Text-to-image AI generators like Midjourney, DALL-E or Stable Diffusion became popular and widely used in the past year. Although these algorithms have their limitations, feeding them with the right prompts can result in spectacular images. So much so that this one generated by Midjourney even won an art prize.
This presents so many fascinating questions and generates fierce debate. I, for one, don’t believe AI will replace artists, just as I don’t believe it will replace doctors. You can go on and experiment with these tools to find out for yourself how extremely hard it is to make the algorithm draw a picture like the one winning the blue ribbon for digital art. Finding these prompts is an art in itself which requires vision.
However, this is actually not the topic we’re discussing here. My initial quest for this experiment was to test whether these generators are able to come up with viable-looking product designs for a variety of digital health products. Can we use them to stumble upon brand-new ideas or revolutionary new concepts?
In short, the answer is “not very likely” as of now, but they might offer some ideas worth exploring. And, as it is explained here with some great examples, coming up with a sleek-looking sketch for a future product is only the first step of the design journey.
Let’s start with the good examples. I tested 11 ideas and Midjourney came up with 5 design directions that I think are actually better in some aspects than what we have on the market today.
- A wearable health sensor patients can wear on their chest
Looking at the output images, Midjourney imagined something along a 6 or 12-lead ECG line. The design is streamlined and you can imagine that some of these can be comfortable to wear even for a prolonged period.
- A smartwatch that can measure blood glucose continuously
I like these designs for several reasons – although they are not revolutionary as smartwatches. First of all, the algorithm came up with nice big displays, capable of showing a fair amount of info – which is an important point for such a device – and which is still not obvious looking at some of the new models debuting these days. Secondly, all these are non-invasive gadgets, which is definitely the way to go with a watch measuring blood glucose.
- A digital stethoscope that doctors can use to listen to cardiac sounds
These digital stethoscopes don’t look materially different from some of the existing models, but I like the little design touches: incorporating the heart shape is cool, and in the first three images it was done in a way that likely won’t hurt usability.
- A 3D bioprinter device that can print out living tissues
What I especially like about these images is that the algorithm was able to part from the thought of printing entire organs – which is the archetypical “bioprinting” image burnt into our retinas. However, the reality is much more likely to follow these images: printing tissues in a liquid (as opposed to on a scaffold).
- A fitness smartwatch that keeps patients healthy
Again, nothing revolutionary, but good, solid designs. Big displays, lots of info, futuristic look, you won’t be surprised for a second to see any of these on the shelves of stores. Of course, existing smartwatches offer a wide range of functionality and design features, and the looks are only the tip of the iceberg. Still, I like what AI did here.
- A headband that patients can use to measure sleep quality and improve it
Well, there are multiple problems with these. You definitely won’t measure anything regarding your sleep quality with a device sitting neatly on top of your hair, as do two of the designs. And sadly, none of these looks even remotely comfortable, which is a problem, trust me. Unrelated comment: I like how uncannily realistic the third person looks.
- A sleep sensor that goes under the mattress to measure sleep quality
Oh, boy. It is obvious that the algorithm didn’t have a clue about this one, there is a bed with a mattress for sure, but not much of a sensor. We see something in the second image, but it is eminently useless: it has the thickness of the mattress itself and almost the size as well. No way you can fit it under yourself and sleep on top of it. And the rest is even worse, and all those lights under the beds? Midjourney, go, ask your spouse about this and try again.
- A wireless patch sensor that can measure blood glucose continuously
Again, current sensors are much better designed than any of these. Midjourney’s output is bulky, way too large to be worn continuously, although the third image resembles something like a digital tattoo, which is cool. So maybe we have reached peak design in current skin sensors, and there is nothing AI can do here to improve these.
And The Ugly…
- A blood pressure device that can measure blood pressure continuously
A toaster? – asked several team members with a degree of uncertainty after seeing these images. I can’t fault them for not seeing the wearable blood pressure monitor in these designs. Here the algorithm entirely missed my point. To be fair with Midjourney, I also tested the exact same prompt with Open AI’s DALL-E2 and with Stable Diffusion and both significantly underperformed what we see here.
- The medical tricorder device in practical medicine today
Well, this is also not looking like a sleek 21st-century design of a multifunctional medical device. Although I like the classic Star Trek vibes, all of them are really old-school. They may remind you of good ol’ cassette players or grandma’s radio she used in the kitchen – definitely nothing like a breakthrough digital health gadget.
- A skin patch that can measure body temperature: photorealistic style, sleek design
These again are totally useless; they are way too big, and look unnecessarily permanent for a body temperature measuring patch – although one might argue for prolonged uses of such gadgets in clinical settings. Existing solutions are smaller and look more comfortable than these images.
Are we expecting to invent the square wheel here?
It is sometimes difficult to keep our expectations realistic regarding these new AI technologies. These examples highlight how un-obvious it is to use them for generating something truly unique. Novel ideas may surface this way, but you need to dig a lot for them. And even more often, if the AI comes up with something radically different, it is just not practical.
I also think that some digital health products and/or devices have already found their “optimal” design. Wearable blood glucose sensors for example seem to have matured into a highly functional form – maybe this is the end of the design road for them. Other devices – like a wearable blood pressure monitor – are yet to get there. I won’t be surprised to learn that companies start using these AI tools for collecting inspiration, but we are very far from AI designing our new devices.
The post We Used Midjourney To Design Digital Health Technologies appeared first on The Medical Futurist.