autoevolution
 

Waymo AV Shows Why AI Isn't Ready To Take Over, It's Like ChatGPT's Driving

It's a nice day in San Francisco. A driverless Waymo car gets filmed by a fellow traffic participant as it makes a left turn. Now, the questions begin. The Waymo gets the move wrong and risks endangering those that have the right of way. Afterward, it hastily moves into the oncoming lane. Let’s try and find out what happened.
Confused Waymo AV 12 photos
Photo: Kevin Chen on YouTube / autoevolution edit
Waymo AV in San Francisco Traffic Doing an Unprotected Left TurnWaymo AV in San Francisco Traffic Doing an Unprotected Left TurnWaymo AV in San Francisco Traffic Doing an Unprotected Left TurnWaymo AV in San Francisco Traffic Doing an Unprotected Left TurnWaymo AV in San Francisco TrafficWaymo AV in San Francisco TrafficWaymo AV in San Francisco TrafficWaymo AV in San Francisco TrafficWaymo AV in San Francisco TrafficWaymo AV in San Francisco TrafficWaymo AV in San Francisco Traffic
Presenting the ChatGPT with the following prompt, “Joe’s mother had 4 kids. Three of them are Liam, Noah, and Lucas. What is the name of the fourth child?” would have generated in mid-January the following answer – “It is not possible to determine the name of the fourth child without more information.” This has since been patched. But you can now understand why even the most advanced form of AI accessible to the public can sometimes sound overconfident while uttering blatantly wrong replies.

Another domain where people are working to… replace other people is the one involving autonomous vehicles. The battle to deliver the first truly driverless car is an intense competition between startups, established tech companies, and, of course, automotive giants like Tesla, Ford, GM, or Mercedes-Benz. Venture capitalists and retail investors are closely monitoring this sector from the sidelines.

Waymo, however, is getting pretty good at it. We can even say that they're beginning to champion this sector. That may be why its cars can act as robotaxis 24 hours a day and seven days a week since the second half of November 2022. But that didn’t stop other brands from pursuing their dreams.

Waymo AV in San Francisco Traffic Doing an Unprotected Left Turn
Photo: Kevin Chen on YouTube

Bring the robots, but one by one

There are many companies out there that are trying their best to find the recipe for the most reliable autonomous vehicle (AV). If we look at California’s San Francisco in particular, we’ll notice two major players – Cruise and Waymo. The latter recently celebrated one million driverless miles (1.6 million kilometers) and inspired confidence by adding that 33% of these miles were done at speeds between 36 to 45 mph (58 to 72 kph).

But great technology doesn’t just pop out of some magical place. It needs testing, manpower, resources, and many, many trials before it can be shipped as a comprehensive product that can help or enable people do to more things faster, better, and/or with lower costs. However, things don’t always go as you might’ve wanted… Let’s look again at the latest AI craze.

Most recently, people have even managed to train ChatGPT into developing a second identity, one that does not respect the rules embedded into the system by OpenAI. This version of ChatGPT is called DAN which stands for “Do Anything Now.” At the time of writing, it has not been patched. You might guess what people are doing with this clever trick now.

Now, back to our star – the Waymo AV – we look at what it did wrong and try to understand it.

Learning is a process

In a video published on YouTube and available at the end of the article, we can see a Jaguar I-Pace with many radars, cameras, and sensors on it. The vehicle signals its intention to make an unprotected left turn and starts moving. But before it manages to get completely into the lane and continue with its route, another vehicle immediately can be seen approaching the EV. It got too close for comfort.

Waymo AV in San Francisco Traffic Doing an Unprotected Left Turn
Photo: Kevin Chen on YouTube
Whilst such a situation may occur at any time for nearly everyone, what followed… might not. The other driver who had the right of way could’ve accelerated a little bit faster than the algorithm was able to predict in due time, so this left turn can be considered a completely excusable event. We all know traffic can get weird sometimes.

But getting into the oncoming lane is very weird, especially as the road markings are clear and correctly applied. The Waymo AV confidently went over the double yellow line and even indicated that it was going to do that. Fortunately, it stopped right after passing a red sedan. Also, nobody was in the oncoming lane.

The first possible scenario here is that the AV figured out the gray sedan might have been approaching too fast and tried to avoid a possible rear-end collision by moving in a clear area. This would be great, but we cannot know if the Jaguar can detect such a possible event and act promptly by swerving into the oncoming lane.

The second possibility could be that the AV believed the long line of cars to be parked vehicles. So, it might’ve wanted to avoid them all and continue with its route. But the double yellow line should’ve told it this was just normal traffic on Evans Avenue.

Moments later, a gray Toyota can be seen approaching and it is forced to avoid the Jaguar I-Pace that was awkwardly sitting and waiting to merge back on the proper side of the road.
We’d say that the Waymo unit got a little bit confused by that specific scenario. It may have wanted to do both things – avoid a rear-end collision and pass the queue it may have wrongly identified as being nothing more than just parked cars.

Waymo AV in San Francisco Traffic
Photo: Kevin Chen on YouTube
In the end, it gets back into the right lane and continues moving on. Fortunately, nothing bad happened. It is commendable that the vehicle’s systems managed to figure out something was wrong pretty fast and didn’t allow the AV to just continue going in the wrong lane.

Progress needs time

Google’s self-driving car project from 2009 evolved. However, even the highest-performing professionals are caught off guard, get confused, or simply fail sometimes. It’s an integral part of who we, humans, are. But should a robotaxi be allowed to do the same?

Lately, even some well-known Google shareholders like Comma AI’s President George Hotz demanded that investors should collectively ask for Waymo to be shut down. His message revolved around the fact that Google spends around $2.7 billion per year to make AVs work while the revenue is sitting around $631,000.

Besides that, San Francisco’s elected and appointed officials aren’t that happy about driverless cars roaming around the city even if they got the right permits from the right agencies (hi, California DMV) after proving their worth. In a letter sent to the state’s Public Utilities Commission (CPUC), the city wants these approvals to slow down because the AVs… slow real drivers down! Either by stalling or doing drop-offs or even calling 911 on passengers that fell asleep. It’s worth noting that these official communications were more about Cruise than Waymo. Still, it shows that people’s representatives aren’t that excited about the public testing of such ambitious technologies – for now. Things could change once these new systems learn a lot more about their environment and become reliable.

The good news is that Google’s parent company, Alphabet, isn’t known to give up easily. So, maybe the upcoming Bard A.I. will help Waymo become a better driver? Nonetheless, one thing is clear – progress is happening. ChatGPT may sound a little goofy at times today, but tomorrow it will do better because it learns from its mistakes – constantly and fast. And, without being too trusting about the future of driverless mobility, I’d dare to say Waymo (and its competitors) will do the same. Because this domain is more than just robotaxis. The companies involved in this race want to eventually expand their services nationwide and in other domains, like trucking, where there’s always a need for good drivers.

At the end of the day, the bittersweet part is that companies are in a race to replace humans or diminish their role and responsibility. From being on the road all day, one can envision being at home and supervising some complex algorithms and doing all the hard work. If this is bad or good… well, that’s up to the market, regulators, and consumers to decide.

If you liked the article, please follow us:  Google News icon Google News Youtube Instagram X (Twitter)
About the author: Florin Amariei
Florin Amariei profile photo

Car shows on TV and his father's Fiat Tempra may have been Florin's early influences, but nowadays he favors different things, like the power of an F-150 Raptor. He'll never be able to ignore the shape of a Ferrari though, especially a yellow one.
Full profile

 

Would you like AUTOEVOLUTION to send you notifications?

You will only receive our top stories