Note: This text was first published as an appendix to How I Use AI Tools Everyday; I decided to extract and expand on my current observations that LLMs are constrained to what has been and only humanity can navigate the final frontier.

Why AI Can’t Create the Future

“The reasonable man adapts himself to the world: the unreasonable one persists in trying to adapt the world to himself. Therefore all progress depends on the unreasonable man.” –George Bernard Shaw

LLMs derive their capabilities from ingesting copious amount of finite training data.1 The current wave of generative AI products are therefore constrained to a world shaped by this data. I will be the first to admit that I don’t see a linear path from our existing understanding and execution of LLMs to a so-called AGI event.

What I do know is that humans have a strong desire to create and shape the world around us. From the earliest cave paintings to the tallest buildings, humanity has advanced by asking “Why not?” Even in a world of perfect AGI, the universe is still expanding, and I trust humanity will be at the forefront of the unexplored frontier.

Perhaps this could even become the most lucrative societal position of all: data gatherer.

Standing on the Shoulders of Giants

Below is one of my favorite charts visualizing how expanding the entirety of human knowledge involves a nucleus of core knowledge and a focused approach toward the outer limits. With LLMs, my hope is that our ability to stand on the shoulders of giants is accelerated such that humanity has more time to expand the circle more rapidly than ever.

the illustrated guide to a phd

To prove my point that LLMs cannot envision a world beyond the context provided, I prompted AI to take the above chart and add itself to the process—well, I think the results speak for themselves: an output that is a jumbled slurry of the original image. The moral is that publication of new ideas will shape the future, and only then will they be incorporated into LLMs.2

Distorted version of PhD knowledge diagram showing AI's inability to expand beyond known data

This really highlights that I know nothing about getting a PhD or how AI would assist. But it does highlight that human contributions are still important to expand that outer boundary.

Significant Revisions

  • Feb 20th, 2025 Originally published on https://www.jsrowe.com with uid 4D4D7537-841B-49AA-8AD4-17FC73FE923D
  • Feb 18th, 2025 Draft created

Footnotes

  1. I wrote elsewhere about dataset limitations. ↩

  2. I acknowledge that Sam Altman feels that AGI will be upon us within 10 years and will replace even the need for humans. ↩