LEGO AND AI

[Yoggington]

I work in a related field. Not specifically AI, but automation of (parts of) people's jobs. e.g. if you spend an hour a day copy/pasting info from this spreadsheet into some system, then I'm the guy who writes a program that will do it for you.

I think the OP (beep boop) is speaking specifically about generative AI - which is exciting of course, but it's also at the peak of a hype cycle. There are a lot of marketing promises of it's potential to do ~everything~, whereas in reality a lot of these things are very difficult. @evank's advice to consider this kind of AI as a tool or an assistant is the way to go here. I will not be fully trusting any task completely to AI any time soon. Can it make suggestions for a Lego designer to noodle upon? Certainly. Hell I'll use it myself for MOCs. "Hey ChatGPT, give me a list of buildings you'll find in a medieval village", oh nice, I hadn't thought about a candle-maker, I can build from that..

Generative-AI is not the only kind though - that's just hot right now as an emergent technology - your ChatGPTs, or your Midjourneys (and all the big companies versions of the same). 

You can also think about applications of the far more mature types of AI, which also have a much higher confidence rating.

1) Classification-AI: So think things that need binary answers or classification. Is this a brick red or blue? Think of the image tool app that scans your pile of bricks and suggest MOCs that contain a lot of them.

2) Predictive analysis AI, or pattern recognition: This has been in use for at least a decade in housing, stock markets etc. I'd be surprised if this isn't already in use at Lego. You throw all the Sales data of the last year into this kind of model and see if some unexpected results come back. Example, if they found that sets with new animal molds sell triple than those without set without - they might decide the price of a new mold in this case is worth more than a new mold for a one-off minifig accesory, and start cranking those numbers up. Let it crunch the numbers on whatever data is available instead of manually trawling through that data yourself.

3) Conversion-AI is the other type I know a bit about, but I'm struggling to think of a use-case. This covers translation services and that kind of thing. It could be used in the Customer Services department to for example, scan incoming emails & phonecalls for tone & level of discontent to more quickly route these to the relevant department.

[Alexandrina]

Moral issues regarding AI aside (as some models are definitely unethical in how they're trained, especially image generative AI - and no, something being posted online does not give carte blanche for it to be used in AI models) it's just not good enough to supplant humans, nor is it likely to be any time soon.

I've made some use of it as a tool - I've previously generated images for tokens in an online DND game, and I did experiment with using one of the AI voice models to make guide lines for stop-motion, so I could time shots more accurately. The results are usable in these cases as a sort of 'scaffold' for the creative process; they're by no means good enough to publish, but they can be put in place as a temporary part of the process. (This goes especially for voices in stop-motion. My workflow means that it's not practical for me to seek out, record and pay my cast until post-production, so having a means of experimenting with alternative lines/takes during the production phase is helpful)

In some fields it's even worse. Take writing. It's a matter of two seconds to get an AI model to generate a few paragraphs of prose for you, but the prose is terrible. Sure, technically it's fine on a sentence-by-sentence level - but it has no sense of rhythm, so you get sentences of equal length, usually three to a paragraph. It has a pedestrian vocabulary. And there's never any nuance beyond what is explicitly put into the prompt. You'd get a similar outcome if you asked a group of Year Six students to write a sentence each, following specific guidelines of what to include in that sentence, then stitched together what they wrote (except the kids would probably be more creative here). Even the idea of 'AI as a tool to aid the creative process' isn't universally true. There are fields where using AI actively hinders you. I experimented with using it to structure a chapter of my book, but the results were so terrible that I didn't get any usable material from them, and actually made a net loss of time when you factor in the time spent to write the prompts in the first place.

When it comes to Lego, I don't think AI will get there. Even if you could get it to consider all the physical factors (how different bricks can combine with other bricks, weight, sturdiness of construction, aesthetics) you're limiting it to combinations that have been thought up by a human to program in. Just today on the front page I saw a genius combination of bricks to make a sloped wall which didn't involve studs or any conventional connection. I fail to see how AI could conceive of that. Not to mention, TLG would still have to build and stress-test the designs, and there are bound to be bricks which don't serve an obvious purpose (or any purpose at all; even if you find a way to get it to factor in legal combinations and stuff, that doesn't mean it's not going to stick a random 2x2 brick in an unseen place that doesn't functionally affect the build but is also entirely pointless)

[evank]

ZeroGPT.com says Cater1977's comment is 100% written by AI. I suspected it, based on the tone.

[icm]

I already reported that post as being from an AI bot.

[danth]

1 hour ago, evank said:

ZeroGPT.com

Whoa, that's a thing? Awesome. Just ran some of my own posts through it...certified human! 🙂

[evank]

This is off-topic to Lego, but yes, AI writing checkers are a thing. The one I used is among the most popular. They're all flawed, but when it comes up 100% you can be sure.

Edited March 2 by evank