Creativity is often defined by three criteria,

• The idea or product is novel (original) and,
• Has value or utility (usefulness),
• Within a specific context or task (task appropriateness)

AI can certainly devise novel works of music, art and writing.  It’s true AI has no cognition, emotions or experiences and therefore, its originality is limited to the patterns and structures learned from the data.

But the Old Testament reminds us, “What has been is what will be, and what has been done is what will be done; and there is nothing new under the sun.”

The meek borrow, genius steal and in this way perhaps, AI is genius and also, creative.

A more quantitative measure of creativity is called the Alternative Uses Test that prompts humans or AI to generate multiple, original uses for five everyday objects (pants, ball, tire, fork, toothbrush).

This test is used frequently to measure human creativity and has good predictive validity.  This chart shows how AI stacked up against humans.

Here are the take-aways:

• Green line: This is the top humans, the most creative per the AUT.
• Yellow line: The average for all humans
• Light blue line:  This is yesterday’s AI, GPT 3 that is slightly above the average human
• Pink line: Today’s AI, GPT 4. Much better than yesterday’s AI and only slightly worse than the best humans.

Many jobs will be lost from AI, more will be enhanced with productivity and efficiency boosts.

Today’s version of AI is still a 3 yr old on a tiny bike with training wheels.  The future is flying cars by comparison.  And the future is moving at an exponential rate.

There is an upcoming book recounting the experience of a person who’s both a data scientist and endocrinologist.  This doctor was given six months of pre-release, hand-on’s time with GPT-4 to trial it in medical use cases.  Recognize, medical students and residents routinely do Google searches now.

Among the many findings;

• GPT-4 gave a correct genomic diagnosis for a condition that occurs in one in one million people
• GPT-4 was used to write pre-authorizations, discharge summaries to edit and approve, recommendations for a clinical trial and review care plans for patients
• And last but not least, this quote from the doctor, “I’ve thought it would be pretty darn difficult to see machines express empathy, but there are many interactions that suggest this is not only achievable but can even be used to coach clinicians to be more sensitive and empathic with their communication to patients.”

Empathy?  Creativity?  If it walks like a duck and quacks like a duck…And remember this is a tiny bike with training wheels and GPT-4 wasn’t trained for the medical domain.

The training bit brings me to a final point for now but one we’ll continue to expound on; you have to train AI on your use case.  The seemingly endless recommendations and Top 10 lists of prompts to give GPT to achieve X, Y or Z are a distraction, at best.

It’s like giving me instructions on how to cook instead of a step-by-step recipe to make beef wellington.  The former will never get me there.

Same for asking GPT to write for you.  It needs to be trained on your style to start matching it.

Kevin