F1 computational restrictions.

[Stu]

Social media and basic algos have had a far larger impact upon soc/pol/culture yet had about zero red flags raised while it was being spun up by state intel agencies and funded via inflationary mechanisms.

I had/have the same feelings regarding the ‘socials’, so after dipping my toe in to both of the ‘big’ ones, I am down to viewing a business only platform.

Anyway, back to F1 uses…
At what point does ML become AI?
ML is/can be used in combination with FEA to maximise strength & minimise mass within parts (uprights, hubs, carbon lay-up, etc), presumably the same can said (to a degree) with CFD (applicable within ICE, cooling and bodywork).
My understanding is that a ‘base’ design needs to be generated manually before the ML gets to work within defined parameters for this.
For AI to work it would only require the allowed parameters and required performance parameters; it would then design the part itself?

* use of word “only” is to refer to system requirements for AI & not an attempt at oversimplification of an incredibly complex concept

[maygun]

Social media and basic algos have had a far larger impact upon soc/pol/culture yet had about zero red flags raised while it was being spun up by state intel agencies and funded via inflationary mechanisms.

I had/have the same feelings regarding the ‘socials’, so after dipping my toe in to both of the ‘big’ ones, I am down to viewing a business only platform.

Anyway, back to F1 uses…
At what point does ML become AI?
ML is/can be used in combination with FEA to maximise strength & minimise mass within parts (uprights, hubs, carbon lay-up, etc), presumably the same can said (to a degree) with CFD (applicable within ICE, cooling and bodywork).
My understanding is that a ‘base’ design needs to be generated manually before the ML gets to work within defined parameters for this.
For AI to work it would only require the allowed parameters and required performance parameters; it would then design the part itself?

* use of word “only” is to refer to system requirements for AI & not an attempt at oversimplification of an incredibly complex concept

In theory, you can start from a random point (e.g. a random 3D shape of a part). If you can describe all of your constraints in a differentiable manner (regulations, weight, surface area etc), you can optimize this shape with a cost function (whatever you want to optimize).

If your cost function is differentiable and easy to calculate this optimization is easier. However, while not my speciality, I assume the thing f1 teams would want to optimize (the data that you would get from CFD) is not differentiable.
One option here would be using CFD in the loop to optimize shape, the ML model would predict a shape, then CFD give you a score, and then using this score you would update your model and try to find a better score. However, the current ML models are generally optimized with algorithms that need thousands of iterations, which means thousands of CFD calculations, which is not feasible.

The second approach would be learning an ML model that can approximate CFD. Here, given the history data of 3D shapes and CFD results, you can learn a mapping function that takes a 3D shape as input and output CFD results. Then you can put this function into the loop that I explain in the previous paragraph.

All of these things, you can do with current ML methodologies, so I am not sure what would be constitute as using AI in F1.

[Zynerji]

I'm more concerned about team management, race management, marketing, politically correct interview scripts, HR disciplinary actions, etc.

F1 will lose its human appeal very quickly if this side isn't made off limits to AI.

Let the generative AI go nuts on car design tho. We may actually find relevant breakthroughs that impact road cars...

[Zynerji]

https://www.pcgamer.com/microsofts-ligh ... oores-law/

This could be huge for the CFD guys...

[maygun]

Somehow very basic, but for those interested, an example of how AI/ML can be used for aerodynamic optimization.
https://www.tri.global/research/drag-gu ... generation

[taperoo2k]

My concern is that with the capability of the technology evolving at a Moore’s Law rate, it only takes one genocidal lunatic to create a massive global problem (nuclear weaponry had/has the same potential - it is just far less accessible).

Somebody will figure out how to weaponize AI in a way that becomes a threat to humanity. We might already be in an AI arms race in that regard, as the US and China are busy working on military applications for AI.

I know that I am looking at this very negatively, but with some things it seems to be fairly sensible to look at it from a ‘what is the worst outcome’…

It's called being prudent, AI does require to be regulated. But I don't think it will be or if there is regulation the regulations will have a lot of loopholes for places like Silicon Valley to exploit.

Even looking at it from a positive perspective (performing tasks where a high level of accuracy - zero failure rate - is important; what jobs do those people then move into?
You end up in a society where a universal income becomes a prerequisite to survive day to day, which is fairly dark rabbit-hole from a socio-economic/political perspective.

That's the question. What happens to people who lose jobs to AI. Will it mean mass poverty and revolutions triggered by that or will it lead to universal basic income that allows people to survive and pursue hobbies. In F1 terms will it see certain engineering jobs replaced by AI?

As far as AI and CFD goes? This is an interesting article about it.
https://blogs.sw.siemens.com/simcenter/ ... ai-in-cfd/


Anyway AI is pretty much in it's infancy, it might be decades before it reaches the level of a Skynet.

[Just_a_fan]

That's the question. What happens to people who lose jobs to AI. Will it mean mass poverty and revolutions triggered by that or will it lead to universal basic income that allows people to survive and pursue hobbies.

History shows that when systems become available that make making money easier for a few, the rest - who were often doing ok before - tend to end up getting treated badly and end up working for the few in poor conditions and for very little recompense.

A universal basic income would need to be funded and that means the few paying money out - and they generally don't like doing that.

Dystopian stories abound in literature and cinematography because humans have a long history of treating other humans very badly when power and money are in the hands of a minority of people.

[hollus]

Let’s not go into politics?

[DDopey]

This is such a topic where you need to get the terminology right. AI is the umbrella term, it covers anything from linear regression to Generative AI. I dont think F1 will benefit from Gen.AI to design new parts, there is too much innovation needed. Deep learning can have its place, although I think the variety in data from f1 will make it difficult to apply (but I might be wrong about datasets available). Regression and ML will probably already be applied a lot.