BEHIND THE SCENES
Artificial Intelligence
Artificial Intelligence
This is an experiment from late 2022 using AI in multiple areas of development - artwork, code, creative writing and music. Technology is moving fast with various tools (OpenAI, Copilot, Midjourney, Dall-E, ChatGPT, AIVA, etc) and the implications are met with mixed reactions: some people love it, some are afraid it will replace them, and others are horrified by it.
I thought it would be interesting to explore AI with the theme of horror in a lighthearted way and see what would happen if I made a hybrid project: partly created by AI, and partly by me as a human. Afterwards, I could evaluate if I thought it was more of a crutch, an enemy, or perhaps something else...
Here's a tiny glimpse of over 20,000 images I generated with Midjourney so far. From these, I photo-bashed numerous parts together to create even more new images.
I repeated this process over and over, for all of the artwork in the entire project - ranging from characters, cinematics, props, UI elements, and even the mouse cursor!
I tested Copilot to see if it could write useful code. The more I worked with it, the more I saw the results improve. I had to learn which kinds of prompts were effective, and which were too generic to be useful.
I used OpenAI Playground to test its creative writing ability (ChatGPT did not yet exist when I began this). I found that, as with writing code, the more I refined my input prompts, the more the output results improved. Shown here is OpenAI spontaneously creating new content (highlighted in green) from my prompt above, providing context for the AI.
Then I flipped things around and tested Copilot to see how it performed when instead of writing code, it was allowed to participate in creative writing. In this case, Copilot is shown creating player choices which allow more user interactions, and perhaps even extend the story branch in some way.
Likewise, I tested OpenAI's ability to write code. When given the same task over and over, it often creates different implementations. Again, refining my input (so-called "prompt engineering") was paramount. And by studying the different ways the AI solved the same objective, it helped me to see new angles that I had not considered before.
As I learned more about each AI mechanism, I realized I could extend the overall functionality of the experiment by creating a kind of feedback loop. For example, I took pieces of OpenAI's creative writing and used them as input prompts for Midjourney to create new images.
To keep the loop going, I then took images generated by Midjourney, and fed them to a mechanism which attempts to reverse-engineer written prompts from visual information. That created new text descriptions of the image, which and I again looped back into Midjourney as new inputs for even more interesting results.
Using the AI tools also helped me see new ways of designing and arranging assets and scene architecture. For a long time, I have heard the repeated mantra that machines are dumb, and people have to teach them how to think. In this case, I encountered something new - the machines were expanding my mind and sometimes teaching me new ways to think...
In summary, the best way I can describe the experience is that each of the various tools was a force-multiplier, and when combined, they are even more empowering. The quality of the output is proportionate to the quality of my input, so learning how to best communicate with the tools is critical.
I did not perceive any of the tech as a threat. It was quite the opposite - this was a very enlightening, liberating and worthwhile experiment. There are clear and tangible benefits, ranging from creative expression to time management, and dramatic implications to the critical "time to market" aspect of product development. The technology is here - and it improves every day. It is far better to embrace it than to fear it.