Busting common AI myths

“By 2030, AI data centres could consume electricity on the scale of what's used by Japan today.”

This was the revelation with which Dipak Varsani, Public Sector Commercial lead for UKI & European Agencies at Wrangu, opened the company’s presentation at Government Transformation Summit, entitled “Challenging the Myths We Already Have Around AI”. The presentation was jointly delivered by Sharek Ullah, Wrangu’s founder and director.

As AI rapidly gains importance it is crucial to develop familiarity with the technology, the pair stated. It was for this reason that the session was dedicated to busting the widespread misconceptions which surround AI.

The first of these is the myth that AI is ‘intelligent’ in the way that a human being is.

While we can have conversations with AI, it doesn’t think like we do, and in fact its answers are based on "mathematics applied to a large data set", as Ullah phrased it. This means that when it replies to prompts, it isn’t thinking about its answer. Instead, it’s stringing together words based on probability.

It also doesn’t learn in real time. When we have conversations with AI, it doesn’t update it’s knowledge based on what we’ve inputted. Rather, it AI is like a whiteboard that gets wiped after every chat. Training happens at data centres and is a very complex and expensive operation, after which models are released for public use and remain frozen in their knowledge until the next update.

However, it’s still important to carefully consider what we put into AI platforms like ChatGPT and Copilot. While these platforms don’t learn from what we type, their host sites may retain the information and use it to train future models at a later stage.

It’s also important not to assume that AI will always be correct. As AI cannot draw from true knowledge, but instead relies on probability, it can hallucinate. And if the sources on which it has been trained on sources are incomplete, biased or false, that will be reflected in the answers it gives. It therefore requires human oversight to be used safely, especially in a context where its outcome could be used to inform policy or public messaging.

And finally: AI isn’t going to take over, like the Terminator. It can’t think, therefore it can’t self-improve, and agentic AI still has to work within the boundaries we set and with the tools it has been given by humans, towards a defined goal.

Organisations can limit the risks of AI usage by establishing clear guidelines and boundaries, as well as keeping humans in the loop to check output and making sure that data sources are reliable in the first place.

To end the session, the pair emphasised the transformative potential of well-regulated AI.

“The key takeaway here is that we are at a point where AI can truly reshape and transform the way we work and what we can achieve”, Sharek Ullah said.

He added: “It's not something to fear or trust blindly. It's something that should be managed, understood, designed, and guided carefully.

“At the end of the day, the difference between success and failure really comes down to how responsibly we use AI.”