Do you think AI could be used by TLG to design minifigures and even sets?

I recently tried with some classic LEGO characters, like pirate captain, knight, robot and while the results are not quite there yet, they werent terrible either (SciFi characters looked the best for some reason)

 

What would be the point? There's nothing that AI could do design-wise that couldn't be done better or more effectively by a conscious human designer.

I suppose it can and will be used more and more in the future. The thing is to me anyway, half the fun is designing whatever I try to build. It's the same with electric bikes as well, what's the point half the fun is pedaling and it does away with any health benefits, kind of like going to the gym to work out and using self lifting weights.

Sure AI can crank out some cool looking stuff and surely can offer some inspiration but like I said in the other thread it's like an artistic representation of a artistic representation, so it gets a little funky with what it creates.

3 hours ago, Lyichir said:

What would be the point? There's nothing that AI could do design-wise that couldn't be done better or more effectively by a conscious human designer.

Disclaimer: I'm going to ignore the difference between AI and a sufficiently advanced algorithm because I don't think anyone really knows the difference anyway.

I think Lego definitely has some sort of software that analyzes human designs and suggests both changes to make them more stable/robust and changes to make them cheaper.

Beyond that, I think AI could be useful as inspiration or, if you want to talk pie-in-the-sky, something that could translate drawings/concept art into Lego. I mean, I would kill for something like that if I could use it myself. I'd love to imagine something and have an AI do all the hard work of figuring out the Lego design for me.

31 minutes ago, danth said:

Disclaimer: I'm going to ignore the difference between AI and a sufficiently advanced algorithm because I don't think anyone really knows the difference anyway.

I think Lego definitely has some sort of software that analyzes human designs and suggests both changes to make them more stable/robust and changes to make them cheaper.

Beyond that, I think AI could be useful as inspiration or, if you want to talk pie-in-the-sky, something that could translate drawings/concept art into Lego. I mean, I would kill for something like that if I could use it myself. I'd love to imagine something and have an AI do all the hard work of figuring out the Lego design for me.

I don't think Lego's software on that front is as complex as all that—they definitely have more advanced building software that lets them more quickly and easily price out what parts they're using, but as far as I'm aware most of the reworking for cost is still done by a human designer making conscious substitutions and stability adjustments are still made based on physical testing. I don't think that kind of thing could or should be automated, as ultimately things like the physical properties of models are easier and cheaper to test physically than they would be to accurately or reliably simulate.

I also think AI's value for inspiration or concept work is generally overstated, since all it's doing is regurgitating the data it's been trained with to try to respond to a prompt. Actual design and research techniques are more valuable and make it easier to avoid unconsciously referencing or even plagiarizing existing art (a massive ethical issue in the field of generative AI that shows no signs of being resolved any time soon).

20 hours ago, Lyichir said:

What would be the point? There's nothing that AI could do design-wise that couldn't be done better or more effectively by a conscious human designer.

Massive cost savings. Also, I wouldn't underestimate AI. Lego is a very simplistic medium with low complexity relative to things like coding languages and algorithms which AI already excels at.

AI can basically can produce hundreds of designs, properly calculate the cost to produce, and provide analysis against market trends in a fraction of the time it would take current departments.

2 hours ago, LegendaryArticuno said:

Massive cost savings. Also, I wouldn't underestimate AI. Lego is a very simplistic medium with low complexity relative to things like coding languages and algorithms which AI already excels at.

AI can basically can produce hundreds of designs, properly calculate the cost to produce, and provide analysis against market trends in a fraction of the time it would take current departments.

Nah, I think you're overestimating it, if anything. AI, as it exists today, is a pretty deliberate misnomer for an overhyped technology. It relies heavily on plagiarism to generate sub-standard 2D images or text that has to essentially be babysat by an actual artist or writer if you want to ensure anything approaching quality. And it's nowhere near ready to extend that sort of thing to a complex and systematic 3D medium like Lego.

Even if there were cost savings to be had (and I doubt that there are), Lego has not gotten the reputation it has by cutting corners creatively. Perhaps a knockoff company would settle for the sort of creatively bankrupt shortcuts AI provides, but Lego has the resources to actually pay human designers who are better at creating quality products for human customers.

On 5/11/2023 at 12:33 AM, danth said:

I think Lego definitely has some sort of software that analyzes human designs and suggests both changes to make them more stable/robust and changes to make them cheaper.

I don't think so. If the shoddy stability check function in Studio is any indication, it is doubtful they have anything advanced. I would even doubt they actually run their models through the simulation tools in a CAD program. At the end of the day it's probably quicker to just build prototypes, given the nature of the task.

On 5/10/2023 at 9:08 PM, Robert8 said:

I recently tried with some classic LEGO characters, like pirate captain, knight, robot and while the results are not quite there yet, they werent terrible either (SciFi characters looked the best for some reason)

Current AI models are hugely dependent on the number of available samples. By that standard, even a few thousand official minifigure designs are a measly number compared to millions of regular images and if you figure in the fact that some elements have only been used on handful of figures at all, any AI algorithm will struggle to even properly "learn" how to use some of those bits correctly.

Mylenium

Edited May 12, 20231 yr by Mylenium

15 hours ago, LegendaryArticuno said:

AI can basically can produce hundreds of designs,

... of which you then have to manually sort out 99.99 percent. Your statement doesn't make any sense.

