The Promised Land of Safety
In 2019, Elon Musk said that he was very confident that there would be Tesla Robotaxis on the road with no human drivers next year. It’s now been more than four years, and there’s still no Tesla Robo taxi.
In 2023, Google-owned WH and GM-owned CRW beat Tesla to the punch, securing regulatory approval to launch commercial Robo taxi services 24/7 in San Francisco. Technologists were quick to proclaim this as a great victory and a big step towards the widespread rollout of self-driving cars.
Early Rollout, Disastrous Results
However, within just a few weeks of operation, the San Francisco autonomous driving experiment has already been a complete unmitigated disaster. Self-driving cars are causing crashes, killing dogs, creating traffic jams, and even disrupting emergency vehicles.
On August 18th, 2023, just a week after receiving approval for 24-hour operations, a Cruise autonomous vehicle crashed into a fire truck, causing at least one injury. The San Francisco Fire Chief says there have been numerous incidents of Robo taxis disrupting emergency vehicles, driving over fire hoses, and driving into active crime and fire scenes. It’s only a matter of time before something catastrophic happens.
Over the past decade, companies like Google, GM, and Tesla have poured billions of dollars into developing self-driving cars with the promise that they’ll be far safer than human drivers. In this video, we’ll look at the technological challenges of autonomous cars and why full self-driving is probably never going to happen.
In the US alone, around 40,000 people die every year due to traffic-related incidents. Countless more are left with life-changing injuries. Theoretically, a self-driving car could be safer than a human driver and cut down a lot of these deaths. For example, a robot car will never get drunk, never fall asleep, and never text. A computer can run calculations far faster than a human with far fewer mistakes.
When you drive a car, you take in visual and auditory information through your ears and eyes: where are the road lanes, where are the stop lights, where are the other cars, is a car near you honking? If you put a bunch of cameras and microphones on a car, a computer would theoretically have all the information a human driver would have. Thus, in theory, the computer should be able to do everything a human driver does, faster and with less chance of error.
In fact, if you put radar or LIDAR sensors on a car, a computer could have even more information than a human has. Fully autonomous cars are theoretically possible.
Questionable Data and Safety Claims
It makes sense that investors would be interested, as the commercial opportunity is enormous. Consumers would be willing to pay a lot of money for a self-driving car, or even better, you could create an Uber-like service where people hail autonomous cabs without the expense of a human driver. The profit margins would be extremely attractive.
For the past decade, tens of billions of dollars have been poured into the space. In 2011, Google acquired the self-driving startup Weo. In 2014, General Motors acquired Cruise and eventually secured additional funding from SoftBank.
In 2014, Tesla unveiled its autopilot feature, which was its first step towards full self-driving. Between 2017 and 2020, Ford and Volkswagen each invested more than $1 billion into a self-driving venture called Argo AI. These companies collectively spent tens of billions of dollars hiring tens of thousands of talented engineers and AI researchers from around the world.
These startups retrofitted regular cars with the cameras and radars necessary for self-driving. They tested them out on real streets, but with real humans always in the car, ready to take over if necessary. They used the data to train their AI models and refine their self-driving software. In terms of training data, Tesla seemed to have a clear advantage. Since 2015, all new Tesla cars sold have cameras and radars built in.
Driving data is sent back to Tesla HQ over Wi-Fi, giving Tesla billions of miles worth of real-world driving data. According to Elon Musk, this data allows their self-driving technology to increase exponentially, and it’s only a matter of time before they have full self-driving cars safer than human drivers.
By the late 2010s, it looked like things were progressing pretty well. In 2017, WHOS started fully autonomous taxi rides in Arizona without a human safety driver. According to WHOS’s own analysis, their autonomous cars were safer than human drivers. Tesla claims that their autopilot feature is far safer than unassisted driving.
In most quarters, there’s less than one crash per 4 million miles driven with Tesla autopilot, compared to the US average of about one crash every half million miles. The autopilot is only available on highways. Their so-called full self-driving beta is available on city streets as well. They claim that Teslas using full self-driving are 80% less likely to get in a crash than regular cars.
So, everything is great. Autonomous cars are already safer than human drivers, so we should all switch to using self-driving cars today and save tens of thousands of lives, right? Unfortunately, it’s not that simple. The problem with data and statistics is that they can easily be manipulated and taken out of context to give a misleading impression.
