Are all models wrong?

[Peter Holmes]

George Box claimed that ‘All models are wrong but some are useful’. But if that claim is true, then at least one model is not wrong – in which case, the claim is false. Iow: if all models are wrong, then so is the claim that all models are wrong. But, leaving that aside, there are other problems with the claim.

1 To clarify: Box probably didn’t mean all models are immoral. He likely used the word wrong to mean incorrect, inaccurate, imprecise, incomplete, imperfect – and so on. (Unbelievably, it seems necessary to point out that we can use the words right and wrong non-morally.)

2 A model can be said to be wrong only if it makes sense to say it could be right. But what would a model, map or description that is right – correct, accurate, precise, complete or perfect – look like? How much and what kind of information would it have to contain? The absurdity of these questions exposes the absurdity of the claim that all models are wrong.

3 We could not model a reality or map a domain to which we have no access, or of which we have no knowledge or information. In that case, any model or map we produce would be a fiction or fantasy, and its usefulness would be entirely fortuitous.

4 The solipsistic claim that we know or can know nothing about what we call reality is an affectation exposed at every turn by performative contradictions, of which the use of language to express the claim is merely one.

The fact is that all models are models – full stop. They are not and cannot be the things that they model or describe, which are features of reality. A factual assertion and its truth-value – and any assessment of ‘rightness’, accuracy, precision, completeness or perfection – can exist only within a descriptive context. So the claim that all models are wrong is incoherent.

To say we have no objective standard by which to assess how well a model describes reality, or that we can’t know if our claims are true or false, is to misconstrue the actual relationship – and radical separation – between a description and the described. The myth of propositions, the JTB definition of knowledge, correspondence theories, and truth-maker/truth-bearer ideas all demonstrate the conflation of what we say with what we say it about – as does the course of foundationalism – its existence and rejection.

All posited foundations for what we know – and (therefore) the truth of what we say – are merely models. What we call facts are such models, and they constitute the objective knowledge we express in language. We build and repair this knowledge on foundations and with materials of our own making. But that does not mean the edifice has no foundation, and so must be shaky. That we can always say more does not mean we can never say enough.

[Terrapin Station]

Re "If that claim is true, then at least one model is not wrong." Are you using "model" so broadly that any proposition whatsoever is a "model"?

At any rate, what he probably had in mind is simply that any model isn't identical to what it's modeling. That should be pretty obvious.

[h_k_s]

George Box claimed that ‘All models are wrong but some are useful’. But if that claim is true, then at least one model is not wrong – in which case, the claim is false. Iow: if all models are wrong, then so is the claim that all models are wrong. But, leaving that aside, there are other problems with the claim.

1 To clarify: Box probably didn’t mean all models are immoral. He likely used the word wrong to mean incorrect, inaccurate, imprecise, incomplete, imperfect – and so on. (Unbelievably, it seems necessary to point out that we can use the words right and wrong non-morally.)

2 A model can be said to be wrong only if it makes sense to say it could be right. But what would a model, map or description that is right – correct, accurate, precise, complete or perfect – look like? How much and what kind of information would it have to contain? The absurdity of these questions exposes the absurdity of the claim that all models are wrong.

3 We could not model a reality or map a domain to which we have no access, or of which we have no knowledge or information. In that case, any model or map we produce would be a fiction or fantasy, and its usefulness would be entirely fortuitous.

4 The solipsistic claim that we know or can know nothing about what we call reality is an affectation exposed at every turn by performative contradictions, of which the use of language to express the claim is merely one.

The fact is that all models are models – full stop. They are not and cannot be the things that they model or describe, which are features of reality. A factual assertion and its truth-value – and any assessment of ‘rightness’, accuracy, precision, completeness or perfection – can exist only within a descriptive context. So the claim that all models are wrong is incoherent.

To say we have no objective standard by which to assess how well a model describes reality, or that we can’t know if our claims are true or false, is to misconstrue the actual relationship – and radical separation – between a description and the described. The myth of propositions, the JTB definition of knowledge, correspondence theories, and truth-maker/truth-bearer ideas all demonstrate the conflation of what we say with what we say it about – as does the course of foundationalism – its existence and rejection.

