- Autonomy founder Mike Lynch told the CogX conference that researchers still don't know how to build AI systems that can do more than one task well.
- AI has become a buzzword in recent years but the technology still has a long way to go.
- Lynch sold Autonomy to HP in 2011 for almost $11 billion but HP has since accused him of deliberately inflating the value of the company before he sold it.
British tech founder Mike Lynch, who sold a software company Autonomy to HP for almost $11 billion in 2011, believes there's a long way to go before machines start to show general intelligence.
Speaking on a livestreamed video call at the CogX tech conference, Lynch said: "We're nowhere near broad AI at the moment as far as we can tell."
"We're in a situation where what's interesting in AI are these difficult problems that can now be solved by systems of software," continued Lynch, whose appearance was somewhat surprising given he is being sued by HP for $5 billion in damages. HP has accused Lynch of deliberately inflating the value of Autonomy before he sold it. Lynch denies the allegations.
Over the last few years, some tech companies have talked up the abilities of their AI systems and speculated that machines could become generally intelligent in the coming decades.
Artificial general intelligence, or AGI, has been referred to as the holy grail in AI and is widely viewed as the point when machine intelligence passes human intelligence.
"As soon as you see that amazing leap, people then assume that intelligence is on some sort of continuum and because, for example, we've managed to solve something very difficult like image recognition or text generation type problems, we've got the ability to produce sort of super intelligences," said Lynch, who studied early forms of AI at the University of Cambridge.
Lynch, who was described by U.K. media as "Britain's Bill Gates" around the time of the HP deal, said there's a big difference between a narrow AI and broad AI. Narrow AI has got a very defined problem, while broad AI was described by Lynch as "something that would make its way in the real world in general, in the way that an animal does."
Lynch added that one of the major challenges with AI development is researchers can't simply "copy across" the human decision-making process to a machine, partly because scientists don't fully understand how humans make decisions.
Another big problem is bias. As an example, he pointed to an AI that is currently looking at literally trillions of words of text from humanity. "It is going to come out with some of the typical stereotypes… and some of the biases around gender and race, because it's reading everything that we write, and that is inherent," he said.
"That's actually a very difficult problem to solve because if you've got trillions of words, how do you go in and tell the AI, not to include those things?"
The U.S. Justice Department wants to extradite Lynch to the U.S. so that he can face criminal charges. In February, Lynch submitted himself for arrest in London and was released on bail.