The potential perils of AI: should we hit pause on smart cars?
The BMW 740i, one of the smartest cars on the planet, arrived at our offices for testing in the same week that a group of prominent tech industry leaders warned that artificial intelligence (AI) was advancing too rapidly for comfort.
The group, which included Tesla’s Elon Musk and Apple co-founder Steve Wozniak, called for a six-month pause on systems like OpenAI’s GPT-4 and Google’s Bard, the smartest AI systems yet released. (The AI-generated picture used with this article was created using OpenAI’s Dall-E.)
AI is getting scarily clever, they argue, to the point where some of them are worried it might eventually wipe out humanity. A self-aware AI might not feel hate towards us, but might simply decide that humans are a threat to its existence and eliminate us.
‘WE ARE NOT READY’
Ethics philosopher and AI researcher Eliezer Yudkowsky says that AI might neither love nor hate humans, but rather see us as consisting of atoms that could be used for something else. It’s not as easy as programming AI to have ethics. Yudkowsky believes that humanity isn’t prepared for AI’s capabilities within any reasonable time window.
He is calling for an “indefinite and worldwide” pause on largescale AI training projects. In a Time column, Yudkowsky said: “Shut it all down. We are not ready.”
He’s watched too many sci-fi movies and is being alarmist, one might argue. Sure, dystopian sci-fi movies have primed us for scenarios where AIs bent on world domination unleash armies of robots against us, but we should be safe as long as we don’t build smart robots. The world’s most advanced robots, like Honda’s Asimo, can run and do other cute things, but their capabilities are nowhere near those of the gun-toting hellbots from the Terminator and Matrix movies.
But who needs cyborg soldiers if you already have millions of wheeled, programmable conveyances roaming the planet? Wheeled conveyances that are ever more connected and getting ever closer to driving themselves.
If you want to recruit a robot army, look no further than clever cars.
Hackers have already demonstrated they’re able to take remote control of vehicles, and crash them, using just a laptop and an internet connection. If the car were to be controlled by an unpredictable AI with unknown motives, it doesn’t take an alarmist Luddite to see the potential pitfalls.
Cars already use AI-based systems such as voice and speech recognition, gesture controls, eye tracking and virtual assistance to name a few.
Even as I marvelled at being able to use voice control to open the 740i’s electric doors, part of me wondered whether we shouldn’t be hitting the pause button in making cars too smart, especially in terms of self-driving ability.
Cars like the BMW 740i already have semi-autonomous driving ability by being able to keep a safe following distance and not veering from their lanes, but giving them the ability to go full autopilot in all driving situations has proven difficult because computers haven’t been able to adapt to complex, ever-changing road conditions like humans do. The emergence of smarter AI might be able to fix this problem and take cars over that final full-autopilot hurdle.
The original idea behind selfdriving cars, apart from convenience, was to prevent road deaths, driven by the notion that computer-operated cars would obey road rules and wouldn’t succumb to road rage, like humans do. With superintelligent AI, who knows? It may not necessarily have motives that align with human preservation, if you subscribe to Yudkowsky’s cautionary view.