All posited foundations for what we know – and (therefore) the truth of what we say – are merely models. What we call facts are such models, and they constitute the objective knowledge we express in language. We build and repair this knowledge on foundations and with materials of our own making. But that does not mean the edifice has no foundation, and so must be shaky. That we can always say more does not mean we can never say enough.

You really owe your reader an explanation of who "G. Box" is.

It's not like he's Norm Chompsky and everybody knows him.

[Pattern-chaser]

At any rate, what he probably had in mind is simply that any model isn't identical to what it's modeling. That should be pretty obvious.

Yes, "the map is not the territory". And because the map is not the territory, every map (model) is wrong in the sense that it does not (cannot) conform exactly to the territory in every respect.

[Pattern-chaser]

You really owe your reader an explanation of who "G. Box" is.

Explanation here. You're welcome.

[Steve3007]

Open Google. Type "George Box". First result:

https://en.wikipedia.org/wiki/George_E._P._Box

First item in his "known for" section:

"All models are wrong"

suggests we've hit the right guy.

---

I think the OP is correct when it say (albeit tautologically) "all models are models". i.e. it is in the nature of models that they represent some, but not all, of the properties/characteristics/things of which they are proposed to be a model. If they represented all of those things then they wouldn't be a model. They'd be the thing itself.

All of the theories of physics and, more widely, all objects (the things to which objective propositions refer) are arguably types of models.

So, in summary, the OP already covered it.

[Terrapin Station]

Open Google. Type "George Box". First result:

https://en.wikipedia.org/wiki/George_E._P._Box

First item in his "known for" section:

"All models are wrong"

suggests we've hit the right guy.

---

I think the OP is correct when it say (albeit tautologically) "all models are models". i.e. it is in the nature of models that they represent some, but not all, of the properties/characteristics/things of which they are proposed to be a model. If they represented all of those things then they wouldn't be a model. They'd be the thing itself.

All of the theories of physics and, more widely, all objects (the things to which objective propositions refer) are arguably types of models.

So, in summary, the OP already covered it.

How would an object be a model?

[Steve3007]

How would an object be a model?

It would be a model in the sense that it is a concept that we devise in order to describe and predict a related subset of the sensations experienced by ourselves and others.

Take the example of the object that I could describe as the table that is currently in front of me, and on which I am resting my elbows as I type. I'll just call it "table". I experience various sensations (in this case visual and tactile ones). I note patterns in those sensation; common characteristics. I hypothesise that the cause of these common characteristics, or patterns, is "table". That hypothesis is a form of model. I use that model to predict future related sensations, and in so doing I test the model. I use that model to predict that other people will also have related sensations. Not the same sensations. Related ones.

No matter how many sensations I or others have, we will never experience all possible sensations that we could attribute to "table". So our "model" can never be entirely complete (i.e. it's always a model). There is always the possibility that a future sensation will not fit the pattern. That might cause me to modify the model, or even abandon it in favour of a new one. It might make me think that there is not an object called "table" there after all.

A standard illustration. Macbeth's dagger scene:

"Is this a dagger which I see before me,
The handle toward my hand? Come, let me clutch thee.
I have thee not, and yet I see thee still.
Art thou not, fatal vision, sensible
To feeling as to sight? or art thou but
A dagger of the mind, a false creation,
Proceeding from the heat-oppressed brain?"

Macbeth creates a hypothetical model of an object called "dagger" based on the evidence of one sensation (sight). He modifies that model on the basis of another sensation (touch).

[Peter Holmes]

Re "If that claim is true, then at least one model is not wrong." Are you using "model" so broadly that any proposition whatsoever is a "model"?

At any rate, what he probably had in mind is simply that any model isn't identical to what it's modeling. That should be pretty obvious.

My point is that a factual assertion presupposes a model which, if the assertion is true, can't be wrong. How can a model that's wrong produce or generate a true assertion?

