- OpenAI has unveiled a new five-level scale for measuring progress towards artificial general intelligence.
- The company told employees that it is getting closer to creating an AI system capable of reasoning, Bloomberg reported.
- Experts say having a system that can make inference would be a “huge” leap.
OpenAI has introduced a new measure of progress towards artificial general intelligence (AGI).
The company behind ChatGPT shared the new five-tier classification system with employees at a company-wide meeting on Tuesday, Bloomberg reported.
The scale ranked AI systems by level of intelligence, ranging from Level 1 chatbots to Level 5 AI systems that can handle tasks for an entire organization.
Executives reportedly told staff that OpenAI is at Level 1, defined as an AI with conversational language capabilities, but is approaching Level 2, identified as a “reasoner” with human-level problem-solving abilities.
The next level of progress is a sign that OpenAI Chief Executive Sam Altman is inching closer to his professed ambition of creating AGI, or AI systems, that can match or surpass human capabilities at a wide range of cognitive tasks.
Since ChatGPT's launch, this mission has morphed into a fierce competition with billions of dollars of investment pouring into companies vying to be the first to reach the same goal.
Altman said he expects significant progress towards AGI will be achieved by the end of the decade.
Big deal
John Baden, a research fellow at the Leverholm Centre for Future Intelligence at the University of Cambridge, told Business Insider that the move from existing systems to ones that can make inference is “hugely important.”
“I can't overstate how big a deal it would be if we had AI systems that could inferentially anytime soon. We're talking about systems that can draw conclusions that we don't like,” he said.
Baden added that developing AI systems to this level risks machines “reasoning beyond humans”, which could have implications for the workforce.
“If these systems could reason like humans do, they would be much cheaper to keep employed than humans,” he said.
An OpenAI representative told Bloomberg that the criteria also includes “agent” and “innovator” levels, classifying AI systems by their ability to take action and help invent.
However, the validity of the scale itself is also open to debate.
Just an illusion?
Burden said the tech industry still appears to be stuck at Level 1, which covers chatbots currently available, adding that the jump from Level 2 to Level 3 to Level 5 is “essentially trivial.”
“No matter what Sam Altman says to hype us up, we're still only at Level 1,” he said. “We have AI systems that appear to be doing a little bit of reasoning, but it's not clear that that's just an illusion.”
It is also unclear whether the upper end of the scale is feasible.
“The highest level of AI that can do organizational work requires not only reasoning but many other human skills,” Hannah Kirk, an AI researcher at the University of Oxford, told BI.
“To get to these levels, the ability to reason but also to collaborate is really important,” she said. “To get to these levels, you need not only cognitive intelligence but also a lot more of the collaborative element and the social intelligence aspect.”
A representative for OpenAI did not immediately respond to Business Insider's request for comment, made outside of regular working hours.