Scientifically Speaking | Can AI substitute for human creativity?

After asking a series of random questions, I asked ChatGPT a question that underscores human cognition — creativity. I typed in, “Can AI show creativity?” Here's what it spat out.

The story of Alceo Dossena is a sad one. By all accounts, a phenomenally gifted sculptor, Dossena created sculptures that were virtually indistinguishable from Renaissance masters such as Giovanni Pisano, Donatello, and Simone Martini. Museums around the world carried his creations, which they thought were original works of art from the Renaissance masters. But the deception had been carried out by his dealers, not Dossena himself. When Dossena revealed that he was the creator, no one wanted to see his work anymore. Dossena died in poverty.

I’m generally sceptical of all grand claims made for AI. But ChatGPT seems genuinely impressive. (Bloomberg)

It has been a little over a month since OpenAI launched ChatGPT, an artificial intelligence chatbot that generates responses to all types of questions in English. I have been thinking about the line between fakes and originals, and between creativity and imitation. And this year, an even more powerful version of ChatGPT is slated for release by OpenAI.

Does an imitation in the style of a creative genius ever qualify as a masterpiece? I asked ChatGPT the question.

“It is unlikely that an imitation that is in the style of an artistic genius would be considered a masterpiece, as it would not demonstrate originality or creativity on the part of the imitator. While an imitation that is skillfully executed and closely resembles the original work may be praised for its technical proficiency, it would not be considered a masterpiece in the same way that an original, groundbreaking work would be.”

This seemed like a well-articulated response. I nodded in agreement. But I also wondered what would happen if I repeated the query. This time the answer was a bit different.

“It is possible for an imitation of art created by an artistic genius to be considered a masterpiece, but it would likely be evaluated based on its own artistic merit rather than its similarity to the original. Additionally, the imitation would need to be extremely well-executed and demonstrate a high level of skill and creativity on the part of the imitator.”

Both answers are what we humans call “subjective” and fall in the squishy area between concrete answers to math sums and eureka moments of making creative connections between unrelated things.

I’m generally sceptical of all grand claims made for AI. But ChatGPT seems genuinely impressive. It is hard to know when the world is undergoing a transformation when you are witnessing it in real-time, but this does feel like a paradigm shift.

Ask a question on a search engine like Google and it will give you references that are sorted algorithmically that already exist somewhere on the internet. It is then up to you to research these sources and find the answer you are looking for.

Voice-activated personal assistants such as Siri and Alexa are much touted, but the novelty is not in the answers, but in the ability to decipher spoken language and deliver answers that already exist. As someone who uses both Alexa and Google devices daily, I can attest to their limitations too. They’re good for finding out the weather or directions or adding appointments to a calendar, but for more complex questions, they simply regurgitate what they can easily grab from the internet.

Even applications like predictive text (currently trying to populate my sentences as I write this paragraph) have major shortcomings and limited utility. I find them annoying and usually wrong, but perhaps that’s because I write long-winded sentences with strange construction, to begin with.

This is how ChatGPT is different. The interface is like a standard chatbot. How the answers are derived is not clearly known by the engineers who create the systems. There are also no guarantees that the answers won’t be wrong or trite.

But the answers created by the machine based on the queries will usually sound like an intelligent person who knows what they’re talking about. Coherent answers are generated by the AI relying on massive processing power and prior “training” on large datasets.

I’ve spent the past few weeks asking ChatGPT all types of questions (“What pathogens might cause the next pandemic?”, “How can we cultivate a healthy gut microbiome?”, “What are ways to limit climate change?”, “What is the meaning of life?”, and even “What are good topics for a science column for Indian media?”).

The writing has always been impeccable, and the ideas generated in areas in which I know something have typically been appropriate and meaningful. But human thoughts are messy and unpredictable. We argue and our views change over time.

So, I asked ChatGPT a question that underscores human cognition — creativity. I typed in, “Can AI show creativity?” The answer came out word by word on the screen:

“Artificial intelligence (AI) systems can be programmed to generate new and unique ideas, but whether or not they can truly be considered creative is a topic of debate.

AI systems can be trained to generate creative outputs, such as writing, music, or art, using techniques like deep learning, generative models, and evolutionary algorithms. These AI systems can produce novel and unexpected results, but they are still limited by the parameters set by their creators and the data they were trained on.

Some experts argue that because AI systems lack consciousness, self-awareness, and personal experiences, they cannot truly be creative in the way that humans can be. Humans have the ability to draw on their own personal experiences and emotions, which allows them to create something that is truly original and reflects their unique perspective.

On the other hand, some researchers and practitioners believe that AI systems are capable of creativity, in the sense that they can generate new and useful ideas or solutions, but it's still limited by the parameters set by their creators and the data they were trained on.

Overall, AI can produce results that can be considered creative, but the level of creativity and originality of the output depends on the complexity of the model and the quality of the dataset used.”

This text was spit out by an entity that does not understand language or meaning. But it mirrors my views on the topic. AI may or may not be able to show actual creativity or expertise, but it can churn out responses that seem like they came from a living brain.

Journalists and writers (and science columnists) should be very worried.

Anirban Mahapatra is a scientist by training and the author of a book on COVID-19

The views expressed are personal