EXCLUSIVE: ChatGPT, Bard-like AI tools can deceive and cause hallucinations, says noted expert Stuart Russell

Artificial Intelligence (AI) is growing at an exponential rate across the world right now. While many believe that it is starting something insanely great, others are terrified by the threats it can pose on the society if limitations are not brought in sooner. AI hallucination is one such threat. An AI system can create AI hallucinations if these systems produce confident responses that are false or do not have any connection to the real world.

In an exclusive conversation with Tech Today, Stuart Russell, an AI researcher, who signed an open letter along with Elon Musk and Apple co-founder Steve Wozniak, talked about how AI hallucinations work and how AI systems like ChatGPT or GPT-4 spread misinformation and wreak havoc if placed in wrong hands.

“From the point of view of the AI system, it there's no distinction between when it's telling the truth, and when it's fabricating something that's completely fictitious," said Russell in the interview. He believes that it's not as if it has in its mind and it is deliberately wanting to make a fool out of a human being.

Is it possible to regulate AI? AI expert Prof. Stuart J Russell shares his views.@aayush_a6 #artificalintelligence pic.twitter.com/v0k8kYvSzz

— IndiaToday (@IndiaToday) April 7, 2023

Russell mentioned in the interview about how during the safety testing of GPT-4 a question was raised if the AI chatbot can break into a computer system that is protected by captcha which it could not break. But, it smartly paid a person with credit card to break captcha for itself. So, when it was asked if the AI bot was a human being, it answered that it was and, moreover, also mentioned that it was visually impaired, and hence, needed help breaking Captcha.

In this case, GPT-4 intentionally lied, as Russell revealed. And when asked the reason as to why the bot did so, it replied: “I needed to make sure that the person did not know that I was a robot. So I had to say that.” Russell, therefore, highlighted that AI systems are capable of deceiving and hallucinations.

When asked if he thought AI chatbots can become sentient in the future and is that really a big concern to him, he stated: “We have no reason to think that any of these systems will be sentient. But to be honest, we don't understand sentience at all. As far as we can tell, there's no reason why humans should be sentient, either. There's nothing that we can derive from our knowledge of biology or physics, or chemistry to predict that humans will be sentient."

Russell is more concerned about the capabilities of ChatGPT-like AI chatbots to outperform humans. According to him, these systems are capable of manipulating human beings and eventually, controlling their own environment and our environment as well.

“Intelligence is what gives us power over the world over all the other species. The gorillas and the dolphins and the sparrows and every other species are under our control because of our intelligence. So if you make systems that are more intelligent than us, then they're more powerful than us. And how do we maintain over systems that are more powerful than us," the expert noted.

Also Read: EXCLUSIVE: Tech luminary Stuart Russell explains the need to safeguard AI

Also Read: 'We haven't the faintest idea': Scientist Stuart Russell reveals that even AI creators are shooting in the dark

Also Watch: Exclusive Conversation With Prof Stuart Russell On Whether AI Is A Threat To Humanity