What is creativity? And can LLMs help?
Someone said I was “very creative” this week - I don’t know if I am or not, but it started me thinking about what creativity really is, and what parts are easy/hard with LLMs.
Creativity is some mixture of randomness (noise), combination/juxtaposition, craft, iteration, and probably the deep knowledge you need to both look in the right places and recognize when you have something good.
Some of these are easy for anyone, if they want to do them: anyone can put random things together or make small random changes to something existing. Usually the results aren’t great, but it’s fairly easy to come up with things like “Star Wars except it’s all ducks”.
Some are amenable to LLMs - iteration isn’t hard to do, with a little bit of code. I did an experiment once called “Make Creative” which consisted entirely of this parameterized prompt, which was called iteratively with it’s own output 10 times: “Add or change an element of the following idea: {idea}”. It gets weird fast.
Recommended by LinkedIn
Craft is a bit harder to pin down. It’s easy to find random things to combine, but harder to have the taste to combine things that result in good work. Perhaps this is also a job for iteration - isn’t that how we do it? Try something, see if it’s good, tweak it again, rinse and repeat? Maybe few-shot examples of “good” would be helpful here. That’s not universal but neither is creativity - acceptable for a kid painting in their room is different than acceptable for an artist displaying professionally or producing professional goods.
I think creativity is a more mechanical process than we like to admit. Certainly, many of the images that have come out of things like DALL-E in the past year would have been rated as “highly creative” by our own selves a few years ago, before we’d heard of these programs. Now we like to discount them as not being particularly creative, for a bunch of reasons, but they’re getting even better.
I suspect the line will continue to move. Creativity seems to be in the “I know it when I see it” category, but the reality is that all creation is the output of the mundane processes described above: take something and combine, mutate, iterate, elaborate, juxtapose or use some other repeatable technique to get out into the unknown, until you have something new enough and good enough. Just like other “no way a computer can do that” tasks that are now routine, I suspect creativity will be redefined and expanded over time as we learn to automate and improve parts of it.
Principal Software Engineer at Comcast
10moTo me creativity is a scouting mission and LLMs help me move faster through unknown territory 🙂🙂
Technical Documentation Pro -- Retired
10moStarting from my working definition of information, creativity is easy to define as a unique combination of existing affordances to produce a new affordance. I think that gets at your concerns about "craft". The directing force in this is affordance. And affordance is ultimately biological. The imperative of life drives you to recognize and develop affordances. Sure, creativity is mechanical in lots of ways. Cutups a la William Burroughs or David Bowie come to mind. But as it is with anything LLM, the heavy lifting is done by the living organism that recognizes an affordance in the result.
Senior Manager Technology Risk, Business Technology & Innovation | Cybersecurity | IT leader | Yale School of Management | SAFe Agilist
10moInsightful article Sam. will the LLMs commoditize creativity sooner or later ? ie the plane/threshold once reached deemed as good enough output , become constrained by what LLMs deem as a successful/likeable output of mundane iterations ? Eureka moments of creativity - not sure if LLMs can synthesize those for human experience or perhaps human perception of what is creative may be tethered ever closer over time to what LLMs are capable of
Senior Manager Technology Risk, Business Technology & Innovation | Cybersecurity | IT leader | Yale School of Management | SAFe Agilist
10moInsightful article Sam. how would one extrapolate this thought framework onto discoveries such as fundamental discovery of zero ?
Engineering & Product Leadership | Entrepreneurship | Software Architecture | AI & Cyber-security Expert
10moI think 'creativity' is to broad term to practically answer how it's will be affected by LLMs. We should scope that question to more specific types of creativity like the one mentioned with image generation... And I think we will get different answers per each specific type of creativity. But over time we will be left with much fewer areas where human creativity can compete with technology. Seems that music creativity is very close to that already....