In the most recent quarter available, Tesla claims that vehicles with the autopilot function engaged record one crash every 4.85 million miles driven. This compares to one crash for every 1.4 million miles driven for Tesla without autopilot engaged and one crash per 652,000 miles driven for all cars in the US. This suggests that with autopilot, Teslas are more than seven times safer than regular cars, and even without autopilot engaged, they’re more than twice as safe.
There are a few major problems with this analysis. Firstly, autopilot is only engaged on highways. The vast majority of automobile accidents are fender benders, which happen on city streets. Thus, it’s an apples-to-oranges comparison, which is completely meaningless.
Secondly, Teslas are expensive, much more expensive than the average gasoline-powered car. People who buy them are wealthier and older. On average, wealthier and older people are less likely to get into crashes anyway. We know this from Tesla’s own data. Even without autopilot engaged, Teslas get into less than half as many crashes compared to the US average.
Tesla has access to a huge amount of driver data. If they wanted to, they could create an apples-to-apples comparison of vehicle safety with and without autopilot, but instead, they only give us useless apples-to-oranges comparisons. Why don’t they give us better data? Probably because they themselves don’t like the results.
In April of 2020, the company disclosed that Tesla drivers have driven an aggregate 3 billion miles since the product was launched in 2014. There’s a website called TeslaDeaths.com, which keeps track of all Tesla-related fatalities and whether or not autopilot was involved. According to their data, there have been 15 fatalities with suspected autopilot involvement during that period. That gives a rate of 0.5 deaths per 100 million miles driven.
So how does that compare to unassisted human driving? In 2019, there were 3.26 trillion total vehicle miles traveled in the US and 36,096 total traffic-related fatalities. That translates to 1.11 fatalities per 100 million miles driven. So, based on this data, Tesla autopilot is about 55% safer than the average car. But we have to make some adjustments. Even when a Tesla has autopilot engaged, the driver is still required to keep their hands on the wheel and be prepared to take control at any time.
As we explained before, Tesla drivers are safer than average due to their socioeconomic and age characteristics. In the first quarter of 2020, which is the last full quarter in the relevant period, Tesla drivers without autopilot engaged were 2.6 times less likely to crash per mile driven as compared to the US average.
Another factor is the weather. According to the US Department of Transportation, roughly 16% of all traffic-related fatalities happen during inclement weather conditions. Tesla autopilot refuses to operate during times of poor visibility, such as rain, fog, or snow. It also refuses to operate on narrow or winding roads or other unfavorable conditions.
After applying the driver quality adjustment and the inclement weather adjustment, the average US death rate goes down to 0.36 deaths per 100 million miles. This is significantly lower than the autopilot death rate of 0.5. Thus, autopilot is more dangerous, not less.
The Challenges of Real-World Driving
It makes sense that investors would be interested, as the commercial opportunity is enormous. Consumers would be willing to pay a lot of money for a self-driving car, or even better, you could create an Uber-like service where people hail autonomous cabs without the expense of a human driver. The profit margins would be extremely attractive.
For the past decade, tens of billions of dollars have been poured into the space. In 2011, Google acquired the self-driving startup Weo. In 2014, General Motors acquired Cruise and eventually secured additional funding from SoftBank. In 2014, Tesla unveiled its autopilot feature, which was its first step towards full self-driving.
Between 2017 and 2020, Ford and Volkswagen each invested more than $1 billion into a self-driving venture called Argo AI. These companies collectively spent tens of billions of dollars hiring tens of thousands of talented engineers and AI researchers from around the world.
These startups retrofitted regular cars with the cameras and radars necessary for self-driving. They tested them out on real streets, but with real humans always in the car, ready to take over if necessary. They used the data to train their AI models and refine their self-driving software. In terms of training data, Tesla seemed to have a clear advantage.
Since 2015, all new Tesla cars sold have cameras and radars built in. Driving data is sent back to Tesla HQ over Wi-Fi, giving Tesla billions of miles worth of real-world driving data. According to Elon Musk, this data allows their self-driving technology to increase exponentially, and it’s only a matter of time before they have full self-driving cars safer than human drivers.
