s right there. You can’t miss it.” (Note to potential foreign visitors, most kids in Manhattan would not intentionally mislead you. But they point is, someone could. They are not engaging some automatic encoding process that takes their knowledge and translates into English. Absurd!
You design every communication. I think that’s a much more useful way to conceive of communicating. Yes, of course, there are occasions when your “design” behavior is extremely rudimentary and seems almost automatic. It isn’t though. It just seems that way. Let’s go back to our question-asking example. Suppose you work at an information booth in New York City. People ask you this same question day after day, year after year. You’re seemingly giving the answer without any attention whatsoever. Suppose someone asks you the question, but with a preface. “Look here, chap! I’ve got a gun! And if you give me the same stupid answer you’ve given me every time before, I’ll shoot your bloody brains out!” You are going to modify your answer. It only seemed as though it was automatic.
When you design your answer you take into account at least these things: some knowledge that you communication about, the current context (which itself has hundreds of potentially important variables), a model of the person you’re creating this communication for, a set of goals that you are trying to achieve (e.g., get them safely to their goal, mislead them, entertain them, entertain yourself, entertain the people around you, demonstrate your expertise, practice your diction, etc.). The process is inherently creative. In many circumstances (writing, playing, exploring, discovering, partying), you can choose how creative you want to make it. In other cases, circumstances constrain you more (though likely not so much as you think they do).
Many readers think this is a classic example of a straw man argument. “No-one believes communication is a coding-decoding process.”
Well, I beg to differ. I worked for relatively well-managed companies. I’ve talked to many other people who have worked in different well-managed companies. We’ve all seen or heard requests like this: “I need a paragraph (or a slide or a foil) on speech recognition. Thanks.”
What??
Who’s the audience? Are they scientists, investors, customers, our management? How much do they already know? What are your goals? What other things are you going to talk about with them? The people who have left me such messages were all smart people. And, providing the necessary info would have only taken a minute or two. But it would have substantially improved the outcome. It’s not a straw man argument.
Sit-com plots often hinge on the characters doing poorly at designing and/or interpreting communications. A show based on encoding-decoding? No. What could be funny — indeed what often is shown in comedy — are people failing to do good design and in the extreme case, that can arise by having an actual robot as a character or someone who behaves like one.
People also interpret what was said in terms of their goals, the context, what they believe about your goals and capacity, what they already know, and so on. And, even though this may seem obvious, millions of people believe what advertisers or politicians say without questioning their motives, double-checking with other sources, or even looking for internal inconsistencies in what is being touted as true. In other cases though, the same people will not believe anything the “other side” says no matter what. Just as one can do faulty design, one can also do faulty interpretation.
In any case, I decided that it would be good to “show” in a controlled laboratory setting that the Encoding-Decoding model was woefully inadequate. So, I brought in “subjects” to work in pairs at a simple task about communicating Venn diagram relationships. The “designer” had a Venn diagram in front of them. “The “interpreter” was supposed to draw a Venn diagram. The “designer” was constrained to say something true and relevant. In addition to a “base” pay, the “interpreter” subjects would be given a bonus according to how many relationships matched those of the “designer.” The designer’s bonus depended on condition. In the “cooperation” condition, their payoff would also, like the interpreter’s, be determined by how much agreement was shown in the two diagrams. In the “competition” condition, the designer’s bonus depended on how different the two diagrams were.
I ran about half the number of subjects I had planned to run when the experiment was ended by corporate lawyers.
What?
IBM had no unions at that time. And, they didn’t want any unions. One of their policies, which they believed would help them prevent the formation of unions was that they never paid their workers for piece-work. Apparently, somehow, IBM CHQ had gotten wind of my experiment. People were being paid different amounts, based (partly) on their performance. They couldn’t have this! People might think were paying people for piece-work!
It hardly needs to said, I suppose, that IBM definitely tried to pay for performance. This was true in sales, research, development, HR, management, and so on. No-one in IBM would argue that your pay shouldn’t be related to your performance. That was exactly — in one way of describing it — was going on here. By the way, these were not IBM employees and each subject only “worked” for about an hour.
Basically, regardless of how irrelevant this experimental set-up might have been to the genuine concern of unions not to pay people in an insanely aggressive and ever-changing piece-work scheme, the lawyers were concerned that it would be somehow misrepresented to workers or in the press and used as evidence that IBM should unionize. In a way, the lawyers were proving the point of the experiment in their own real-life behavior even as they insisted that the experiment needed to be shut down.
Lessons Learned: #1 Corporate lawyers are not only concerned about what you actually do or how you represent your work; they are also worried about how someone might misrepresent your work.
Lessons Learned: #2 Even when constrained to say something true and relevant, ordinary people are quite capable of misleading someone else when it’s to their benefit and considered okay to do.
It is this second aspect of the experiment that I myself felt to be “edgy” at the time. Sure, people can mislead, but I was providing a context in which they were being encouraged to mislead. Was that ethical? Obviously, I thought it was at the time. On reflection, I still think it’s okay, but I’m glad that there are now review boards to look at “studies” and give a less biased opinion than the person who designed the study would do.
I view the overall context of doing the study as positive. As adults, these people all already knew how to mislead. I was letting them, and many other people, know that we know you know how to mislead and we’ll be on the lookout for it.
What do other people think about studies wherein the experimenter encourages one person to deceive another?
2026 Update: I’ve spend a fair amount of time recently “chatting” with ChatGPT and with Claude. It’s clear that the implementers of these systems (or, at least some of them) were quite familiar with the idea of designing and interpreting text. Claude, in particular, seems like a “nice guy” and an honest one at that. It acknowledges some of its own limitations and often praises the questioner. This shouldn’t come as a complete shock. After all, good sales people and advertisers have relied for centuries on “Design and Interpretation.” Nonetheless, I sometimes wonder whether AI systems might be more ethically employed if they were operating on principles closer to “Coding and Decoding.” What do you think?
———————-
References published literature that describes some of the research that was done around that time.
Malhotra, A., Thomas, J.C. and Miller, L. (1980). Cognitive processes in design. International Journal of Man-Machine Studies, 12, pp. 119-140.
Carroll, J., Thomas, J.C. and Malhotra, A. (1980). Presentation and representation in design problem solving. British Journal of Psychology/,71 (1), pp. 143-155.
Carroll, J., Thomas, J.C. and Malhotra, A. (1979). A clinical-experimental analysis of design problem solving. Design Studies, 1 (2), pp. 84-92.
Thomas, J.C. (1978). A design-interpretation analysis of natural English. International Journal of Man-Machine Studies, 10, pp. 651-668.
Thomas, J.C. and Carroll, J. (1978). The psychological study of design. Design Studies, 1 (1), pp. 5-11.
———————
Other essays that touch on communication.
Freedom of Speech is not a License to Kill
The Sound of One Hand Clasping
Fool Me
