Disambiguating expected likelihood of discrete outcomes

I’ve been writing about disambiguation techniques. Here’s another one. This one applies when the question cannot be disambiguated into a continual variable.

Suppose Adam and Bob and a bunch of friends live in a house together. They all enjoy living together and hope to continue for a long time to come. But a complication has emerged. Adam is allergic to dogs, and Bob’s girlfriend has a dog. Bob would like to have his girlfriend come over more — and maybe even move in some day — but it’s hard for her to do that because she has to leave her dog at home.

This poses a risk to Adam and Bob’s goals of living in the house together. If Bob can’t have his girlfriend over, he might have to move out. Or Bob might not move out, but might someday find that his relationship with his girlfriend has been strangled by the fact that she doesn’t spend much time at his house. Or if Bob brings his girlfriend’s dog over anyway, it may cause Adam to have constant allergies, and Adam might have to move out.

There’s also a possibility that some happy compromise can be found — perhaps if the dog is bathed regularly it won’t produce an allergic reaction. Or maybe the dog can be limited to certain areas of the house that Adam doesn’t care to spend time in.

But Adam and Bob find themselves having a hard time talking about the issue. If they were to disambiguate using this technique, they might find out why.

The first step is to list the possible outcomes. They agree that there are 4 possible outcomes worth discussing. Next, they each record their likelihood estimate for each outcome.

Here’s what they come up with.

OutcomeAdam’s expected likelihood of this outcomeBob’s expected likelihood of this outcome
Bob moves out25%10%
Bob stays but his relationship with his girlfriend is strangled50%35%
Adam moves out20%5%
A happy solution is found whereby all of the above are false5%50%

Finally, they compare results. Looking at the last row, they see a huge delta on their expected likelihood of a happy solution. Now it makes sense to both of them that Bob has been excited to talk about solutions, while to Adam this has felt wrong.

From here, with a greater understanding of one another’s point of view, they can further disambiguate to find out why they have such different likelihood estimates for this outcome. They’re doing productive disagreement.

Comparing disambiguated views

I’ve been writing about disambiguation and the high-dimensionality of superficially low-dimensional phenomena like abortion. 

The continuum from pro-life to pro-choice can be visualized as a single dimension. 

But it’s probably more helpful to think of one’s position on abortion as high-dimensional. It’s composed of your views on questions like “when does life begin?” and “how much should we value the preferences of the would-be mother?” We can say that your positions on those dimensions project to a position on the single pro-life/pro-choice dimension. 

We can disambiguate one person’s views this way, and we can also disambiguate a second person’s views this way. And we can map both of their views onto a single graph. 

For many issues, this would go a long way towards getting two people to understand one another and to disagree productively rather than unproductively. 

We could also compare one person’s actual views with what the other person thinks that person’s views must be. This type of misunderstanding/ambiguity is responsible for a lot of unproductive disagreement. 

Disambiguation is disentangling dimensions

A few days ago I wrote about Eric Weinstein’s discussion of the “middle” position on political issues. He used abortion as an example. We have the terms “pro-life” and “pro-choice”, and most people’s position is probably somewhere in the middle.

Weinstein said something interesting about this so-called “middle”. 

I don’t think it’s at the middle. …. I think that there’s this very flat, low dimensional plane where these positions [pro-life and pro-choice] live. And what we’re calling the middle is not the thing between these. It’s in a higher dimensional space that combines these crappy low resolution, moronic positions, and it projects to the middle, but it isn’t the middle.

I’m interested in visualizing this. Here’s a sketch. 

Current solutions to the ambiguity problem

In recent posts I’ve been talking about a gap in our ecosystem of communication tools. My last post used an example from Eric Weinstein to illustrate the problem. Here I want to build on that post by looking in more depth at the current solutions that Weinstein has available to him. 

What solutions are available to him right now? 

He can say “well I’m pro-choice, generally”. This is succinct but not accurate.

He can give a monologue articulating his actual view. This is accurate, but not succinct. 

