ChatGPT is being used weekly by over 700 million people around the world, and it turns out that it has some users in India’s local assemblies as well. And they’re giving the company product feedback from the halls of Parliament.
Abhay Singh, the Member of Legislative Assembly from Gosaiganj in Uttar Pradesh, detailed his experiences with ChatGPT in an interesting speech in the UP Assembly. Singh said that he’d used ChatGPT for a variety of applications, including finding out who India will side with in World War 3, and how long he’d remain in power. Singh said that ChatGPT kept flipping on its answers, and said that too much reliance on the product could lead to trouble for everyone involved.
“I have been a Google user for 10-15 years,” Singh told Parliament. “Now we have this AI product called ChatGPT. It’s a sort of a scam or deception,” he said, as the rest of the members laughed. “It takes information from other parties and gives it to us; it takes information from us and then gives it to other parties,” he claimed.
“Speaker Sir, you’d just said that Mr. Mata Prasad Pandey had become an MLA in 1974. I put this into ChatGPT. But ChatGPT told me he’d become an MLA in 1980. Speaker Sir, I will trust your word,” he said, folding his hands, as the other members giggled. “I’m not going to trust ChatGPT,” he added.
“ChatGPT or other AI programs seem to be taking information from different sources and rotating it around,” he said. “I tried asking ChatGPT what the mood is in my constituency, which party the people are backing. Just the other day, I was bored at night, and I started chatting with ChatGPT. I asked ChatGPT, if there’s a third world war, which side will India fight on. ChatGPT told me India would side with the US. But I pushed back and asked it that given how the US was now supporting Pakistan, how could India and US end up on the same side. Then ChatGPT immediately changed its tune, and said that India would side with Russia. Is this AI?” he asked to titters from the other MLAs.
“Then I shared my kundli (astrological chart) on ChatGPT. I’m not kidding,” as other MLAs laughed even harder. “I asked ChatGPT that based on my chart, how long would I be in power. AI told me I’d be in power till this and this date,” he said. Singh even took out his phone and showed it to the speaker, saying that they could read his chat logs and verify anything he was saying. “Then I told ChatGPT that my own pandit (priest) had said I’d remain in power till a different date. ChatGPT immediately said it had been wrong, and the priest was right. ChatGPT said that it had overlooked a part of my horoscope to give me the wrong prediction,” he said.
“Speaker Sir, I submit that ChatGPT doesn’t actually tell you anything. It’s simply collating information from different sources, combining it, and then presenting it as its own insights,” Singh said.
“But you need to be mindful of the questions you can ask ChatGPT,” the speaker told him, hinting that ChatGPT wasn’t designed to give exact answers to the kinds of questions he was asking.
“If we base everything on AI,” Singh went on, “just as how Google Maps often sends people off bridges with wrong directions, AI could end up doing the same. It’s good to use AI to get an initial feel of an issue, but if we start relying on it, and start using it for our work, we’d be in soup,” he said.
“So my question to the Parliament is, how long will it take for AI to upgrade itself to give real answers to important questions. When can it actually help us in administration? When can I ask it where to build a road so that I get the most votes,” as members laughed and even the speaker couldn’t resist a smile.
While Abhay Singh’s speech was delivered in everyday language that was very distant from the tech specifics that follows the usual commentary on AI, he did hit upon a few important points. He highlighted ChatGPT’s well-known issues with sycophancy, in which it tends to agree with whatever the user asks — ChatGPT changed its answers on World War 3 and Singh’s horoscope when pushed back. Also, he highlighted that ChatGPT didn’t have any information of its own — it was merely compiling information received from other sources, and its answers would be only as good as the information it was trained on. These are both important issues for AI models to address, and now they’ve been raised in the state assembly of India’s most populous state which has a population nearly as much as the US. And with Parliamentarians across the world increasingly beginning to use AI — Sweden’s PM has said that he uses AI to get a broad first pass at issues — it becomes even more important for lawmakers to understand the potential, along with the many drawbacks, of this new technology.