We can redefine ‘hard to vary’, but we’d need still a working implementation in the form of computer code.

… Demeter scores 25% and axial tilt scores 100%.

Now do this universally, for any given theory.

We can redefine ‘hard to vary’, but we’d need still a working implementation in the form of computer code.

… Demeter scores 25% and axial tilt scores 100%.

Now do this universally, for any given theory.

Have some thoughts, which might be way off. But interested in your response. It seems to me that "hard-to-vary" is itself the criterion that a theory should be as programmable as possible. As you note, the goal of a theory should be to make it as explicit as possible, and a program is explicitness in its most complete form. Any theory with ambiguous components automatically has a breaking point that is changeable without detection. A programmable theory has strict causal relations all the way from the axioms to the prediction, which makes any change to the components detectable. In other words: a theory is hard to vary to the extent that its components and the couplings between them can be specified as a program. If a theory is vague, you cannot tell when it has been varied.

This gives a concrete operationalization. A breaking point is any place in the formalization where the chain stops being programmable: a primitive with no implementable type, a coupling between components that cannot be specified, or a step that requires implicit theories to fill the explanatory gaps. A mathematical theory with no remaining gaps has zero breaking points and is maximally hard to vary. A theory in natural language is already worse, because words carry ambiguity and vary from mind to mind. This does not rule out better and worse theories in natural language, since we can use more or less ambiguous words and relations. But it does create a hierarchy of hard-to-vary explanations, where the share of the explanation that is programmable, or at least unambiguous, forms the basis for the criterion.

This is probably too crude a formalization. But evaluating the two theories of Demeter's emotions and axial tilt as explanations, you could check how much of each is programmable. Detecting seasons is programmable in both cases through temperature and changes in weather. Demeter's emotions and the causal link from them to the weather, which is the entire explanation, are not programmable. In the axial tilt theory, every component is. So on this measure Demeter scores 25% and axial tilt scores 100%.

Have some thoughts, which might be way off. But interested in your response. It seems to me that "hard-to-vary" is itself the criterion that a theory should be as programmable as possible. As you note, the goal of a theory should be to make it as explicit as possible, and a program is explicitness in its most complete form. Any theory with ambiguous components automatically has a breaking point that is changeable which is hard to detect. A programmable theory has strict causal relations all the way from the axioms to the prediction, which makes any change to the components detectable. In other words: a theory is hard to vary to the extent that its components and the couplings between them can be specified as a program. If a theory is vague, you cannot tell when it has been varied.

This might give a concrete operationalization. A breaking point is any place in the formalization where the chain stops being programmable: a primitive with no implementable type, a coupling between components that cannot be turned into a function, or just a step that requires implicit theories to fill the explanatory gaps. A mathematical theory with no remaining gaps has zero breaking points and is maximally hard to vary. A theory in natural language is already worse, because words carry ambiguity and vary from mind to mind. This does not rule out better and worse theories in natural language, since we can use more or less ambiguous words and relations. But it does create a hierarchy of hard-to-vary explanations, where the share of the explanation that is programmable, or at least unambiguous, forms the basis for measuring the "hard-to-vary" criterion.

This is probably too crude a formalization. But evaluating the two theories of Demeter's emotions and axial tilt as explanations, you could check how much of each is programmable. Detecting seasons is programmable in both cases through temperature and changes in weather. Demeter's emotions and the causal link from them to the weather, which is the entire explanation, are not programmable. In the axial tilt theory, every component is. So on this measure Demeter scores 25% and axial tilt scores 100%.

It seems to me that "hard-to-vary" is itself the criterion that a theory should be as programmable as possible. As you note, the goal of a theory should be to make it as explicit as possible, and a program is explicitness in its most complete form. Any theory with ambiguous components automatically has a breaking point that is changeable without detection. A programmable theory has strict causal relations all the way from the axioms to the prediction, which makes any change to the components detectable. In other words: a theory is hard to vary to the extent that its components and the couplings between them can be specified as a program. If a theory is vague, you cannot tell when it has been varied.

