AI Like It's 1930

Cite as:

Geoffrey Bilder (2025, June 24). AI Like It's 1930. Retrieved from https://doi.org/10.59347/5ehqx-91420

After reading another messianic sermon about how AI is going to accelerate science, I had an idea for an experiment:

• Build a modern LLM-based AI using only content from before 1930 (or another cutoff that gives you enough data).
• Restrict researchers interacting with the LLM to using prompts that do not include information, terminology, or ideas that have been introduced since the cutoff date.
• See if the LLM can help the researchers recreate, in essence, any of the major paradigm-shifting creative developments - scientific breakthroughs, inventions, art/music movements - that humans have come up with since that cutoff date.

In short, let’s test the idea that LLMs can be creative by building an anachronistically time-boxed LLM and seeing if researchers using the LLM can innovate outside of the box.

And let’s do the experiment before we make knowledge-production and creative work inviable career paths.