He can write an essay, publish it on the internet, and give the view articulated therein a name — eg, he could call it “pro-utility” (if he thinks his view is utilitarian). Then when asked, he could say “I’m pro-utility. You can read my blog post about it.” This gets him succinctness and accuracy. But it introduces another problem. Asking someone to read an essay on your blog is a tall order. He’ll say “I’m pro-utility, you can read my blog post.” And his audience will, for the most part, not read it. So now he’s failed to get them to understand at all, let alone understand succinctly and accurately. 

If he could get his new word, “pro-utility” into the popular lexicon, then he could say “I’m pro-utility” and not have to refer to is blog post because people would already know what it meant. Now he’s getting understanding accurately and succinctly. One obvious difficulty with this approach is that it’s difficult to get a new word propagated into the popular lexicon. A second difficulty is that if he manages to succeed at popularizing the word, he’ll soon run into the same problem he started with: people will use the word “pro-utility” to name a range of views which aren’t actually what he meant. They might be closer to his actual view than “pro-choice”, but there will be ambiguity resulting in loss of accuracy. 

What else could he do? 

He could make a short video and put it on YouTube. But this seems to have the same problems as his essay. Maybe the video is shorter — but this would seem to come at the cost of accuracy or specificity.  

He could find someone — a scientist or politician perhaps — who has already written an essay, and refer people to that. But again, people aren’t going to read it. 

The unfortunate fact, I believe, is that he doesn’t have a good solution to this problem given current tools. 

The missing tool of the fifth estate

If the three traditional estates of the realm are the clergy, the nobility and the commoners, and the fourth estate is the mainstream media, then we might say, as William Dutton has, that networked individuals enabled by the Internet are the fifth estate. 

But the Internet has not provided all of the tools for communication and democratic action that the fifth estate requires. There’s at least one important gap. 

I’ve been writing recently about Eric Weinstein’s podcast episode with Timur Kuran. We can draw from that episode an illustration of the missing tool. 

Giving conversation about abortion as an example, Weinstein says that he wants to escape the “enforced conversation of morons” in which we identify our views as one of two positions in a single dimension: either pro-life, or pro-choice. Weinstein says that he caucuses with the pro-choicers, but that his real position is “a plague on both your houses”. 

I’m not pro-choice to the extent that I’m willing to call a child four minutes before its birth ‘fetal tissue’. Nor am I pro-life to the extent that I’m going to call a blastosphere a ‘baby’. Both of those seem patently insane to me.

my key point is, look, I’ll drop these people in a heartbeat if you give me some nuanced room in which to maneuver. Let’s talk about the neural tube formation.

Let’s talk about what we think of as life…nowhere do I get to discuss Carnegie stages in embryonic development, which would be sort of a more scientific approach to what quality of life is it that we’re trying to preserve.

What Weinstein wants is to be able to convey his actual position on abortion, succinctly. Currently, he has to choose between the two: he can succinctly convey something that isn’t his actual position (“pro-choice”), or he can laboriously convey his actual position. But he can’t do both. 

His inability to do both forces him to choose. And it prevents him from accomplishing a number of important things. It makes it hard for him to be seen and understood accurately. It makes it difficult for him to have truth-seeking conversations with reasonable people. It makes it difficult for he and others, if they have different views, to find out whether they might want to caucus together for some particular purpose. 

I believe that there’s a tool that’s within our power to build which would let Weinstein and everyone like him communicate about their actual positions succinctly. Let’s call it the Disambiguator. I’ll talk about it (probably) in a future post. 

Precision vs accessibility in natural language

The ambiguity of natural language is often a benefit: it gets us in the right neighborhood with parsimony. And it helps us to caucus around views that may not actually be as similar as our language makes it sound.

But one thing ambiguity is not good for is specificity. Sometimes we don’t want the broad, general, multiply-interpretable word. Sometimes we need a precise semantic package. In those cases we can use — or invent — jargon. But jargon is, by definition, inaccessible.  

How might we get the best of both worlds — accessibility and precision? 

I think we need new tools.