Is ChatGPT Creative?

AI isn't creative, right?

The commonly held view is that AI cannot itself be creative. It can mimic or replay creativity pulled from training data, but it cannot recognise or produce creative thoughts or ideas on its own.

In the following article, scholars in science and technology consider creativity to be "the capacity to trace links in unconnected elements" :

“Scholars in science and technology studies consider the capacity to trace linkages between heterogeneous and previously unconnected elements as a distinctive human social activity,” Oleinik writes.

Oleinik continues to say AI will not meaningfully change or rewrite the imposed rules.

However, that doesn’t mean that neural nets aren’t excellent mimickers of creativity. “In the words of a sociologist,” Oleinik writes, “a robot powered by neural networks may be a good actor, i.e. someone who closely follows the script, but not a subject, i.e. someone who meaningfully changes and rewrites the imposed rules.”

Lets just make sure

Well, ChatGPT is performing so well in many areas and many views on AI's ability to be creative were pre ChatGPT - so why not try and test it for creativity just to be sure?

So to test creativity of a machine that effectively has the entire internet in its brain, we require a creative idea where the probability of someone else having already documented that idea online is extremely unlikely or impossible.

So to do that, I used an idea that I think only a few (if any) in the world would have also considered - as it sits at the intersection of law and computer science (the field I'm in). Furthermore, I researched various related search terms on Google to see if anyone had written about the idea (with no results).

The idea

In software engineering, loose coupling means elements of code should be weakly associated and have breakable relationships.

In law, severability means the remainder of a contract's terms remain effective, even if one or more of its other terms are found to be unenforceable.

By connecting these ideas across two domains - one can learn from the other.?

For example, severability states that "if any area of the agreement is declared invalid or unenforceable, the remainder shall remain enforceable." but the word area is vague. To ensure you can cut away the invalid element of the agreement, lawyers might want to follow the software engineering design pattern Single Responsibility Principle - which is the idea that every element of code should only have one responsibility.

So, I asked ChatGPT to tell me whether it thought there was a relationship between loose coupling in software engineering and severability in law.

I was astonished with the response.

"There is similarity in the sense that both concepts refer to the idea of separating or isolating individual parts of a larger system"

The machine hasn't learned this from training data (there is none!). If you Google these concepts, no one has written about them in a related context, but the technology behind ChatGPT is aware of the relationship.

Going back to the definition of creativity "the capacity to trace linkages between heterogeneous [diverse] and previously unconnected elements." - I think I'd have to argue that ChatGPT was able to do this.

The ability to understand context is the ability to create

The creativity, the innovation, the hard bit, the lightbulb moment - comes from the light that goes off in your head when you learn a new concept and it triggers a memory "this is similar to something else I've seen". Once that link has been created, learning the concepts across the two linked domains to then find relationships and forge new ideas - well in my view, that's the easy bit.

The connection is the hard bit. It requires a lot of diverse reading, a good memory and a creative link to be made within short succession of reading (and understanding) the connected domains. That is why humans who are able to do this are the most rare and often the most successful (insert name of any inventor).

ChatGPT the inventor!

So I then thought ok ChatGPT - taking the context of my question, give me an idea from another connected domain :

Interestingly, architecture and biology are fields that some are bringing together under the term biomimetic architecture?however I haven't been able to find anything online around the modularity relationship. You'd have to assume there would be things that one could learn from the other re modularity.

Summary

The word Isolation is the bridge that connects loose coupling and severability.

I think its extremely impressive that ChatGPT is able to calculate that.

I think the machine is about to outperform human beings on one of the most valuable human abilities - creativity.

What is rather scary is that it also has the ability to connect every domain.