By the late 2010s, it looked like things were progressing pretty well. In 2017, WHOS started fully autonomous taxi rides in Arizona without a human safety driver. According to WHOS’s own analysis, their autonomous cars were safer than human drivers. Tesla claims that their autopilot feature is far safer than unassisted driving.
In most quarters, there’s less than one crash per 4 million miles driven with Tesla autopilot, compared to the US average of about one crash every half million miles. The autopilot is only available on highways. Their so-called full self-driving beta is available on city streets as well. They claim that Teslas using full self-driving are 80% less likely to get in a crash than regular cars.
So, everything is great. Autonomous cars are already safer than human drivers, so we should all switch to using self-driving cars today and save tens of thousands of lives, right? Unfortunately, it’s not that simple. The problem with data and statistics is that they can easily be manipulated and taken out of context to give a misleading impression.
In the most recent quarter available, Tesla claims that vehicles with the autopilot function engaged record one crash every 4.85 million miles driven. This compares to one crash for every 1.4 million miles driven for Tesla without autopilot engaged and one crash per 652,000 miles driven for all cars in the US. This suggests that with autopilot, Teslas are more than seven times safer than regular cars, and even without autopilot engaged, they’re more than twice as safe.
There are a few major problems with this analysis. Firstly, autopilot is only engaged on highways. The vast majority of automobile accidents are fender benders, which happen on city streets. Thus, it’s an apples-to-oranges comparison, which is completely meaningless.
Secondly, Teslas are expensive, much more expensive than the average gasoline-powered car. People who buy them are wealthier and older. On average, wealthier and older people are less likely to get into crashes anyway. We know this from Tesla’s own data. Even without autopilot engaged, Teslas get into less than half as many crashes compared to the US average.
Tesla has access to a huge amount of driver data. If they wanted to, they could create an apples-to-apples comparison of vehicle safety with and without autopilot, but instead, they only give us useless apples-to-oranges comparisons. Why don’t they give us better data? Probably because they themselves don’t like the results.
In April of 2020, the company disclosed that Tesla drivers have driven an aggregate 3 billion miles since the product was launched in 2014. There’s a website called TeslaDeaths.com, which keeps track of all Tesla-related fatalities and whether or not autopilot was involved. According to their data, there have been 15 fatalities with suspected autopilot involvement during that period. That gives a rate of 0.5 deaths per 100 million miles driven.
So how does that compare to unassisted human driving? In 2019, there were 3.26 trillion total vehicle miles traveled in the US and 36,096 total traffic-related fatalities. That translates to 1.11 fatalities per 100 million miles driven. So, based on this data, Tesla autopilot is about 55% safer than the average car. But we have to make some adjustments. Even when a Tesla has autopilot engaged, the driver is still required to keep their hands on the wheel and be prepared to take control at any time.
As we explained before, Tesla drivers are safer than average due to their socioeconomic and age characteristics. In the first quarter of 2020, which is the last full quarter in the relevant period, Tesla drivers without autopilot engaged were 2.6 times less likely to crash per mile driven as compared to the US average.
Another factor is the weather. According to the US Department of Transportation, roughly 16% of all traffic-related fatalities happen during inclement weather conditions. Tesla autopilot refuses to operate during times of poor visibility, such as rain, fog, or snow. It also refuses to operate on narrow or winding roads or other unfavorable conditions.
After applying the driver quality adjustment and the inclement weather adjustment, the average US death rate goes down to 0.36 deaths per 100 million miles. This is significantly lower than the autopilot death rate of 0.5. Thus, autopilot is more dangerous, not less.
A Long Road Ahead for Self-Driving Cars
The technologists claim that they just need a little bit more time and data, and eventually, Robo taxis will be safer than humans. But they’ve already been working on this for over a decade, collected billions of miles of data, and spent tens of billions of dollars. How many more people need to die before we wake up to the reality? Self-driving cars are never going to happen. All right, guys, that wraps it up for this video. What do you think about autonomous cars? Let us know in the comments section below.
A seasoned software engineer with more than eleven years of experience who writes about news and international topics on the side. Afolabi, who holds a degree in Electrical/Electronics Engineering, combines technical know-how with a sharp awareness of global events to offer a distinctive analytical viewpoint to his work. Afolabi is the one to turn to for perceptive commentary on world affairs.