[1propionyl]

> Even in extremis: A fist-fight is a sequence of biomechanical optimization problems, and there's always a "perfect move" at any given moment in time.

No, it is not. And no, there isn't.

This is exactly the sort of reductive mode of thought the article is calling out.

[jstanley]

Just because you don't know what the best move is doesn't mean it doesn't exist.

[ysofunny]

there is only a perfect move in certain kinds of constrained "games", like a fight.

[jstanley]

Enumerate your possible moves.

Evaluate your personal utility function over all possible future timelines of the universe, conditional on each of your possible moves.

The move that has the highest score is the best move.

[1propionyl]

At that point the map is the territory.

[jstanley]

Yes. The territory has an optimal move. If your map is not good enough to identify it, that's a property of your map not of the territory.

[1propionyl]

No, it is not. First of all the territory isn't fixed. Second of all the existence of a fitness landscape (let alone ocean) doesn't guarantee the existence of unique optima, nor that the ones you identify aren't in a second or less your loss because your opponent read you. The other person's behavior is unpredictable but can be guided, and likewise. Feints are a huge part of fighting.

To think you can identify a model in this situation is pure hubris. Absolutely no one who fights thinks this way. Fighting is NOT LIKE CHESS.

This fundamental faith in modeling is dangerous. It overestimates its own applicability, and ignores its predisposition to only focus on the most available data (under the incorrect assumption that if you collect more and more it will eventually "average out").

[A_D_E_P_T]

You can literally model this.

Sure, there's a predictive aspect to it. What if your opponent zigs instead of zags, etc. But this is basically a matter of forecastable probabilities and can be added to your model. The optimal move still exists, no question about it.

Any problem of bodily motion through space has an optimal solution. In athletic situations, humans often can't think fast enough to find/utilize it, or aren't coordinated enough to move in the optimal way. And a biomechanically-perfect savant may still lose to an opponent vastly physically superior.

[strken]

I'd bet you $20 that none of the people who have ever won a fistfight have done so by modelling it as a biomechanical optimisation problem, at least on the fly while it was happening.

The comparison is unintentionally funny because it's the exact same "I can ignore the experience of the people who my work impacts because my models are perfect" mentality that produces unlivable apartments in dead lifeless streets.

[abc123abc123]

This is the truth! Added to the purely mechanical process, is an enormous psychological aspect. But... we can do even better! The above _assumes_ that fist fighting is indeed what we want to do, but perhaps the best (TM) solution is to avoid the fist fight alltogether?

Since ultimately, living, and living well, is about values, how do I choose to live, according to which values, science will never be able to capture that dimension.

I feel that scientists and technologists, and designers for that matter, should study more philosophy. It will open up their eyes to the fact that not every question is solvable by science.

[1propionyl]

Technologists in particular, taken as a group, have a very specific philosophical outlook that they don't tend to interrogate in themselves because it's so pervasive and intrinsic to what they work on and how they do so. Fish unaware of water, so to speak. It's a set of assumptions that make sense when you're programming software, but break down when applied to other things in the real world.

The tend to assume the universe is deterministic.

They tend to assume (incorrectly) that because it's deterministic a good enough model will be able to predict or explain.

They tend to ignore or not even be aware of the inherent bias towards available and measurable data, or that what we can measure must capture the essential dimensions of it.

The most naive tend to assume that given enough data, a model will get better, that the noise will "average out" (it doesn't).

I don't have a good name for this, but it has all the trappings of a good -ism otherwise.

Beyond philosophy, they should study art, music, literature, and whatever else interests. They should spend time with others who do and not only with people who work in technology. Unfortunately, increasingly college CS programs have cut out general education requirements in favor of questionably useful skills training, leaving graduates in a state where this seems daunting.

Computer scientists are building the world we all have to live in. Is it so much to ask that they be educated in the humanities before they're turned loose to do so?

[strken]

I mean, I think that the universe is close enough to deterministic at a macro scale, but even in a deterministic world you can still get smashed in the head by a falling brick with zero warning. Whether the the sequence of events that led to the brick falling down is deterministic or random is irrelevant to your ability to predict its descent. You only have a feeble human brain, a soft pink organ that's incapable of fitting more than 5 to 7 items in its working memory even before the collision.

[zaphar]

This is the sort of statement that is both true and also completely useless in a fist-fight. A fist fight is both biomechanical movement and also a mind-game comprising physical, mental, and emotional stamina.

[travisjungroth]

> You can literally model this.

Show me.

[A_D_E_P_T]

What would you like to see?

At this point, people are even modeling figures on Ancient Greek pottery to determine the biomechanical merit of their fighting stances: https://www.mdpi.com/2075-4663/12/12/317

The same or similar techniques, of course, can apply to any combination of fighters (or dancers, or swimmers, etc.) at any particular moment. At the highest levels of sport, biomechanics analysts are employed, e.g.: https://pubmed.ncbi.nlm.nih.gov/34402417/

In any case, I don't think that I made any extraordinary claims. There are a lot of unknowns, though, as the most valuable analyses tend to be extremely computationally demanding.

[travisjungroth]

> What would you like to see?

A model that shows the optimal move for a fighter at any point in time.

You don’t actually have this. It can maybe be theoretically done, but not in practice.

[A_D_E_P_T]

It can assuredly be done in practice, with currently available technology. It would, however, be very expensive and time-consuming.

I'm thinking of putting together a set of general biomechanical models for foil or kendo fencing. Both forms feature a highly constrained ruleset, which simplifies things. Hobby project, though, so maybe one of these days...

[user____name]

I don't know if this was your intention, but I read this in the voice of Morpheus.