By Dr. Lance B. Eliot, the AI Insider for AI Trends and a regular contributor
I’ve been speaking at several self-driving car conferences and noticed that the in-vogue thing to do is to show a video clip of a self-driving car in action. Various developers of self-driving cars are eager to showcase their self-driving car videos. Similar to something you might see at any car show, these videos depict a smooth ride and you watch in amazement that there isn’t a human driver in the car. The video is usually shot from the backseat and so you have the perspective of looking out the windshield, and can see that there isn’t a person seated behind the wheel. The steering wheel turns back and forth on its own, as though a ghost is sitting there. The movement and grace of the car appears to be akin to having famous race driver Mario Andretti behind the wheel, driving the car better than any average car jockey could drive it.
These are essentially sizzle reels. If you aren’t familiar with the phrase “sizzle reel” it refers to a relatively short video, typically 3 to 5 minutes in length, and commonly is used for marketing purposes. Also known as a promotional video, pitch reel, montage video, or demo reel, these are usually produced with high-production values. In other words, the cost to make these short videos can be quite high, because the company using the reels considers it an essential form of marketing. If the short video doesn’t get the viewer excited about the product or service, and if it doesn’t look professionally made, it can lead to lost sales and buyers that won’t buy.
Years ago, when I was an academic researcher in Autonomous Vehicles (AV), we’d make quite simple videos and show them among other fellow researchers. These had almost no production value and the camera work was shaky and often out-of-focus. Fellow researchers didn’t care whether the video looked slick or not. Instead, we all were simply eager to see what the AV technology looked like and how it worked. Usually, the video would show not only what was working, but also show stuff that was being prototyped and so it might not be working or working incorrectly. We all knew that’s the way that these kinds of advances occur. You try something, it flops, you fix it, you continue on.
Not anymore. Now, with everyone wanting to attract Venture Capital (VC) money to their self-driving car startup, the “research” video has morphed into a modern day ultra-slick sizzle reel. It pretty much makes sense because if you were to show a video to a VC and it showed your self-driving car driving down the road and, if all of a sudden it veers into a ditch, they’d be walking out of the room thinking you must have been crazy to waste their time on coming to see you.
I want to warn you about falling victim to these sizzle reels. It is easy to get sizzled, and believe that the video is showing you the real truth of what the self-driving car can do. I provide next some handy tips of how to be watchful of the trickery involved.
Misleading Editing. Most of these self-driving car sizzle reels do not show you the self-driving car over any lengthy period of time. This can be excused by the sizzle maker by pointing out that the video is only 3 to 5 minutes in length. This is true, but what it does is shows you only those clips that had the self-driving car working perfectly. If you saw a longer video, maybe one of a self-driving car on a several hours drive, I assure you that’d see imperfections as it violated some rule-of-the-road.
No Human Appears in the Car. These sizzle reels will usually omit things like the human test driver that is supposed to be in the car in case the AI fails in some fashion or cannot figure out what to do. For nearly all states in the USA, any self-driving car maker must get a permit to have the car go on the public roadways, and must agree to have a human driver in the car that can take over control of the car at a moment’s notice. The human driver sometimes sits in the passenger seat next to the driver, but if you want to make a nifty video you instead have the human driver sit in the backseat, which then is not shown on camera because the camera is pointing forward from the backseat.
Crucial Omissions. The self-driving car will appear flawless in these videos. This is due to cutting out the video of the self-driving car when it inexplicably came to a sudden halt even though there wasn’t any reason to do so. Or maybe the self-driving car ran a red light, but that was left on the cutting room floor. All of these imperfections are typically omitted.
Pristine Weather Conditions. Usually, the self-driving car will be driving during ideal weather conditions. It is a sunny day. There isn’t any snow on the ground. You might not think about the weather conditions when you see the sizzle reel, and so you might mentally generalize that the self-driving car can work in any kind of weather condition. Don’t bet on it. The worse the weather, the harder it is for a self-driving car. Most self-driving cars right now cannot handle adverse weather and expect the human driver to take over control.
Open Roadways Only. The self-driving car videos normally show the self-driving car as it is zooming along the highway. We are used to car commercials showing us the joys of escaping to the open roads. This same kind of imagery is handy for the self-driving car sizzle reels. What it is doing though is misleading you into thinking that the self-driving car can go anywhere. In fact, the toughest kind of driving is not the open roadway, which is exceedingly simple, but instead the inner city driving is the toughest. I think we all know instinctively that driving in an inner city with cars within inches of each other, pedestrians coming at you from all directions, bicycle riders, potholes, cars illegally parked, and so on has got to be the toughest driving situations. Sizzle reels usually don’t show this.
No Nighttime Driving. Here’s another harder kind of driving, namely driving at night. Think about how hard it can be to see the roadway ahead. Is that a pedestrian walking across the street or not? Where does the road curve? Most of the sizzle reels will only show the self-driving car at daylight. Don’t be fooled into assuming it can equally drive at nighttime.
Purposeful Good-Driver Examples. This is the sneakiest of the sizzle reel elements. The video will show you a circumstance of a pedestrian that leaps from the sidewalk and the self-driving car magically comes to a halt. Proof that the self-driving car is not only good, but maybe even better than if a human driver was driving the car. This is usually sneaky because the self-driving car happened to encounter a circumstance that it was programmed to deal with. I am not suggesting that the makers hired a person to jump onto the roadway (which I suppose some might do), but I am saying that the self-driving car “knew” of that particular situation and so it dealt with it well. I would want to see other circumstances of when something happened that the AI did not anticipate, and see how it handled those situations.
Perfect Suburban Driving. Besides the open roadway, another handy aspect of these videos is they will show you a self-driving car driving in a suburban neighborhood. At first glance, it makes you believe that the self-driving car can handle inner city driving, since the suburb is seemingly similar. If you watch closely, you’ll usually see very little traffic on the roadway and it is as though they picked a suburb from the old “Leave It To Beaver” television show. Try driving that self-driving car in a busy suburb first thing in the morning when every driver is striving to drop their kids at school and are driving in a panic to then get to work.
I want to emphasize that I appreciate seeing the sizzle reels and welcome that the self-driving car makers are trying to strut their stuff. A presentation that is only a PowerPoint with bullets, or just still images of a self-driving car, all of that is certainly much less interesting and informative in many ways in comparison to showing a video of the self-driving car in action.
What I am trying to point out is that we cannot allow ourselves to be misled by these sizzle reels and falsely believe that a self-driving car can do more than it realistically can do. Propaganda videos are good for getting people excited about self-driving cars, and can attract more talent and money to invest in self-driving cars. On the other hand, hyping what self-driving cars can do is going to have a boomerang impact. Ultimately, investors will get wise to these shenanigans. I am not advocating that we return to the old days of grainy videos that have cheap production values. Let’s just try to restrain somewhat on the gee-golly wizardry of some of these sizzle reels. Keep the sizzle but add some reality.
This content is original to AI Trends.