This gives a concrete operationalization. A breaking point is any place in the formalization where the chain stops being programmable: a primitive with no implementable type, a coupling between components that cannot be specified, or a step that requires implicit theories to fill the explanatory gaps. A mathematical theory with no remaining gaps has zero breaking points and is maximally hard to vary. A theory in natural language is already worse, because words carry ambiguity and vary from mind to mind. This does not rule out better and worse theories in natural language, since we can use more or less ambiguous words and relations. But it does create a hierarchy of hard-to-vary explanations, where the share of the explanation that is programmable, or at least unambiguous, forms the basis for the criterion.

This is probably too crude a formalization. But evaluating the two theories of Demeter's emotions and axial tilt as explanations, you could check how much of each is programmable. Detecting seasons is programmable in both cases through temperature and changes in weather. Demeter's emotions and the causal link from them to the weather, which is the entire explanation, are not programmable. In the axial tilt theory, every component is. So on this measure Demeter scores 25% and axial tilt scores 100%.

Have some thoughts, which might be way off. But interested in your response. It seems to me that "hard-to-vary" is itself the criterion that a theory should be as programmable as possible. As you note, the goal of a theory should be to make it as explicit as possible, and a program is explicitness in its most complete form. Any theory with ambiguous components automatically has a breaking point that is changeable without detection. A programmable theory has strict causal relations all the way from the axioms to the prediction, which makes any change to the components detectable. In other words: a theory is hard to vary to the extent that its components and the couplings between them can be specified as a program. If a theory is vague, you cannot tell when it has been varied.

This gives a concrete operationalization. A breaking point is any place in the formalization where the chain stops being programmable: a primitive with no implementable type, a coupling between components that cannot be specified, or a step that requires implicit theories to fill the explanatory gaps. A mathematical theory with no remaining gaps has zero breaking points and is maximally hard to vary. A theory in natural language is already worse, because words carry ambiguity and vary from mind to mind. This does not rule out better and worse theories in natural language, since we can use more or less ambiguous words and relations. But it does create a hierarchy of hard-to-vary explanations, where the share of the explanation that is programmable, or at least unambiguous, forms the basis for the criterion.

This is probably too crude a formalization. But evaluating the two theories of Demeter's emotions and axial tilt as explanations, you could check how much of each is programmable. Detecting seasons is programmable in both cases through temperature and changes in weather. Demeter's emotions and the causal link from them to the weather, which is the entire explanation, are not programmable. In the axial tilt theory, every component is. So on this measure Demeter scores 25% and axial tilt scores 100%.

I'm not a programmer, so the code below is 100% AI-generated. But here is an attempt. If we normalize a theory into the parts that can be put on a computer, the types it uses, the nodes (specific values) it commits to, and the functions between them, we can score the theory by how many of those parts run.

from dataclasses import dataclass

@dataclass
class Item:
name: str
kind: str # "type", "node", or "function"
programmable: bool # does it compile and run?

@dataclass
class Theory:
name: str
items: list[Item]

1
2
3
4
5

def score(self) -> float:
    if not self.items:
        return 0.0
    ok = sum(1 for i in self.items if i.programmable)
    return ok / len(self.items)

def compare(a: Theory, b: Theory) -> None:
print(f"{a.name:<20} {a.score():.0%}")
print(f"{b.name:<20} {b.score():.0%}")

Demeter theory

demeter = Theory("Demeter", [
Item("Latitude", "type", True),
Item("Temperature", "type", True),
Item("Goddess", "type", False),
Item("Emotion", "type", False),
Item("Demeter", "node", False),
Item("emotionstate", "node", False),
Item("emotionat", "function", False),
Item("emotion_weather", "function", False),
])

