Moving Beyond Substitution: Imagining the Future Uses of Generative AI
One does not need to look hard to see the hype surrounding AI in 2023, and generative AI in particular. If the articles, press releases, and science fiction novels are to be believed, generative AI is poised to completely alter our world. Technologies like GPT-4 and DALL-E3 are easily accomplishing feats that just 2 or 3 years ago might have seemed decades off. These changes have — reasonably — sparked existential speculation about the future. We are being forced to confront a world in which many of the skills we felt were uniquely human can be done by computers in a fraction of the time. Chatbots, image generators, scriptwriters, music composers, etc…, AI has dominated visions of the future with worries that it will replace human jobs and creativity.
I asked GPT-4 to give me a list of the most impactful future uses of generative AI. Here are a few of the suggestions it gave:
- Personalized Education: Generative AI could be used to create personalized educational content and curriculums tailored to individual students’ learning styles, progress, and preferences.
- Creative Industries and Entertainment: From creating music, and writing scripts, to designing graphics, generative AI could potentially generate human-like content, revolutionizing the entertainment and creative industries.
- Automated Journalism: News articles and reports could be automatically generated by AI, especially for coverage of financial earnings, sports results, or other largely data-centric stories.
- Sustainable Architecture: Generative AI could be used to design sustainable and efficient building structures based on numerous variables like climate, location, available materials, etc.
AI is already being used in some capacity for all of these. There is no doubt in my mind that these are some of the most promising or current prospects for generative AI, but despite the relevance and likely truth of this list and those like it, they are all limited by the way that they imagine these use cases. The current scope of generative AI speculation seems to be limited to ways that it can accomplish tasks that previously relied on humans. These types of innovations are called “technology substitutions.” Substitutional use cases often present themselves much more readily as they fit into already imagined tasks and problems, but they rarely represent the full set of uses that mature technologies find.
I do not mean to diminish the significance of these use cases, as they have and will continue to make possible incredible new things that go beyond what a human could ever achieve, but this type of speculation I believe can close us off from another world of harder to imagine but equally impactful, “non-substitutional” use cases. By considering the inherent qualities of generative AI, we can create a sort of framework for considering its less obvious use cases.
Historical Precedent
This is not an issue unique to AI, but rather a manifestation of a pattern that has repeated itself in many of the revolutionary new technologies of the modern age. Take photography for example. Photography’s first popular use cases involved capturing the same things that painting and drawing had before it. Early photographers substituted photography for painting to capture the same subject matter that had been explored by artists for centuries prior, namely landscape, portrait, and still life. Imagination surrounding the new technology was limited by the possibilities afforded by the “previous” medium.
One of the landmark evolutions of photography came in 1878 when English photographer Eadweard Muybridge created the now famous “The Horse in Motion”, a collection of sequential photographs capturing the stride of a horse galloping. Muybridge created something wholly unique to photography. He took advantage of photography’s unique ability to objectively preserve these moments in time. It required an understanding of photography as more than just a faster way to do what the painters of the day did. By treating photography as more than a substitution, Muybridge laid the groundwork for the first motion picture a decade later, a new medium that would have been near inconceivable and wholly impossible before the advent of photography.
So how can we apply this to generative AI?
We must move past the qualities of generative AI that make it powerful as a substitution for humans and other technologies, and try and conceive of its inherent properties. “The Horse in Motion” was possible because of an understanding of photography’s ability to almost instantaneously freeze and preserve moments in time, and separating it from what styles and forms emerged from the properties of painting or drawing. This understanding of this unique property of photography could have been found by looking at the differences between photography and paintings created by humans, and as such, one could potentially do the same with AI.
For the same reason that AI is and will continue to find so many substitutional uses, it is particularly difficult to imagine what its non-substitutional uses will look like. Machine learning has been and will continue to be so disruptive and revolutionary because it does one of the things we hold most essential in our identity as humans: thinking. This is different from photography, which just challenged the abilities of a small subset of humanity that could paint realistically. To imagine AI not just acting as a substitution for human ideation, creativity, reasoning, etc… is in many ways to imagine an AI that goes beyond the human mind itself, which is unsurprisingly difficult to imagine.
Perhaps non-substitutional uses could emerge from the fact that the neural networks that power generative AI work by extrapolating patterns. Studies suggest that much of human cognition is similarly based on grasping and extrapolating patterns. A difference comes from the ability of AI to hold far more information with much greater fidelity than even the smartest human could. Uses could come from this ability to interpret patterns in media or research without the limitation of the limited scope of human perspective. AI also differs from human cognition in that, as a computer program, it can be readily edited and changed. Not limited by the fragile and immutable nature of a flesh and blood brain, we could find use in the ability to work with cognition and imagination that is malleable.
In Conclusion…
If the above paragraph doesn’t make it clear, the already wishy-washy nature of tech speculation is only magnified when imagining in this way. We can’t rely on the proven necessity of existing use cases that substitution can. I think it is unlikely anyone can accurately predict what entirely new use cases will emerge for generative AI. Nonetheless, I believe that an understanding of generative AI’s inherent properties and capabilities can help our imaginations avoid being limited by the needs of the present and look into the possibilities of this new technology.
