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

[Ra1zel]

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

You would first need to define what is a perfect speaker... that's not possible imo

 

[Emlin]

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

And schools in the UK...

Ministers face calls to explain delay in closing schools over concrete crisis

Labour calls for audit of handling of longstanding safety fears as government pledges to fund repairs expected to cost at least £100m

www.theguardian.com

 

[CapMan]

And schools in the UK...

Ministers face calls to explain delay in closing schools over concrete crisis

Labour calls for audit of handling of longstanding safety fears as government pledges to fund repairs expected to cost at least £100m

www.theguardian.com

Can’t believe we have schools that may collapse without warning - how far we have fallen!

 

[CapMan]

You would first need to define what is a perfect speaker... that's not possible imo

So, how would you start to define that goal
Size
Price
Distortion
Flatness of reponse
Directivity
What else ?

Could you ask the AI to take a best in class archetype and make it perfect ?

 

[CapMan]

Mushroom pickers urged to avoid foraging books on Amazon that appear to be written by AI

Sample of books scored 100% on AI detection test as experts warn they contain dangerous advice

www.theguardian.com

Maybe we just close this thread !

 

[pjug]

I just want a spam filter that works. If that can't be done then I don't have much hope for AI as a go to for advanced tasks.

 

[phoenixdogfan]

I'm sure it could optimize a design based on the parameters it's been given, but given that those parameters would not capture something that would be universally perfect for everyone, I'm equally certain it couldn't do that. It might, however, be able to design a speaker whose design parameters would be least objectionable to the greatest number of people by absorbing the psycho acoustic research and using speaker design and acoustic research to create something that linearly optimizes those qualities.

 

[Salt]

Back on topic:
an ideal speaker would reproduce a given signal from a source (amp or preamp) without distortion primary on axis and secondary without significant directivity.
Agree?

 

[001]

If we absolutely insist on function over form then, eventually given sufficient input, I believe ai could do this. However, apply this idea to cars and i quickly realise that's i do not subscribe to the 'function over form' design model. Quite the opposite in fact if I'm being really honest with myself. I don't think I'm alone in this. Witness the ai generated buildings previously mentioned for example.

 

[ta240]

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

It would need to be specifically trained. Just pulling from all the information on the internet would likely cause it to come up with some errors. So the results of just asking a current AI to design a speaker would be questionable. When AI is trained they feed it the information and then they give it a large set of questions and rate its answers. It 'learns' from that feedback what answers not to give in the future. At this point if the people rating the answers have a different idea of a 'perfect' speaker then the AI will consider those people's opinion to be fact and it will design your speaker based off their personal thinking. This is the point in the process where AIs develop biases.

 

[Salt]

Correct.
AI needs to be speficillaly trained, means feeded with reliable (independent!) data resources of drivers and, not to forget, enclosures.
In the latter may occur the clue.

 

[AudiOhm]

Every speaker/electronics designer thinks their design is the best...

Ohms

 

[coonmanx]

If you have ever been on Fakebook then you know why AI is a failure.

 

[AudiOhm]

There are no better book shelf speakers than the ones I designed and built...

Ohms

 

[Berwhale]

Can’t believe we have schools that may collapse without warning - how far we have fallen!

We had plenty of warning, it's the lack of subsequent action that is the problem.

 

[Berwhale]

It would need to be specifically trained. Just pulling from all the information on the internet would likely cause it to come up with some errors. So the results of just asking a current AI to design a speaker would be questionable. When AI is trained they feed it the information and then they give it a large set of questions and rate its answers. It 'learns' from that feedback what answers not to give in the future. At this point if the people rating the answers have a different idea of a 'perfect' speaker then the AI will consider those people's opinion to be fact and it will design your speaker based off their personal thinking. This is the point in the process where AIs develop biases.

You could swap 'it' for 'they/them' and 'AI' for 'human' in your text and it still makes perfect sense

 

[stubaggs]

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 ?

Here is an example of traditional AI/ML used to design an antenna. As the other posters noted the challenge would be to define how to evaluate 'best', but if we could......

 

[kemmler3D]

AI can only interpolate and not extrapolate

This isn't strictly true - as @nerdstrike hinted at, AI can randomly generate new "ideas" and simulate their performance, gradually combining and selecting among the best performers. It's an "evolutionary" type of algorithm, not sure if "AI" is the right word for it, though. There are examples of new modes of walking coming from algorithms like this.

IN THEORY you could come up with new tech or designs for drivers this way, but I think the simulation would be so demanding that you're better off just asking an engineer to do it.

 

[kemmler3D]

Here is an example of traditional AI/ML used to design an antenna. As the other posters noted the challenge would be to define how to evaluate 'best', but if we could......

Actually, if you just set an optimization algorithm at the task of "Build a speaker that gets a 10.0 preference score" you could probably get pretty far. If the task is "come up with a speaker that I like more than one with a 10.0 preference score" you're going to be spinning wheels forever.

 

[xyvyx]

waiting for BrunoGPT