And my main point is that, since the very idea of a model being 'identical to what it's modelling' - or 'right' - is incoherent, it follows that it makes no sense to say all models are 'wrong'.

[Peter Holmes]

At any rate, what he probably had in mind is simply that any model isn't identical to what it's modeling. That should be pretty obvious.

Yes, "the map is not the territory". And because the map is not the territory, every map (model) is wrong in the sense that it does not (cannot) conform exactly to the territory in every respect.

What I'm trying to get at is the radical difference and separation between a description (a model) and the described. They are such different things that the very idea of conformity of any kind - exact or not - is a delusion.

Features of reality do not and cannot conform to the ways we identify, name and describe them in any way whatsoever. To think they do or can is to get things completely back to front.

[Steve3007]

Features of reality do not and cannot conform to the ways we identify, name and describe them in any way whatsoever. To think they do or can is to get things completely back to front.

If that is true, how do we even know which features of reality these descriptions are intended to correspond to?

[Peter Holmes]

Features of reality do not and cannot conform to the ways we identify, name and describe them in any way whatsoever. To think they do or can is to get things completely back to front.

If that is true, how do we even know which features of reality these descriptions are intended to correspond to?

That's the problem with correspondence theories of truth: 'the claim 'snow is white' is true because real snow really is white'.

There's no way out of that tautology, because there's no foundation, for what we say, beneath our linguistic practices. A name no more corresponds with what it names than an arrow corresponds with its target.

[Pattern-chaser]

A name no more corresponds with what it names than an arrow corresponds with its target.

Yes! As we said earlier:

Yes, "the map is not the territory".

[Sculptor1]

George Box claimed that ‘All models are wrong but some are useful’. But if that claim is true, then at least one model is not wrong – in which case, the claim is false. Iow: if all models are wrong, then so is the claim that all models are wrong. But, leaving that aside, there are other problems with the claim.

1 To clarify: Box probably didn’t mean all models are immoral. He likely used the word wrong to mean incorrect, inaccurate, imprecise, incomplete, imperfect – and so on. (Unbelievably, it seems necessary to point out that we can use the words right and wrong non-morally.)

2 A model can be said to be wrong only if it makes sense to say it could be right. But what would a model, map or description that is right – correct, accurate, precise, complete or perfect – look like? How much and what kind of information would it have to contain? The absurdity of these questions exposes the absurdity of the claim that all models are wrong.

3 We could not model a reality or map a domain to which we have no access, or of which we have no knowledge or information. In that case, any model or map we produce would be a fiction or fantasy, and its usefulness would be entirely fortuitous.

4 The solipsistic claim that we know or can know nothing about what we call reality is an affectation exposed at every turn by performative contradictions, of which the use of language to express the claim is merely one.

The fact is that all models are models – full stop. They are not and cannot be the things that they model or describe, which are features of reality. A factual assertion and its truth-value – and any assessment of ‘rightness’, accuracy, precision, completeness or perfection – can exist only within a descriptive context. So the claim that all models are wrong is incoherent.

To say we have no objective standard by which to assess how well a model describes reality, or that we can’t know if our claims are true or false, is to misconstrue the actual relationship – and radical separation – between a description and the described. The myth of propositions, the JTB definition of knowledge, correspondence theories, and truth-maker/truth-bearer ideas all demonstrate the conflation of what we say with what we say it about – as does the course of foundationalism – its existence and rejection.

All posited foundations for what we know – and (therefore) the truth of what we say – are merely models. What we call facts are such models, and they constitute the objective knowledge we express in language. We build and repair this knowledge on foundations and with materials of our own making. But that does not mean the edifice has no foundation, and so must be shaky. That we can always say more does not mean we can never say enough.

A model can only ever approximate.
In this sense, even the right ones are wrong.

[Peter Holmes]

Hi, Sculptor1.

Why is the claim 'snow is white' only an approximate description of the relevant state-of-affairs? What would an accurate or 'right' description be like? Could there be such a description?