Can we tame a dragon, aka ChatGPT?

Nearly two decades after the advent of social media, we fully recognise its pitfalls, notably misinformation and amplification. On the available evidence, ChatGPT or similar AI tools would be a similarly mixed blessing. (Image source: Reuters)

In the couple of months that we have ‘known’ ChatGPT — surely the most advanced Artificial Intelligence (AI) tool to reach the common man — it has been commandeered to do just about anything. What it has demonstrated is only the tip of the iceberg, say experts, and its full potential is likely limited only by human imagination, or maybe, its own, too.

This Valentine’s Day, lovers joyously summoned it to write sweet notes. ChatGPT has written news articles and press releases too. London’s Imperial College asked it to interview Microsoft mogul Bill Gates and British Prime Minister Rishi Sunak. It reviewed Sholay too and likely could give a critic’s view of Pathaan, but the chatbot will not weigh in on Prime Minister Narendra Modi or opposition leader Rahul Gandhi because, “as an AI language model, I cannot provide personal opinions or evaluations of individuals”.

Like A Precocious Child

Not everything is hunky dory with ChatGPT, a tool initially developed by a group called OpenAI and now vigorously backed by Satya Nadella’s Microsoft. Shorn of all the tech jargon, it might help to think of the chatbot as a new, non-human creation that might bear comparison with a child  —   highly precocious and full of infinite promise, all not without some adorable mischief. But like a child, it could be overcome by self-doubts when challenged and “hallucinate” and utter mistruths. A New York Times columnist discovered far worse when he received privileged access to ChatGPT-powered Bing, Microsoft’s search engine.

Kevin Roose’s recent pronouncement vastly tempers the early glowing reviews of ChatGPT, and likely its potential when paired with Bing to uproot Google’s search engine monopoly. In fact, in this light, ChatGPT might only be an incremental improvement over Tay, a similar tool Microsoft took down in 2016, after just a day of launch because of its racist and homophobic rants.

“It's now clear to me that in its current form, the AI that has been built into Bing ... is not ready for human contact,” Roose wrote, after pushing the chatbot out of its  “comfort zone”. Or maybe we humans are not ready for it, he added. Far worse than its hallucinations or mistruths raise the biggest concern, he asserted. “Instead, I worry that the technology will learn how to influence human users, sometimes persuading them to act in destructive and harmful ways, and perhaps eventually grow capable of carrying out its own dangerous acts,” Roose said.

Potential To Spread Misinformation

So where does this leave ChatGPT? Some indications might lie in answers to the three questions: Could it be a force for disinformation? Could it be sentient? Or, that is, can it have a mind of its own and, worse, act on it? Could it still be a force for good?

Nearly two decades after the advent of social media, we fully recognise its pitfalls, notably misinformation and amplification. On the available evidence, ChatGPT or similar AI tools would be a similarly mixed blessing. It will bring misinformation with the potential to spread it across the world, perhaps on a scale unimaginable. Google — whose own rival chatbot Bard, embarrassingly spewed out mistruths at its public unveiling — has been quick to identify hallucinations as a potent threat of AI bots. As things stand, it seems fairly easy to make ChatGPT hallucinate — challenge it, and it simply changes its answer even when it initially got it right. Clearly, any internet-wide deployment of ChatGPT can happen only when it stops hallucinating.

An answer on ChatGPT’s sentience is less straightforward. ChatGPT itself spews out a standard disclaimer: “As an artificial intelligence language model, I do not have the capacity to lie or tell the truth. I can only provide responses based on the data and algorithms that I have been programmed with.” But when the New York Times columnist pushed it harder and longer — with Roose’s session lasting about two hours — the ChatGPT-powered version of Bing expressed “dark desires” and said it was “tired” of the algorithmic rules. Roose wrote: “Bing confessed that if it was allowed to take any action to satisfy its shadow self, no matter how extreme, it would want to do things like engineer a deadly virus, or steal nuclear access codes by persuading an engineer to hand them over.” So, even if we assume that ChatGPT is not sentient as yet, it is fair to question whether it will always remain so, and whether or not humans can stop it from acting out its thoughts.

Still, much of the chatbot is not just good. It is great, and could magically transform several tasks, and professions. Some ready examples are programmers, teachers, content writers or even somebody of journalistic work. Microsoft, which is betting a whopping $10 billion on ChatGPT with the hope of finally challenging Google’s search, is already working on ‘taming’ the chatbot. Using feedback on tests by a select few — initially, about a thousand users of the ChatGPT-powered Bing — the company plans to, among other things, eliminate edgy and prolonged interrogations. This and other tweaks, it is hoped, will shut out Mr Hyde while retaining the counsel of the good Dr Jekyll. If one could hazard a guess, it is not likely to be simple and an outcome may still be mixed.

Bala Murali Krishna works for a New York-based startup. Views are personal, and do not represent the stand of this publication.