In theory it's being already being done. I think machine learning has great potential to help out in the design/drafting space. I've been looking at how to implement it myself.
My current genre of design (landscape) is mostly 2D. After taking a machine learning class, I've had a difficult time figuring out how I could apply it to my designs. That doesn't mean it can't be done, it just means I'm not there yet. As with many problems, one of the hardest parts is figuring out what question to ask, and how to ask it.
My current thought on how to apply it to CAD looks something like:
Build a machine learning model- Using existing completed designs (dwg files) use LISP/.NET to generate hundreds or thousands of problem and solution datasets (probably text files, possibly images)
- Write some miracle machine learning code, (Python) and generate a really nice ML model as an analysis tool
Use that model for new designs- Using the same code in the first step , generate a problem set from a dwg file you want to design to
- Feed that problem dataset to the miracle model from step 2 and retrieve it's solution (along with it's accuracy)
- Read the solution and write it into the drawing via LISP/.NET.
- Review design and tweak as necessary. When it's what you want, feed it back into the model (as a problem solution dataset) so the next iteration is better.
The first two steps are going to be REALLY hard.
I agree with dgorsman in that there's a lot of marketing hype around it. I do think it will have real world applications within a decade. As far as I can tell, AI is mostly marketing speak and for now, machine learning is where the work is being done. The first applications of machine learning to design aren't about putting out perfect design, they're about helping the designers see and find patterns in existing data that were difficult to see before. Basically, machine learning isn't autonomous, it needs a human driving it.
Want to get started? Take some of the many courses online for machine learning. Install BricsCAD (runs in Linux,and can run LISP). It's going to be a lot of work, it's still early days for this. It'll take a while for the investment of time and effort to return. But
if when it does, I think the returns will be exceptional. Keep me posted on your progress.
If it's written in python, it's machine learning, if it's written in Power Point, it's AI.
[/list]
