How Human's Scientific Work Has Constantly Evolved and AI is the Latest Disruptor or Influencer?

“The Long Evolution of Scientific Work, and the Disruptive Arrival of AI"

by Steven Bates, Ph.D.

The latest generation of AI tools has already begun to be transformative to scientific informatics, and this may be only the beginning stages of an even more profound transformation of the nature of scientific work, and of the products of scientists’ labor. As those changes arrive, people are having to grapple with the adjustments in thinking needed to take the fullest advantage of this technology.

There was an old children’s book that I read growing up, Danny Dunn and the Homework Machine. The Danny Dunn series was from the fifties or early sixties and featured Danny and his two sidekicks and their genius inventor neighbor who introduced them to various sci fi devices. Picture the Machine as a mid-century mainframe computer. When Danny and his sidekicks had the chance to play with it they naturally tried to put it to use to do their homework for them. The punchline was that the effort they needed to make to learn the material so they could program the computer forced them to effectively do self-assigned independent study homework and absorb the knowledge they were supposed to in the first place.

Though they’ve invested far more time and passion into acquiring their deep knowledge, the modern scientist is in a similar position, and the lessons of this Cold War kid’s book are surprisingly relevant today. As AI tools percolate into their workplace, scientists are in a unique position to adapt to their use.

This lesson was brought home to me when I attended the Paperless Lab Academy conference a few weeks ago. https://www.paperlesslabacademy.com/usa/

The place of AI in laboratory informatics was a prominent topic in the presentations, and a theme emerged: laboratory scientists are in a unique position to maximize benefit from these new LLM tools.

As with Danny and friends, their domain knowledge is the secret sauce that can make them prompt engineering wizards, who know how to evaluate outputs to separate the gravel from the gold nuggets. Business analysts and products managers can help bridge that domain expertise with best practices for data governance, and software deployment and validation.

I recently saw a LinkedIn post the mentioned the idea of “semi-agentic” AI. This was based on the user experience, of the LLM apparently accomplishing multiple tasks independently to produce an output, but without too long a sequence of autonomous decisions without human feedback. Although I think the idea can be more precisely defined, I believe this approach will be the guiding spirit for how AI is deployed in the laboratory. With a data and AI strategy roadmap, human intelligence coupled with machine intelligence can open new horizons of scientific discovery.