15 hours ago, LegendaryArticuno said:

Lego is a very simplistic medium with low complexity relative to things

May be so, but with a few thousand elements you still get trillions of combinations. And the point really is not whether a human or an algorithm can slap together a few hundred models. It still comes down to verifying those designs with simulations, human interaction tests and so on.

15 hours ago, LegendaryArticuno said:

properly calculate the cost to produce

You can do that already. It's called ERP and PPM software. Those may of course be based on their own AI algorithms, but they bear no relation to the actual design process.

15 hours ago, LegendaryArticuno said:

and provide analysis against market trends in a fraction of the time it would take current departments.

AI has no magical powers. If the underlying model is based on wrong assumptions and insufficient data, it would make wrong decisions. Duds like VIDIYO would still happen, even if the AI decides. And when it comes to such things I long have been of the opinion that it's much more important to have managers with balls and reliable gut feeling. All the analyses in the world can't avoid that someone somewhere has to make a few tough decisions.

On 5/10/2023 at 9:08 PM, Robert8 said:

Do you think AI could be used by TLG to design minifigures and even sets?

Could it? Perhaps. Generally, though, I feel that the task is much more complex than everyone seems to think. As per my previous replies to other posts my impression is that nobody gives much thought to how you actually would train the AI to produce functional LEGO models and then use that same approach to verify the designs. Those things are completely unrelated to synthesizing shapes and pictures with a generative AI. I also already commented on the minifigure stuff and to add to that, I don't really see the value of what an AI could do that an experienced designer couldn't do. At best I would see those AI-generated designs as inspiration or reference, but to really turn them into actual minifigs you'd still have to put in the extra work.

On 5/11/2023 at 1:17 AM, Lyichir said:

but as far as I'm aware most of the reworking for cost is still done by a human designer making conscious substitutions and stability adjustments are still made based on physical testing. I don't think that kind of thing could or should be automated, as ultimately things like the physical properties of models are easier and cheaper to test physically than they would be to accurately or reliably simulate.

Amen to that. That's why companies still build prototypes of huge machines, aircraft, cars, engines etc. that cost them millions during development. You can simulate such stuff all day, but you only know how it actually works once you put it in the physical world.

Mylenium

Edited May 12, 20231 yr by Mylenium

Imagine one day, you put list of bricks and AI will do best to make object from it. That day, I would be really scared...

10 hours ago, Mylenium said:

... of which you then have to manually sort out 99.99 percent. Your statement doesn't make any sense.

 

Yes, this is the real problem with artificial generated LEGO designs (whether sets or figure designs). The loads of rubbish that AI comes up with needs to be sorted out and discarded, before it can also iteratively learn from what it has generated. Unless they can come up with some formula or scoring function to determine what makes a great minifigure design or what makes a great set design, then that sorting would have to be done by human.

It will be interesting in 10 or 20 years time to see if AI can learn from brick placement (finding common patterns in combinations of two or three bricks or n bricks and the frequency those patterns occur) in official sets and design a set based on those patterns and a given volume / shape of an object, or description of an object.

3 hours ago, Ondra said:

Imagine one day, you put list of bricks and AI will do best to make object from it. That day, I would be really scared...

That is what they promise in the thread right below this one.

I used the free version of ChatGPT and told it, "Design a new Lego car from the Expert Builder series." It replied:

Certainly! Here's a design for a new Lego car from the Expert Builder series:

Model Name: Turbo Racer

Theme: Sports Car

Size: Medium/Large

Features and Functions:

1. Sleek Exterior: The Turbo Racer will have a streamlined and aerodynamic design, with a low-slung body, curved lines, and sporty details like a spoiler and air vents.

2. Opening Doors: The car will feature functional doors that can be opened and closed, allowing mini-figures to easily enter and exit the vehicle.

3. Detailed Interior: The interior will be carefully designed with a dashboard, steering wheel, adjustable seats, and a gear shift. Additional details like a stereo system and speedometer can also be included.

4. Working Suspension: The Turbo Racer can incorporate a working suspension system, allowing the wheels to move up and down, providing a realistic and smoother driving experience.

5. Functional Steering: The car will have a functional steering mechanism that enables the front wheels to turn left and right, allowing for more precise control during play.

6. Removable Engine Cover: The model can include a removable engine cover, showcasing a detailed engine block with visible pistons and other mechanical components.

7. Convertible Roof: Optionally, the Turbo Racer can feature a removable or foldable roof, allowing for both open-top and closed-top driving experiences.

8. Customizable Decals: The set can include a variety of stickers or printed elements to add personalized decals, such as racing stripes, numbers, and logos, giving builders the option to customize their Turbo Racer.

9. Additional Accessories: The set can also include extra accessories like a toolbox, a fuel canister, and a spare tire to enhance the playability and realism of the model.

Piece Count: The piece count for this Expert Builder set could range from approximately 800 to 1200 pieces, depending on the level of detail and complexity desired.

Please note that this is a conceptual design, and the actual availability and inclusion of specific features would be subject to the manufacturing and design processes undertaken by the Lego Group.

I tried the same query in Dall-E. It failed miserably.