A GOOGLE REPRESENTATIVE in Ireland has said that bizarre and inaccurate results produced through AI are “not typical” of the experience of most users on the search engine.

Google recently launched a feature in the US called AI Overview that involves AI generating a summary of results for the search term that a user enters.

However, users have reported cases of receiving inaccurate or nonsensical ‘summaries’.

TDs and senators on the Oireachtas Committee on Enterprise met this morning to discuss the impact of artificial intelligence on businesses with representatives from Google, Amazon, and Microsoft.

Sinn Féin TD Louise O’Reilly raised the problems with the AI Overview feature with Google’s Public Policy Manager in Ireland, Ryan Meade, who was in attendance.

“What I have read shows that it gives results that can be seen as misleading, like adding glue to pizza to keep the cheese on it, or more seriously, returning results that would say that serial killers or nuclear war have benefits,” O’Reilly said.

She said content like that is “misleading and deeply unhelpful” and asked if Google intends to launch the feature in Ireland and if it has any concerns in relation to that misinformation.

Meade outlined that AI Overview was recently rolled out in the US but has not been launched in Europe.

“It provides an AI-generated overview of the search results that you get for a particular query. It’s been in testing for a while and one of the reasons why it’s being rolled out is that it’s very helpful to users and also they are more likely to click on the links for further context that are provided within the AI overview,” he said.

“Since it’s been rolled out, there have been a few stories about specific queries that have been producing misleading results. Just to describe the way the system works – it is a large language model producing a piece of text that is based on the results that are returned for a particular query.

“These cases are not typical of most users’ experience with the product in that they’re in most cases searches for uncommon pieces of information with a limited number of search results.

“But I think the important thing is that we are continually refining these models and continually testing. That product is part of our lab experiments, in which each of these instances is fed back into the systems that review and refine and make the system safer.”

He said he did not have any information about whether Google intends to launch the feature here but that if it did, it would be “in conjunction with discussions with regulators”.

In response, O’Reilly said that reports about certain results “wouldn’t massively fill me with confidence”.

“I appreciate your point in relation to it being a discrete and a small number, but for some of those results, which I’m not going to ventilate here because they don’t deserve an airing, it is incredibly disruptive,” she said.

Meade said that “mixed in to some of those genuine cases there have been some mock-ups of results that weren’t actually returned”.

All three tech companies present made cases to the politicians about why they believe AI will be an important part of work in the years to come.

Sasha Rubel, Head of AI/Gen AI Policy for the EMEA region at Amazon Web Services, said that regulators should support innovation and experimentation in AI.

“We need innovation-friendly regulation; effective-risk based legislation for AI  that protects citizens and their rights and encourages trust while also allowing for experimentation  and practical application of AI,” Rubel said.

Similarly, Kieran McCorry, National Technology Officer at Microsoft Ireland, said Microsoft believes there is merit in regulators setting up a committee “focused on the use and deployment of AI, open to all voices and all communities”.