A British tech start-up has unveiled an artificial intelligence system for automated driving that gives feedback to its riders.
The Lingo-1 model, created by Wayve, can answer questions that clarify why it made certain decisions or took particular courses of action.
Wayve's chief executive Alex Kendall told the CogX Festival that when the system was asked about the hazards driving down a street in north London, it fed back that a row of parked cars was a potential risk.
“Lingo commented that there was a row of parked cars,” Mr Kendall said.
“Why are you concerned about those parked cars? And Lingo said, 'Because there's a chance that a pedestrian might slip out from between them, might step out unexpectedly. I should be cautious about that.'”
“So it's really neat to see that Lingo was capable of reasoning about risks that maybe weren't immediately obvious to me.”
The conversation between man and machine may seem simple, but it actually cuts to the core of a major concern regarding AI – why certain systems do what they do and the transparency of their reasoning.
For many observers, it is this transparency that is fundamental to building the trust of the general public and allaying any fears they might have about AI technologies.
Mr Kendall told the CogX Festival that being able to interrogate Lingo-1 about why it made some decisions and not others is crucial to AI's development.
“One of the interesting things is that people in autonomous driving believe that perfect driving is all about seeing things,” he explained.
“What we've really learnt over the last couple of years is that decision-making problems are really complex.
“The reason why five-year-olds don't have a driving licence, even though they can see and recognise what a car or traffic light is perfectly, is that it's not until you're an adult that you develop the risk assessment, the judgment.
“One of the things we're learning is there's no better technology to develop robust and performance decision-making than AI.
“I think the inflection and rising capabilities we're seeing in AI technology this year, and just recently, is going to be what unlocks an application like autonomous driving at scale.”
When The Financial Times took Lingo-1 for a test drive recently, the system was able to give feedback on its reasoning for travelling at certain speeds. But when asked if it could park on the side of the road where there was a bus stop, Lingo-1 said it could, prompting Mr Kendall to comment that the model might need “a little more training”.
Nonetheless, Mr Kendall told the CogX Festival that he was convinced that eventually AI would render autonomous driving accident-free.
“1.3 million people lose their lives around the world each year due to road accidents,” he said.
“I think AI technology can drive that down to zero.
“Road accidents actually cost the UK 2 per cent of its GDP [gross domestic product], due to all the associated costs.
“So, with AI we can make a much more sustainable and efficient transportation system, with autonomous driving as a specific example.”