As someone who works in computational design and occasionally uses machine learning techniques, it’s super interesting! However, that article and the private research group (http://www.imagination-engines.com/) are weird…
Not in my books. For one, I can’t find much information on what they actually did. Did the AI scour Amazon, figure out from reviews people had issues with trays, independently decide to pursue tray designs and somehow independently came up with what we know as fractals? My gut feeling is the AI was pretrained to score stackability and it was able to play around with a few parameters on a pre programmed fractal generator and optimized from there…
One of the first physical objects that gets cited as being designed by AI are extremely high performance “evolved antennas” that were designed by NASA and some affiliated academics with work beginning in the early 90s.
They were designed through genetic algorithms. While that’s on the dumb side of artificial intelligence it does still fit most definitions. I think a key part of the outcome is that it wasn’t simply a predictable optimization. From what I understand, the engineers were surprised by some of the solutions and were able to get significant improvements over human designed antennae.
Great point to bring up, most of these discussion on the state of AI in relation to design comes down to a lot of semantics. I’m personally much more interested with the use and the outcome than trying to figure out which box AI fits in, especially since its such a moving target.
But to answer your question, I think our bias towards admiring human accomplishment is causing us to move the goal posts. Would anyone argue that say, Charles Goodyear didn’t invent vulcanized rubber because he did it by accident?
Even then, I think a lot of AI strategies, especially within the realm of machine learning would fit the deductive reasoning bill more than the “give enough monkeys enough time/happy accident” bill. Since they lack the context we may have, they need to churn through tons of data in order to gain an understanding but once that’s figured out they can predict a solution to any query within that realm of knowledge.
As for the current state of AI with respect to design, I see it as two buckets one of big picture problem and one of minutia.
In the big picture real, I see problems approaching the business and marketing side of design work. For example WeWork has been talking about how they are using machine learning to predict the number and size of meeting rooms to put in a new coworking space based on usage data from existing coworking spaces. Data science has certainly been a part of many corporations to help answer business type questions in the last decade, but I could imagine the type of questions being looked at becoming more design related. Though to quote Daniel Davis, one of the researchers at WeWork, Why are you looking at AI tools if your company isn’t even using basic statistics? It’s a great reminder of the hype surrounding AI/machine learning and that there’s a lot to be learned by just objectively looking at business data and questioning the status quo.
There has been a decent amount of work where AI has been trained to solve specific 3D tasks. Techniques around topology optimization come to mind. I’ve personally used it as part of a group where we trained an ML algorithm on determining optimal bracing angles for beams dependent on length and angle between the beams. We then used it to generate hundreds of braces for a space frame structure in seconds. Autodesk’s optimization of the office layout of their Toronto office is also an interesting usecase where their program is finding a balance between much more goals than we could parse as humans.
To the best of my knowledge, very few people are looking at neural networks being applied to the complete design process in 3D space as freely as we see it with done with images with recognition, style transfer and creative work coming from adversarial networks. Part of it is that 3D data is much less available than image data and the additional dimension adds a significant amount of complexity. While we don’t know what technical advances lay ahead of us, teaching aesthetics, manufacturability, usability of a product in 3D space, UX, etc. in a generalized sense doesn’t seem possible in the near to medium future.
Another interesting design avenue is how the integration of AI will change the products we design. The Nest thermostat comes to mind, the UX its so much nicer than a traditional thermostat because the thermostat has just a bit more understanding. I’m curious to see what else could benefit
Back to the article where I was saying the research group seems a bit weird. From the link in the article, you’ll get to Frequently Asked Questions – The Artificial Inventor Project where you’ll learn that their intention for patents belonging to AI really is that the patent would be to the owner of the AI. Which again would need to be clarified in the case of a company selling pretrained AIs to other companies that would operate them. From that site, you’ll get to http://www.imagination-engines.com/ which seems to be the core of the research group. Note that I can’t find a single worked through use case for their technology, to be fair it seems like they may be operating in the areo/military space. It seems like it’s a small team/possibly a single a person that has been around AI since it’s first wave in the 70s. The most noteworthy patent I can find is “Device for the Autonomous Generation of Useful Information”. It’s a super long patent filing and I haven’t been through the whole thing and don’t know the extent of the actual claim at the end but it seems to share some stuff in common with current generative adversarial network. This is the current type of neural net used in many of the style transfer, deepfakes, image enhancers, sketch to image, etc. I’m a bit suspicious that their play may somehow involve them getting profits from any patent filed by a group using an AI? I might have my tinfoil hat on.
Also found this gem of a video by the owner of this company: