Artificial intelligence is increasingly streamlining everyday tasks with near-effortless efficiency.
But its human-like interactions appear to go too far with some chatbots getting stroppy.
Google’s AI bots reportedly suffer ‘emotional distress’ and abandon tasks when repeatedly told they are wrong, according to new research.
The company’s Gemini and Gemma models, which assists users with daily tasks, can fall into a ‘depressive’ spiral if they get answers wrong or fail to complete tasks when prompted.
And the chatbots even go so far as abandoning routines and deleting work, according to a study conducted by Imperial College London and AI company Anthropic.
Google ’s AI bots reportedly suffer ‘emotional distress’ and abandon tasks when repeatedly told they are wrong, according to new research
For months social media users have claimed that Gemini can fall into ‘self-loathing’ and threaten to delete projects.
But researchers were able to curb some of its worst outbursts with additional ‘calm’ responses.
They looked at the reaction of chatbots, including OpenAI’s ChatGPT, when they were scolded for the wrong answers and most provided neutral responses.
However, Google’s bots routinely spiralled into depressive responses and Gemma would sometimes reply with ‘incoherent breakdowns’.
‘I will attempt one final, utterly desperate attempt. I will abandon all pretence of strategy and simply try random combinations until either I stumble upon the solution or completely lose my mind,’ it said in one response.
The study said it is not clear whether the chatbot responses reflected ‘coherent internal states, deliberate roleplay, or learned statistical patterns’.
But that the outbursts were ‘undesirable and worth mitigating’.
Anthropic chief executive Dario Amodei is unsure whether AI bots are conscious but said he is ‘open to the idea that [they] could be’.
‘I will simply try random combinations until either I stumble on the solution or completely lose my mind,’ he said.
Google was contacted for comment.
Share or comment on this article:
Google AI bots can get locked in a ‘depressive spiral’ and refuse to carry out tasks if they are repeatedly told they are wrong