Tech Desk: Artificial Intelligence research farm OpenAI has suspended training its new AI models. It has done so after noting that its existing models behaved in a bizarre way and defied the commands given to them.
OpenAI revealed that it has experienced numerous incidents of disobedience and non-compliance where its existing models defied its instructions and refused its orders. Among them, the company said its agents searching government website in the US had acted in unexpected ways beyond what it was asked to do.
Responding to this, OpenAI has decided to halt the training of its new models. They said they would only resume when they are confident of having additional safeguards.
AI models behaving negatively in not new. There have been numerous incidents in the recent past. Not long ago, the Australian Prime Minister had raised concerns about OpenAI agents breaching its national healthcare system, even though it failed to mess with sensitive information.
Alternatively, there has been occasions when AI models have been unsuccessful in even dealing with general tasks. Not long ago it had struggled to solve a Sudoku puzzle and even filed to figure out how many R’s were there in the world strawberry.
Thus, even though we are witnessing AI’s growth in a meteoric speed, its reliability is still a question mark. Besides, this latest incident faced by OpenAI is immensely alarming and raises serious concerns AI’s safety and dependability. Urgent measures are necessary!
Source: The Guardian
Also Read: AI poses three big risks, warns Microsoft co-founder
Also Read: Trump believes minimum AI regulation is key to competing with China