ChatGPT Is Creative — There, I've Said It

Key insights of today’s newsletter:

• Todays’s AI systems, including ChatGPT, challenge the notion of creativity being exclusive to humans by demonstrating the ability to synthesize new ideas from existing knowledge.

• Ironically, the ones now facing the threat of automation are the white collar workers and creatives who were traditionally considered immune.

• Of course, being creative is more than ‘generating ideas’ and in many cases, AI still falls short in that regards.

↓ Go deeper (8 min read)

---

---

I’ve thought about this long and hard and I didn’t think I would be uttering these words anytime soon, but I am. I think ChatGPT is creative.

Now, you’re probably shaking your head. And I get it, I’ve been guilty of arguing the opposite side myself, more than once.

“Large language models are nothing more than stochastic parrots.”

“It’s just 0’s and 1’s.”

Instinctively, being creative should be the last thing computers are good at, right? It turns everything we think we know upside-down.

The narrative for decades was that the workers facing the threat of automation were the truck drivers, the factory workers, retail clerks etc. Who would’ve thought that instead the graphic designers, photographers, copywriters and white collar workers would be the first to be walked to the proverbial guillotine?

§

Large language model’s propensity to hallucinate is simultaneously their biggest flaw and their biggest feature. Without any hesitation it’ll pull sources out of thin air, engage in role play, and can even be tricked or persuaded to act against its own instructions.

But it will also just as easily organize, remix, blend, transform, or synthesize whatever thoughts or ideas you give it. It can even generate ideas from scratch.

In a recent post Automating Creativity,

Ethan Mollick

, professor at the Wharton School of the University of Pennsylvania, reviews three interesting papers (1, 2, 3) that suggest the following:

• AI can generate creative ideas

• The ideas AI generates are generally better than what most people can come up with, but very creative people still beat the AI

• The ideas that AI comes up with have more in common with each other than the ideas you’d get from a big group of people (i.e. the ideas are less diverse)

Mollick writes that it would be foolish for us not to include AI in the creative process, especially if you don’t consider yourself highly creative, and I tend to agree.

---

---

§

Now you might argue that AI cannot generate any new ideas, as it is humans (with the help of computers) coming up with new inventions and pushing the scientific frontier forward, and not the computers by themselves.

While that may be true, we also have to admit that most problems don’t require new ideas. Most problems have already been solved by someone, somewhere, in some shape or form. On top of that, most innovative ideas are two or more old ideas combined in a novel way. Because these large language models are trained on virtually all of the Internet, AI is very good at this. It takes what is there, adds a certain degree of randomness to it — en voilà.

There’s a classic Steve Jobs quote that goes something like this: “To me, ideas are worth nothing unless executed.”

Of course, being creative is more than ‘generating ideas’ and in many cases, AI still falls short in that regards. Even if it were to generate truly novel ideas, it would still take a human to go off and do something with it.

§

Back to ChatGPT. While it can’t do much yet, it can do some things. It can write and run code, create and look at images, and converse with you through voice and text.

So, without further ado, I would like to present you with an anecdotal piece of evidence. An exchange with ChatGPT that I had on Friday, 2nd of February. To test its creative execution powers, I focused on ChatGPT’s ability to take an idea and do something with it. To be clear: any attribution of creativity for the idea falls on me, the execution is all ChatGPT.

I asked ChatGPT to create “moral code”, an imaginary piece of computer code that describes some of the universal moral codes that we, as humans, hold dear in our societies, and that are at the heart of peaceful coexistence with each other.

Programmed to be helpful, ChatGPT responded:

“Creating a “moral code” in the form of computer programming is a fascinating and thought-provoking challenge!”

Within seconds, it generated a few lines of code describing four core positive values: kindness, honesty, respect, and personal responsibility.

class HumanBehavior:
    def show_kindness(self, towards_others):
        if not towards_others.feeling_good:
            offer_help(towards_others)
            listen_to(towards_others)
            encourage(towards_others)

    def be_honest(self, in_communications):
        truth = extract_truth(in_communications)
        if truth.is_difficult:
            deliver_with_compassion(truth)
        else:
            openly_share(truth)

    def respect_others(self, all_individuals):
        for individual in all_individuals:
            acknowledge(individual.rights)
            value(individual.differences)
            give_space_for(individual.voices)

    def take_responsibility(self, for_actions):
        if actions.have_consequences:
            admit_fault(for_actions)
            seek_to_repair(for_actions)
            learn_from_experience(for_actions)

# Implementing the Moral Values
human = HumanBehavior()
human.show_kindness(towards_others="everyone")
human.be_honest(in_communications="all interactions")
human.respect_others(all_individuals="every person")
human.take_responsibility(for_actions="own actions")

Even now, when I re-read the output, only one word comes to mind and that is poetic. ChatGPT beautifully captured the idea.

I then asked it to write some lines of code that reflects how people are hurtful to one another:

class HumanBehaviorNegative:
    def show_hate(self, towards_others):
        if towards_others.different_from_self:
            inflict_harm(towards_others)
            spread_negativity(towards_others)
            isolate(towards_others)

    def be_dishonest(self, in_communications):
        lies = fabricate_truth(in_communications)
        if lies.benefit_self:
            deceive_others(lies)
        else:
            hide_truth()

    def show_prejudice(self, against_individuals):
        for individual in against_individuals:
            judge_based_on_preconception(individual)
            deny_opportunities(individual)
            disrespect(individual.differences)

    def avoid_responsibility(self, for_actions):
        if actions.have_negative_consequences:
            deny_involvement(for_actions)
            blame_others(for_actions)
            refuse_to_learn(for_actions)

    def consequences_of_negative_actions(self):
        social_isolation = increase()
        trust_loss = increase()
        personal_unhappiness = increase()
        societal_harm = increase()
        return social_isolation, trust_loss, personal_unhappiness, societal_harm

# Implementing the Negative Moral Values
human_negative = HumanBehaviorNegative()
human_negative.show_hate(towards_others="certain groups")
human_negative.be_dishonest(in_communications="when convenient")
human_negative.show_prejudice(against_individuals="based on bias")
human_negative.avoid_responsibility(for_actions="own mistakes")

# Observing the Consequences
consequences = human_negative.consequences_of_negative_actions()
print(consequences)

Again, the execution is striking. My favorite part is the section describing dishonesty:

def be_dishonest(self, in_communications):
        lies = fabricate_truth(in_communications)
        if lies.benefit_self:
            deceive_others(lies)
        else:
            hide_truth()

Is that not creative?, I ask you.

It was even willing to run its own code and summarized its efforts:

[The] output illustrates an increase in negative outcomes across four critical areas as a direct result of showing hate towards others. Each increment signifies a detrimental impact in that area, highlighting how such behavior contributes to broader social and personal issues. This symbolic execution underscores the ripple effect of hate, emphasizing its damaging consequences on both individual and societal levels.

You can read the full conversation here.

While it’s obviously a silly program (it doesn’t do anything), the fact that ChatGPT engaged with me in this strange thought exercise with verve, filled me with a profound sense of joy and excitement.

It made me realize that, like beauty, creativity lies in the eye of the beholder.

---

Join the conversation 🗣

Leave comment with your thoughts. Or like this post if it resonated with you.

Get in touch 📥

Have a question? Shoot me an email at jurgen@cdisglobal.com