(Senator John McCain (R-Arizona) voting against the “skinny repeal” health-care bill on July 28th, 2017)
LC: So what did you end up working on in graduate school?
SS: My idea was to try to create a program that simulated the way people make decisions. Fortuitously or obviously or whatever, I ended up with politics as the domain. So I created a program that simulated the way a member of congress voted on a particular bill. You would give the program a member of congress and a particular bill and it would look at the member’s goals, which could be expressed in terms of issues like being for or against gun control or balancing the budget or the environment or abortion, and then you would look at the consequences for voting for or against it.
So part of the model was that a person has his or her own constellation of issues that are important to them and they have a history of how they have made these decisions in the past from which you can infer behavior. They also have relationships with other agents–friends, family, colleagues. The strengths of those relationships, if it’s a positive relationship, would allow the agent to adopt the other person’s goals.
So in the case of members of congress they had relationships with various lobbying groups like the NRA, the ACLU, the republican party, or the democratic party, and each of those groups had their own constellation or hierarchy of goals. And so if a member of congress has relations with dozens of groups, suddenly there are hundreds or possibly thousands of goals that he or she has to keep track of when they are voting.
And what’s interesting about this is that sometimes the goals can be in conflict. Now in a mathematical logic system, if you have both p and not p that’s a contradiction and the system falls apart. Like Reductio ad Absurdum, that’s how you prove the halting problem–you reach a contradiction. But the point is that people have to deal with contradictions all the time. So in that sense, this wasn’t strictly speaking a logical model of decision making, it was a psychological model because it would handle the fact that people are often of two minds on a given topic.
And it finds a way to weight these various goals?
Well that’s one way to do it, but actually that didn’t seem psychologically plausible to me. I don’t people are up there doing math in their heads. And actually, talking to politicians and looking at the political science literature, at least for congressional voting, what they report they do is they vote in a way that has the best story, the best explanation. Because they know that when they vote, they have to offer an explanation in a press release or something.
And so that’s one explanation. Another interesting thing in the literature is that there are cases of minorities voting against civil rights legislation. And you might be confused about why they do this, voting against their and their constituents’ interests. But they do it when the passage of the vote is not at stake to draw attention to themselves. They can have a press conference and say “This doesn’t go far enough, we have to keep fighting” or whatever. And interestingly, the same argument you can find with conservative members of congress who are sort of fiscally responsible, and they might vote against a spending reduction bill. And you think “well aren’t they in favor of spending reduction.” It’s the same thing. Their vote doesn’t endanger the passage, but they can register a protest. That’s a compelling explanation; they like that.
It’s similar to what happened when Nixon became president in 68/69. One of the first things he needed to do was raise the debt ceiling so the government could keep going. But when the Democrats were in the White House, congress had always voted against raising the debt ceiling. So now they were being asked by Nixon to vote in favor of it. And they wanted to help Nixon out, but they were in a quandary–they would look like hypocrites.
Well they had to come up with a story. The story was that yes, the deficit is too high, and we need to get it under control, but the new republican administration will be more fiscally responsible, and we need to give them a chance… so we are going to raise the debt ceiling.
The point is that the decisions are mitigated not explicitly by the weights but by these explanations. So when you make a decision, I argue, you have to decide what story you are going to tell. And often the story you are going to tell is different based on who you are talking to. So in the case of the politician, the program would make a decision and then it could generate a natural language explanation for why it made the decision. And you could tell it to tailor it to a particular audience. And so congressperson, now you’re talking to republicans. How do you explain it? Now you’re talking to democrats or the NRA, how do you do it?
So I did that, I had been teaching at Yale, I had written some books here, but I left and went to NYU to the business school. My thought was that, in business school, the sort of unifying paradigm of business school education is how to make a decision. And the dominant paradigm in business school for decision making is economic decision theory, which does do the weighting thing. So you calculate the expected value based on the probability and the payoffs and you get an optimal choice, which is good right?
And my theory, there’s no optimal choice. There’s no concept of a right answer, which may be disturbing, but I think it’s probably more realistic, that people make decisions all the time and they want to think that they found the right answer.
You can argue about this. There is a professor at Stanford, I think his name is Hoffman, who argues that people confuse the quality of the decision with the outcome of the decision. You know if a doctor treats a patient and the patient dies, you think that’s a bad decision. Well maybe the doctor was doing the right thing and 99 times out of a hundred the patient would die and maybe this patient would die regardless of what the doctor did. So maybe looking at the outcome isn’t the best way to evaluate the decision.
So you look at hospitals around the country. What state has the hospitals with the highest mortality rate? Off hand do you know?
No I don’t.
Florida. Does that mean they have bad hospitals in Florida? They’re probably really good hospitals. But you have this really old population which skews the statistics. So the outcome is not necessarily the best way to judge the quality of the decision.
So anyway I’m at business school and I decide I can’t do politics anymore, so I hit upon security analysis, where you have people deciding not whether to vote for or against a bill but whether to recommend for or against buying or selling a stock. They often use quantitative models like discounted cash flow models. So for a company you predict what are the revenue streams for the foreseeable future, then you take the net present value of that, and if the net present value is above the current stock price then you would buy the stock and if it’s below then you would sell the stock. That’s pretty straightforward. So I decided to see if maybe the people who did the analysis could offer some insight, reflect on my model.
So I talk to analysts on wall street, and I would tell them what I did. And they would generally say yes that is what we do. We look at the numbers, we do the quantitative analysis, but when they make a recommendation they have to tell a story. They can’t just say “Buy IBM I’m 93 percent sure.” And one reason for that is that the client… You have to understand that security analysts are not buying or selling stocks, they’re making recommendations to people who do buy and sell stocks, customers or portfolio managers. And they have to convince them that they’re offering good advice. And if they just said “I’m 93 percent sure.” Well maybe if the guy is really good that’s fine, but they’re not going to be right all the time.
But if they tell a story, they say “Buy IBM, we think interest rates are going to be low, inflation is going to be low, they have a new market in Europe and they are able to hedge the foreign currency risk and their competitor has a new machine but it’s having trouble…” The client, the portfolio manager can follow this line of reasoning and decide if they agree with it. So in a sense they can validate or invalidate the chain of reasoning and then they decide if they agree with the recommendation. And if it turns out that it’s not correct, it’s probably because one of the assumptions was wrong. But the portfolio manager was making the same assumption so he might be less likely to blame the security analyst. So anyway security analysts tell stories and recognize this as an important thing.
So the idea is that you have a model of decision making that takes into account narrative or justification. You can use that model of decision making to develop a model of persuasion. If you want to change someone else’s mind, and you have a model of their decision making process, you can try to mold your argument in a way that takes advantage of what you know. So maybe you can highlight a goal that they weren’t thinking about or you can point out they have another relationship that should influence them in some way.