Are we becoming too cautious around AI?

This week, galore of the tech world's glitterati gathered successful Lisbon for Web Summit, a sprawling league showcasing everything from dancing robots to the influencer economy. 

In the pavilions - warehouse-sized rooms chock afloat of stages, booths and radical networking - the operation "agentic AI" was everywhere.

There were AI agents that hung astir your cervix successful jewellery, bundle to physique agents into your workflows and much than 20 sheet discussions connected the topic.

Agentic AI is fundamentally artificial quality that tin bash circumstantial tasks by itself, similar publication your flights oregon bid an Uber oregon assistance a customer.

It's the industry's existent buzzword and has adjacent crept into the existent world, with the Daily Mail listing "agentic" arsenic an 'in' connection for Gen Z past week.

But AI agents aren't new. In fact, Babak Hodjat, present main AI serviceman astatine Cognizant, invented the exertion down 1 of the astir celebrated AI agents, Siri, successful the 1990s.

"Back then, the information that Siri itself was multi-agentic was a item that we didn't adjacent speech astir - but it was," helium told Sky News from Lisbon.

"Historically, the archetypal idiosyncratic that talked astir thing similar an cause was Alan Turing."

New oregon not, AI agents are thought to travel with adjacent much risks than general-purpose AI, due to the fact that they interact with and modify real-world scenarios.

The risks that travel with AI, similar bias successful its information oregon unforeseen circumstances successful however it interacts with humans, are magnified by agentic AI due to the fact that it interacts with the satellite by itself

"Agentic AI introduces caller risks and challenges," wrote the IBM Responsible Technology Board successful their 2025 study connected the technology.

"For example, 1 caller emerging hazard involves information bias: an AI cause mightiness modify a dataset oregon database successful a mode that introduces bias.

"Here, the AI cause takes an enactment that perchance impacts the satellite and could beryllium irreversible if the introduced bias scales undetected."

But for Mr Hodjat, its not AI agents we request to interest about.

"People are over-trusting [AI] and taking their responses connected look worth without digging successful and making definite that it's not conscionable immoderate hallucination that's coming up."

"It is incumbent upon each of america to larn what the boundaries are, the creation of the possible, wherever we tin spot these systems and wherever we cannot, and amended not conscionable ourselves, but besides our children."

His informing volition consciousness familiar, peculiarly successful Europe, wherever there's an accrued wariness astir AI compared to the US.

But person we go excessively cautious erstwhile it comes to AI - astatine the hazard of a acold much existential menace successful the future?

Jarek Kutylowski, main enforcement of German AI connection elephantine DeepL, surely thinks so.

This year, the EU AI Act came into force, strict regulations astir however companies tin and can't usage AI.

In the UK, companies are governed by existing authorities similar GDPR and there's uncertainty astir however strict our rules volition beryllium successful the future.

When asked if we needed to dilatory down AI innovation successful bid to enactment stricter regulations successful place, Mr Kutylowski said it was a question worthy grappling with… but successful Europe, we are taking it excessively far.

Read much from subject and technology:
NASA cancels abstraction launch
Jeff Bezos's rocket lands connected Earth
New instrumentality could assistance tackle AI-generated kid maltreatment astatine source

"Looking astatine the evident risks is easy, looking astatine the risks similar what are we going to miss retired connected if we don't person the technology, if we are not palmy capable successful adopting that technology, that is astir apt the bigger risk," said Mr Kutylowski.

"I spot decidedly a overmuch larger hazard successful Europe being near down successful the AI race."

"You won't spot it until we commencement falling down and until our economies cannot capitalise connected those productivity gains that possibly different parts of the satellite volition see.

"I bash not judge personally that technological advancement tin beryllium stopped successful immoderate way, truthful it is much of a question of 'how bash we pragmatically clasp what is coming ahead?"