Axial tilt theory

axialtilt = Theory("Axial tilt", [
Item("Latitude", "type", True),
Item("Temperature", "type", True),
Item("Angle", "type", True),
Item("Insolation", "type", True),
Item("axialtilt", "node", True),
Item("solarconstant", "node", True),
Item("solarangle", "function", True),
Item("insolation_at", "function", True),
Item("temperature", "function", True),
])

compare(demeter, axial_tilt)

Output metrics:
Demeter 25%
Axial tilt 100%

If we normalize a theory into the parts that can be put on a computer, the types it uses, the nodes (specific values) it commits to, and the functions between them, we can score the theory by how many of those parts run.

The program goes through each item in a theory, counts how many are marked True, and divides by the total. That fraction is the score of how hard the theory is to vary.

An item is True if it can be put on a computer, either by reusing an existing type (Float, Int) or by defining a new one that compiles. It is False if no working type system can express it. The user fills in the labels; the program just counts.

Demeter: 2 of 8 items program (Latitude and Temperature). Demeter, her emotions, and the functions linking them to weather can't. So the score is 25%.

Axial tilt: 9 of 9 items program. Standard types, measured constants, and functions from standard physics. So the score is 100%.

If we normalize a theory into the parts that can be put on a computer, the types it uses, the nodes (specific values) it commits to, and the functions between them, we can score the theory by how many of those parts actually run.

from dataclasses import dataclass

@dataclass
class Item:
name: str
kind: str # "type", "node", or "function"
programmable: bool # does it compile and run?

@dataclass
class Theory:
name: str
items: list[Item]

1
2
3
4
5

def score(self) -> float:
    if not self.items:
        return 0.0
    ok = sum(1 for i in self.items if i.programmable)
    return ok / len(self.items)

def compare(a: Theory, b: Theory) -> None:
print(f"{a.name:<20} {a.score():.0%}")
print(f"{b.name:<20} {b.score():.0%}")

Demeter theory

demeter = Theory("Demeter", [
Item("Latitude", "type", True),
Item("Temperature", "type", True),
Item("Goddess", "type", False),
Item("Emotion", "type", False),
Item("Demeter", "node", False),
Item("emotionstate", "node", False),
Item("emotionat", "function", False),
Item("emotion_weather", "function", False),
])

Axial tilt theory

axialtilt = Theory("Axial tilt", [
Item("Latitude", "type", True),
Item("Temperature", "type", True),
Item("Angle", "type", True),
Item("Insolation", "type", True),
Item("axialtilt", "node", True),
Item("solarconstant", "node", True),
Item("solarangle", "function", True),
Item("insolation_at", "function", True),
Item("temperature", "function", True),
])

compare(demeter, axial_tilt)

Output metrics:
Demeter 25%
Axial tilt 100%

I'm not a programmer, so the code below is 100% AI-generated. But here is an attempt. If we normalize a theory into the parts that can be put on a computer, the types it uses, the nodes (specific values) it commits to, and the functions between them, we can score the theory by how many of those parts run.

from dataclasses import dataclass

@dataclass
class Item:
name: str
kind: str # "type", "node", or "function"
programmable: bool # does it compile and run?

@dataclass
class Theory:
name: str
items: list[Item]

1
2
3
4
5

def score(self) -> float:
    if not self.items:
        return 0.0
    ok = sum(1 for i in self.items if i.programmable)
    return ok / len(self.items)

def compare(a: Theory, b: Theory) -> None:
print(f"{a.name:<20} {a.score():.0%}")
print(f"{b.name:<20} {b.score():.0%}")

Demeter theory

demeter = Theory("Demeter", [
Item("Latitude", "type", True),
Item("Temperature", "type", True),
Item("Goddess", "type", False),
Item("Emotion", "type", False),
Item("Demeter", "node", False),
Item("emotionstate", "node", False),
Item("emotionat", "function", False),
Item("emotion_weather", "function", False),
])

