Inaccurate images generated by AI chatbot were ‘unacceptable’, says Google boss

28 February 2024, 12:44

A Google building
Google-Reddit-Partnership. Picture: PA

The tech giant said it had been working ‘around the clock’ to address the issue.

The historically inaccurate images generated by Google’s Gemini AI chatbot were “unacceptable”, chief executive Sundar Pichai has said in a memo to staff.

Last week, users of Gemini began flagging that the chatbot was generating images showing a range of ethnicities and genders, even when doing so was historically inaccurate – for example, prompts to generate images of certain historical figures, such as the US founding fathers, returned images depicting women and people of colour.

Some critics accused Google of anti-white bias, while others suggested the company appeared to have over-corrected over concerns about longstanding racial bias issues within AI technology which had previously seen facial recognition software struggling to recognise, or mislabelling, black faces, and voice recognition services failing to understand accented English.

Following the Gemini image generation incident, Google apologised, paused the image tool and said it was working to fix it.

But issues were then also flagged with some text responses, with an incident highlighted where Gemini said there was “no right or wrong answer” to a question equating Elon Musk’s influence on society with Adolf Hitler’s.

Now Mr Pichai has addressed the issue with staff for the first time and promised changes.

In his memo, Mr Pichai said the image and text responses were “problematic” and that Google had been working “around the clock” to address the issue.

“I know that some of its responses have offended our users and shown bias – to be clear, that’s completely unacceptable and we got it wrong,” he said.

“No Al is perfect, especially at this emerging stage of the industry’s development, but we know the bar is high for us and we will keep at it for however long it takes. And we’ll review what happened and make sure we fix it at scale.”

He said Google had “always sought to give users helpful, accurate and unbiased information” in its products and this was why “people trust them”.

“This has to be our approach for all our products, including our emerging AI products”, he added.

Going forward, Mr Pichai said “necessary changes” would be made inside the company to prevent similar issues occurring again.

“We’ll be driving a clear set of actions, including structural changes, updated product guidelines, improved launch processes, robust evals (sic) and red-teaming, and technical recommendations. We are looking across all of this and will make the necessary changes,” he said.

By Press Association

More Technology News

See more More Technology News

Google CEO visit to UK

Google boss ‘cautiously optimistic’ about protecting elections from deepfakes

Radio Frequency Directed Energy Weapon

New weapon being developed to blast drones out of sky with radio waves, says MoD

Google I/O developer conference

Google tackles phone theft with new safety features coming to Android

Apple iPhone 15s on display

Apple brings eye tracking to iPhone and iPad in accessibility update

Google I/O developer conference

What do Google’s AI updates mean for everyday users?

Person running

Technology cannot replace human coaches in obesity treatment – study

Google I/O developer conference

The key announcements from Google’s I/O conference

Google discussing new AI tools during its Google I/O developer conference

Google AI updates a ‘direct response’ to ChatGPT dominance, experts say

Google chief executive Sundar Pichai speaking on stage at the firm's Google I/O developer conference

Google rebuilds search engine around AI and unveils new image and video tools

A person using a smartphone

Children ‘blackmailed into stripping for strangers’ on daily basis, says MP

Intelligence community ethnic minorities positions

China poses ‘genuine and increasing’ cyber risk to UK, GCHQ chief warns

Someone at a laptop

Research team finds AI slashes carbon capture and storage design time

AI chatbot eye problem assessing

New free ChatGPT model can teach maths, sing and even flirt

Open AI ChatGPT Update

OpenAI unveils new ‘smarter, faster’ version of ChatGPT

Google Stock

Google I/O: Tech conference expected to be dominated by AI

Claude 2 AI and ChatGPT icons seen in an iPhone

OpenAI rival Anthropic launches AI assistant Claude in European Union