Three formal arguments for why there cannot be a finite set of cognitive biases. The lack of a finite set of biases is one reason I believe Heuristics and Biases cannot survive long term as a prescriptive paradigm. A paradigm of deviations must eventually collapse under its own weight. - ThreadSky

jaredpeterson.bsky.social • 90 days ago

Three formal arguments for why there cannot be a finite set of cognitive biases.

The lack of a finite set of biases is one reason I believe Heuristics and Biases cannot survive long term as a prescriptive paradigm. A paradigm of deviations must eventually collapse under its own weight.

1 / 3

Comments

conjugateprior.org•90 days ago

Premise 3 of the NFLT sequence seems sensitive to the assumption, roughly, that the distribution of future decision problems has support over all possible problem space of problems. If it doesn't, an 'heuristics and biases' list indirectly indexes the relevant subset, which fixes what's (ir)relevant

jaredpeterson.bsky.social•90 days ago

That's one of the core arguments against the NFLT. There could be consistency in problem space such that only a small subset of prescriptive models are actually optimal in real life. I'm sympathetic to this. At one point I even wrote a blog arguing the same thing, and applied it to...

jaredpeterson.bsky.social•90 days ago

...evolution. You can treat a species as an algorithm striving to optimize in their niche. And despite impressive biodiversity, there is also convergent evolution, suggesting that decision space is constrained and NFLT isn't 100% correct in the real world. Perhaps if we discover alien life...

jaredpeterson.bsky.social•90 days ago

...it will look remarkably like Earth life, and that would be another blow against NFLT.

Nevertheless, I tend to think NFLT is *mostly* correct. There is not infinite diversity, but probably still too much variety to be managed in the way H&B tries.

fragri.bsky.social•90 days ago

📌

jaredpeterson.bsky.social•90 days ago

I don't normally write formal arguments, but thought I would try something new. Apologies to the philosophers who I'm sure will cringe at my attempt.

jaredpeterson.bsky.social•90 days ago

Useful chart from Jonathon Baron's Thinking and Deciding. It shows how each bias corresponds to a prescriptive model, and then explains the heuristic that is the cause of the deviation

jaredpeterson.bsky.social•90 days ago

Baron also makes a distinction between prescriptive and normative models. I'm ignoring that distinction so that I don't have to use the term "normative" which I think confuses people. But happy to discuss the difference if someone thinks it's relevant to the argument I'm making

gregsasso.bsky.social•90 days ago

will the heuristic that there is a finite set be useful though

jaredpeterson.bsky.social•90 days ago

Ha! Perhaps we will have to pretend as if there are if we want to continue to use H&B prescriptively.

mattwallaert.com•89 days ago

And there it is. This is why I don’t talk about these things when doing applied behavioral science: they are descriptive at best and better descriptions often exist for in-situ behavior change.

jaredpeterson.bsky.social•89 days ago

Arguably better descriptions ALWAYS exist. It is always better to describe a behavior using a heuristic than a bias. The former is an actual explanation, whereas a bias is merely explaining what a behavior is not (ie not "economically rational").

mattwallaert.com•89 days ago

That’s very well said.

milkos.bsky.social•90 days ago

Dude. But what about dynamic set of heuristics? We crate them and change them on the go.

jaredpeterson.bsky.social•90 days ago

Not sure what you mean

milkos.bsky.social•89 days ago

You say that HB will fall under its own weights because HB is not finite. However, notice that your argument could also support the claim: the HB changes on the go. So it’s not finite because it’s dynamical, and also the space of possible HB is changing.

milkos.bsky.social•89 days ago

But maybe I don’t understand what you mean by HB program, and why impossibility of finite set is so devastating to such program

jaredpeterson.bsky.social•89 days ago

The problem arises if we view the paradigm as trying to catalogue biases. Right now there are 200+ biases on wikipedia, but if the arguments above are correct, then that list will never stop growing.

jaredpeterson.bsky.social•89 days ago

We could select a finite set of prescriptive models to compare human decisions against. Arguably this still wouldn't fix the problem, but even if it did, it would mean that we would be unsure if a given biases actually represents a deviation from rationality/optimality.

chrisschuck.bsky.social•90 days ago

What's the difference between "unbounded range of possible decision-making scenarios/diversity of real world contexts" (NFLT) & "vast range of SIFs such as option order/ framing effects/contextual cues" (upper-right)"? Aren't SIFs just examples of the former; why treat it as a separate argument?

jaredpeterson.bsky.social•90 days ago

NFLT argument is about the prescriptive models which serve as benchmarks. For every given prescriptive model, we can introduce a new way in which we might deviate from that model. So we have at least as many biases as there are prescriptive models.

jaredpeterson.bsky.social•90 days ago

The SIF argument is about descriptive models. The argument is that even if we had only one prescriptive model there could be a non-finite set of ways to vary from that model. ie there are a non-finite set of correct descriptive models of Decision-Making which deviate from Expected Utility.

chrisschuck.bsky.social•90 days ago

Oh I see, thx. Might be helpful to clarify that. My impression reading was that the whole reason NFLT holds is precisely because of all these SIF cases, and so all the variability in prescriptive models boils down to the variability resulting from SIF. But it sounds like these are still independent?

jaredpeterson.bsky.social•90 days ago

Still independent. The SIF argument presumes there's a thing that is optimal, and the SIF takes you away from optimality. The NFLT is a statement about the number of optimal models

chrisschuck.bsky.social•90 days ago

I see. So the NFLT doesn't really refer to deviations, but to non-finite number of models where there are zero deviations?

chrisschuck.bsky.social•90 days ago

I'm especially intrigued by the one on the left (personal beliefs, values, and philosophical frameworks), since these observations are so foundational to interpretive social science and reflexive qualitative research, but not necessarily brought into direct dialogue with the H&B program per se.

Comments

Posting Rules

Reply