Designing with AI
It is a well known fact that the rapid growth of artificial intelligence has quickly brought moral questions and debates over whether or not AI can actually create new things, or even more importantly, if any person in the creative field should be worried that AI is going to take over their occupations and artists or designers are going to cease to exist. Nevertheless, these more pessimistic views may be failing to grasp the concept of AI and human interaction.
For starters, AI is more than just a simple tool. In fact, I support the idea that it can be creative. From my experience using the diffusion and GAN models over the past few weeks, there have been a few images generated by these models that have surprised me and that have made me realize AI has some sort of degree of independence. For example, in the first GAN model I made, I wanted to replicate the art of making Japanese Hina dolls in a way that represents how all the original dolls have a similar and consistent structure, yet they differ in a few details such as head size, clothing, and ornaments. Thus, after taking more than 400 images of Hina dolls, I was able to create a model that consistently generated images of dolls that have a recognizable and consistent structure, but some dolls may have different color clothes, others may have a different body frame just like the traditional Hina dolls. Hence, looking at both the GAN model and the traditional art together, given enough training data, it is difficult to tell apart whether a doll was created by a human being or by AI. In this way, it seems as if AI had some sort of degree of independence in creating art, making it more than just a simple tool.
Nonetheless, I wouldn’t go to the other extreme and say that AI could replace artists in the future. With the previous example most people would be able to tell whether the doll was created by a human or not, yet compared to other models with more than 6000 images of training data, these models make it hard to tell whether the image was created by a human or not. Nonetheless, without the proper training data and inputs, the GAN model wouldn’t have been able to generate the desired results. In other words, there was some kind of human to AI interaction in order to generate the previous Hina dolls.
Perhaps a more clear scenario of this AI-human interaction would be a case where a product designer wants to create a piece of furniture, accessory, or object to sell to households, like let’s say, a desk lamp. Now, instead of designing any desk lamp, why not try to speculate about what a desk lamp would look like in a modern or futuristic concept. This is exactly what I tried to convey by first using Dall-e mini to create around 200 images of “futuristic desk lamps” and then combining the results of this diffusion model with other 200 conceptual images of modern desk lamps to create a GAN model that creates images of modern and futuristic desk lamps. Looking at the results, I then selected a few desk lamps that I found the most appealing and fitting to this theme.
Furthermore, even I myself was surprised by a particular desk lamp design that the GAN model generated: a suspending desk lamp. From the training data, most images followed a certain design pattern consisting of a head sustained by a support that’s connected to the base of the lamp. Nonetheless, the GAN model created a desk lamp that is suspended in the air right on top of the base. In this case, I thought to myself, what if some day in the future it’s possible to have a suspended desk lamp perhaps floating in a magnetic field on top of its base. This was an example that made me think that AI can be creative and produce results that change the typical model of how something should look like.
Now, taking this AI-human interaction one step further into the future, perhaps in 20 or 30 years, I personally think that services and the way products are served to the customer will experience changes. Other than the previous example where AI is used to create new products, I think that models such as the diffusion model could even be used to give customers and people the creative power to input the specific product they’re picturing in their head, and then manufacture what they have created and deliver it to their home. Thus, I personally visualize scenarios where AI is used to make more highly personalized and self-tailored services.
Lastly, I think the use of AI is going to be so pervasive that it will get to the point where everyone in school will have to learn the basics of what AI is and how it works. Other than using AI for services or creating new products, it’s already being used in the web applications we use, the music we listen to, the art we appreciate, etc. For this reason, it will soon get to the point that it’ll be mandatory in schools’ curriculums to include the basics of AI. Students from a young age will need to have a basic understanding of how almost all the things they’re using may have some interaction with AI and how they work.