Axial tilt theory

axialtilt = Theory("Axial tilt", [
Item("Latitude", "type", True),
Item("Temperature", "type", True),
Item("Angle", "type", True),
Item("Insolation", "type", True),
Item("axialtilt", "node", True),
Item("solarconstant", "node", True),
Item("solarangle", "function", True),
Item("insolation_at", "function", True),
Item("temperature", "function", True),
])

compare(demeter, axial_tilt)

Output metrics:
Demeter 25%
Axial tilt 100%

So my criticism is that the HTV criterion is not a computational task (but a principle, universal statement) and Deutsch's criterion of understanding (you need a program) only applies to computational tasks.

With principle/ universal statement/ theory, I mean for example: for all masses, there is a force proportional to the inverse square of their distances/ for all integers, addition is commutative/ for all species, their evolution is governed by variation and selection, for all interpretations of moral actions, these are moral relativistic one/ ....

• Principles/ universal statements/ theories are not computable because they speak about sets of (possible) transformations (not 1 in particular which would be a computation) and they offer a constraining criterion to those transformations in the set.
• Whereas a computer program is an abstraction capable of causing 1 particular transformation (between sets of inputs and sets of outputs)

There may be a way to quantify HTV, and thus deal with specific evaluations of how HTV of one theory is higher than another. That would be a computational task. But that is different from the criterion for HTV (which is by definition not computable). And having no program for that computational task does not imply that the criterion for HTV is irrelevant or not usable, or even fluff.

Compare for example to the theory of evolution: the theory of "variation and selection" is the criterion for a set of allowable transformations (of species), but not having a specific program (e.g. for how a particular species can evolve in some particular niche) does not imply that the criterion is useless or fluff.

I think the usefulness of the HTV criterion becomes clear when you link it to Constructor Theory, then one can argue that HTV criterion adds more than criticisms alone can do. But that's a whole other story we could get into.

So my criticism is that the HTV criterion is not a computational task (but a principle, universal statement) and Deutsch's criterion of understanding (you need a program) only applies to computational tasks.

With principle/ universal statement/ theory, I mean for example: for all masses, there is a force proportional to the inverse square of their distances/ for all integers, addition is commutative/ for all species, their evolution is governed by variation and selection, for all interpretations of moral actions, they are moral relativism when ... applies to that interpretation/ ....

• Principles/ universal statements/ theories are not computable because they speak about sets of (possible) transformations (not 1 in particular which would be a computation) and they offer a constraining criterion to those transformations in the set.
• Whereas a computer program is an abstraction capable of causing 1 particular transformation (between sets of inputs and sets of outputs)

There may be a way to quantify HTV, and thus deal with specific evaluations of how HTV of one theory is higher than another. That would be a computational task. But that is different from the criterion for HTV (which is by definition not computable). And having no program for that computational task does not imply that the criterion for HTV is irrelevant or not usable, or even fluff.

Compare for example to the theory of evolution: the theory of "variation and selection" is the criterion for a set of allowable transformations (of species), but not having a specific program (e.g. for how a particular species can evolve in some particular niche) does not imply that the criterion is useless or fluff.

I think the usefulness of the HTV criterion becomes clear when you link it to Constructor Theory, then one can argue that HTV criterion adds more than criticisms alone can do. But that's a whole other story we could get into.

That's because a good explanation for Deutsch is not an explanation with good points, but an explanation that is harder to vary compared to any other explanation. So again relative to other explanations.

The word "good" is indeed misleading in that sense, but he clearly qualifies it as performing better, relative to other explanations, on his HTV criterion, and as the explanation having scored high points.

That's because a good explanation for Deutsch is not an explanation with good points, but an explanation that is harder to vary compared to any other explanation. So again relative to other explanations.

The word "good" is indeed misleading in that sense, but he clearly qualifies it as performing better, relative to other explanations, on his HTV criterion, and not as: the explanation having scored high points.