DesigningwithExtendedIntelligence

Generating Images from prompts

One of the first exercises we did was taking an AI system create through Dottod.net. In this system you take a picture and that is then created into a visual description that you can not see. Afterwards you can choose to generate a new image using a prompt. This was possible via 3 setting: technical, abstract or poetry. The fun thing with this is that I began to realize the impact of writing a strong prompt could bring. As it had a first decription and we were adding additional information it meant that it didn't always that the description you wanted. Also we noticed descrepencies with the image created of two people. My partners portrayal varied from a white man, to a black man to a woman. Yet the other persons portrayal had much less variation. It was funny to see that the AI could have such a large descrepencie with this. Below you can see the generated images + the original photos description based on AI + the prompt given

Rapid Prototyping AI

We also got to play around with a tool by Chris and Pietro in which we could rapidly prototype and AI system. Oliver and I are working together on our thesis project and were really interested in seeing if we could simulate a part of the interaction. The goal was that the data would be inputted to the AI ,ideally through a micro-controller, but here we did it as a text block. Furthermore it would get the input of a text message asking for energy reduction tips. We did it in such a way that the response was still quite serious but after presenting I am also curious to explore different ways of adding personality to our bot to make the interaction even more natural. For now I did a small additional test when adding a line about its personality. In the end we got the AI to do what we wanted and while we want to continue to use AI in the future, we most likely will be using an AI API in our system and then using the prompt form to create the personality we desire for our bot.

Reflecting on Designing with Extended Intelligence

This seminar has raised quite a few questions for me and also opened my eyes to the possibilities by using AI. I had some experience using generative AIs in the past and have always been curious to know more, so this course was perfect.

I think for me the best part was using ModMatrix to quickly in a no-code environment be able to test ideas that were in my head and see the possibilities. I feel that while I will continue to experiment with AI and using them in my thesis project it might not be by using ModMatrix. As I am currently working in NodeRed together with Oliver it might make more sense to use the ChatGPT API directly from that environment. What I did find extremely helpful to learn from Chris and Pietro was the way you can either create a custom GPT and then call that API so we could also create a personality for our AI that people then could download and use at home as well. I feel that I am now more understandable in the ways that one could just go out and try AI and use it in projects in a variety of ways that doesn't require a software background Which was the way I saw the future of our project going, so am excited to see if we could create similar results with the tips and tools we were given.

While it was fun playing around with the AI tools, for me a lot of my learnings from this seminar also came from the discussions we held about the ethics of using AI. There were two main points that really opened my eyes to the reality we are living in and the potential issues we could face quite quickly. The firs question raised was about the intellectual property and for me this wasn't just the question of who owns that art, the prompter or the AI, but also say in a business context. What are the problems that might arise if while working on a design, a designer uploads a sketch to generate a model, this sketch is then of course stored and used to further the learning of the AI. But could this then cause the business to not have the rights to their own designs? Another interesting thing that I will continue to think about during the coming weeks is the way that AI generations are becoming so realistic that it is becoming hard to tell real from generated. This could have large effects on society as we have grown accustom to using videos/images as proof but as this technology grows we discussed how can we create watermarks or ways to tell them apart? In the EU I learned about the laws they are currently working on but if we look deeper at what is being used now to tell apart the videos, then a thing like a mark in first layers of generated video can easily be scrubbed and removed. With this in mind I feel that as a society we must consider ways to solve this issue quickly before it becomes a problem at the judicial level.

Finally, in my thesis project we wanted to use AI and algorithms but an important point was raised during our presentation, when is the energy expenditure of the AI/system more then the question? Is using this technology really the way further? I feel this is something I had considered before, however I got a bit swept up in the amazement of the possibilities that I didn't really stop and think about this. I feel that it does still make sense to us the system we want, however it might be good to build in a small loop that if the questions answer will cost more to answer then it will save then it might not be worth it in the long run. I feel that this part of my research is something I still need to look into more, and also really understanding the difference of using a small scaled AI versus using an API call to a larger model.