Could AI design the ‘perfect’ speaker (or any other audio component !)

[CapMan]

I am engineer and work in technology, but I am not an electrical / audio engineer. I actually think I suffer from imposter syndrome in this forum

As the question says - could AI take everything we know and do better than a human designer ?

 

[fpitas]

Probably not.

 

[CapMan]

Why ?

 

[fpitas]

Why ?

Why would it be qualified? In essence someone or something has to teach it.

 

[nerdstrike]

Whether Newton-Raphson or genetic algorithm it amounts to an optimisation approach. Presuming we talk about realistic contemporary expressions of AI, rather than post singularity hypotheticals, then the novelty would only come from happy accidents.

If you cut loose with a large parameter space, it may only suggest crap, or iterate toward the contemporary bookshelf (the source of a great deal of training data?). How do you set fitness criteria? Will it just rediscover all the compromises we discuss here?

There is room for ML-driven structure optimisation, but then you usually need fancy 3d printing to make what it proposes. What do you get out of it? Maybe a nicer basket for the driver to sit in, maybe a stiffer cabinet or whatever. In any event, similar outcomes can be achieved with less buzzwordy perturbed regressions toward the goal.

 

[kemmler3D]

could AI take everything we know and do better than a human designer ?

IMO yes and no.

"AI" of the generative LLM type that's been in the news lately - no. It's not especially good for that type of work. It creates things that seem right (and by virtue of seeming right very accurately, often are right) but can't evaluate whether something is actually right or not.

"AI" of a "genetic algorithm / machine learning" type, which has actually been around for a few years, probably could. This is akin to multi-physics goal-seeking optimizer / solvers that are already available in packages like COMSOL. It could iteratively simulate a large number of possible designs and compare them to a goal performance spec.

This takes a hell of a lot of computing power but there's nothing stopping us from designing a speaker that way, in principle.

Why it can't actually design "the perfect speaker":

Speaker performance is already pretty good and is largely limited by the performance of materials that go into the speaker. The designs of drivers are pretty efficient and close to the limits of physics in many cases. Voice coils are an old technology that's quite mature.

Totally novel technologies could be in scope for a "genetic algorithm" of some kind, but the competition is very steep already, and the laws of physics don't change just because AI is involved.

Also, "the perfect speaker" is a subjective notion, in real life. What IS the perfect speaker? Tell us your definition and we can have a more detailed conversation about whether AI can achieve it. However, my prediction is once you define your perfect speaker, the thread will instead devolve into an argument about what "perfect speaker" means... it always does.

 

[fpitas]

Also, "the perfect speaker" is a subjective notion, in real life.

Bingo

 

[fpitas]

In any event, similar outcomes can be achieved with less buzzwordy perturbed regressions toward the goal.

The only advantage being you can charge idiots, I mean discriminating buyers, more for the perfect speaker.

 

[CapMan]

My sense is that AI can only interpolate and not extrapolate - so in this sense I agree with comments that it might optimise within a set of constraints and a target, BUT would be unlikely to create something new or unheard of !

 

[CapMan]

The only advantage being you can charge idiots, I mean discriminating buyers, more for the perfect speaker.

Harsh, but probably fair !

 

[fpitas]

My sense is that AI can only interpolate and not extrapolate - so in this sense I agree with comments that it might optimise within a set of constraints and a target, BUT would be unlikely to create something new or unheard of !

Yet you really could charge more. As a marketing exercise it's a good idea.

 

[CapMan]

Yet you really could charge more. As a marketing exercise it's a good idea.

There is enough foo foo in the world already !

 

[fpitas]

There is enough foo foo in the world already !

It might produce a useful speaker, albeit as you noted very derivative. So not snake oil.

 

[BlackTalon]

If AI could do it we would have come full circle, as the first loudspeaker that could produce intelligible voices was patented by Al (Graham Bell)

 

[Ron Texas]

The term "perfect speaker" is too vague to be a design target. However, maybe the AI could be asked to design a speaker for a price target with dynamic range of X, dispersion characteristics of Y, and not larger than a volume of Z.

 

[Doodski]

The term "perfect speaker" is too vague to be a design target. However, maybe the AI could be asked to design a speaker for a price target with dynamic range of X, dispersion characteristics of Y, and not larger than a volume of Z.

That would be cooL! A sooper computer that can strategize with calculus.

 

[Rednaxela]

Agree the idea of a perfect speaker is not very well defined. On the other hand, in every one of Amir’s reviews we are very well able to point out where a speaker deviates from a certain ideal.

Maybe a better task (that would still be interesting and valid) would be to ask AI to perfect a speaker given its existing implicit or explicit design goals.

 

[Doodski]

Agree the idea of a perfect speaker is not very well defined. On the other hand, in every one of Amir’s reviews we are very well able to point out where a speaker deviates from a certain ideal.

Maybe a better task (that would still be interesting and valid) would be to ask AI to perfect a speaker given its existing implicit or explicit design goals.

I was more turned on by it being able to strategize with any calculus and form stronger bullet points. Rather than being programmed to use a specific use case like speakers crossovers etc.

 

[Emlin]

Maybe it would be a good idea to ask the AI what its definition of the perfect speaker might be before asking it to design it.

 

[Berwhale]

As the question says - could AI take everything we know and do better than a human designer ?

I don't see why not, they are already using AI tools to design and build hotels in China...

‘It’s already way beyond what humans can do’: will AI wipe out architects?

It’s revolutionising building – but could AI kill off an entire profession? Perhaps not, finds our writer, as he enters a world where Corbusier-style marvels and 500-room hotels are just a click away

www.theguardian.com