By Lance Eliot, the AI Trends Insider
Julius Caesar is famously known for his radical act in 49 BC of defying authority by marching his army across the Rubicon river. Unless you happen to be a historian, you might not be aware that the Roman Senate had explicitly ordered Caesar to disband his army, return to Rome, and not to bring his troops across the Rubicon. His doing so was an outright act of defiance.
Not only was Caesar defiant, he was risking everything by taking such a bold and unimaginable act. The government of Rome and its laws were very clear cut that that any imperium (a person appointed with the right to command) that dared to cross the Rubicon would forfeit their imperium, meaning they would no longer hold the right to command troops. Furthermore, it was considered a capital offense that would cause the commander to become an outlaw. The commander would be condemned to death, and — just to give the commander some pause for thought, all of the troops that followed the commander across the Rubicon would also be condemned to death. Presumably, the troops would not be willing to risk their own lives, even if the commander was willing to risk his life.
As we now know, Caesar made the crossing. When he did so, he reportedly exclaimed “alea iacta est” which loosely translated means that the die has been cast. We use today the idiom “crossing the Rubicon” to suggest a circumstance where you’ve opted to go beyond a point of no return. There is no crossing back. You can’t undo what you’ve done. In the case of Caesar, his gamble ultimately kind of paid-off, as he was never punished per se for his act of rebellion, and he led the Roman Empire, doing so until his assassination in 44 BC.
I’m sure that most of us have had situations where we felt like we were crossing the Rubicon.
One time I was out in the wilderness as a scout master and decided to take the scouts over to a mountain area that was readily hiked over to. While doing the hike, I began to realize that we were going across a dry streambed. Sure enough, when we reached the base of the mountain, rain began to fall, and the streambed began to fill with water. Getting back across it would not have been easy. The more the rain fell, the faster the stream became. Eventually, the stream was so active that we were now stuck on the other side of it. We had crossed our own Rubicon.
At work, you’ve probably had projects that involved making some difficult go or no-go decisions. At one company, I had a team of developers and we were going to create a new system to keep track of VHS video tapes, but we also knew that DVD was emerging. Should we make the system for VHS or for DVD? We only had enough resources to do one. After considering the matter, we opted to hope that DVD was going to catch-on and so we proceeded to focus on DVD’s. We got lucky and it turned out to be one of the first such systems and even earned an award for its innovation. Crossed the Rubicon and luckily landed on the right side.
Of course, crossing the Rubicon can lead to bad results. Caesar was fortunate that he was not right away killed for his insubordination. Maybe his own troops might have even tried to kill him, since there were bound to be some that didn’t want to get caught up in the whole you-are-condemned to death thing. The recent news story about the teenage soccer team in Thailand that went into the caves and became lost, and then the rain closed off their exit, it’s something that they all easily could have died in those caves, were it not for the tremendous and lucky effort that ultimately saved them.
What does this all have to do with AI self-driving cars?
At the Cybernetic AI Self-Driving Car Institute, we are developing AI for self-driving cars. As we do so, there are often very serious and crucial “crossing the Rubicon” kinds of decisions to be made. These same decisions are being made right now by auto makers and tech firms also developing AI self-driving cars.
Let’s take a look at some of those kinds of difficult and nearly undoable decisions that need to be made.
LIDAR is a type of sensor that can be used for an AI self-driving car. It makes use of Light and Radar to help ascertain the world around the self-driving car. Beams of light are sent out from the sensor, the light bounces back like a radar wave, and the sensor is able to gauge the shapes of nearby objects by the length of time involved in the returns of the light waves. This can be a handy means to have the AI determine if there is a pedestrian that is standing ahead of the self-driving car and at a distance of say 15 feet. Or that there is a fire hydrant over to the right of the self-driving car at a distance of 20 feet. And so on.
For my assessment of LIDAR for AI self-driving cars, see my article: https://aitrends.com/selfdrivingcars/lidar-secret-sauce-self-driving-cars/
AI self-driving cars tend to use conventional radar to try and identify the surroundings, they use sonic sensors to do likewise, and they use cameras to capture visual images and try to analyze what’s around via vision related processing. They can also use LIDAR. There is no stated requirement that an AI self-driving car has to use any of those kinds of sensors. It is up to whatever the designers of the self-driving car decide to do.
That being said, it is hard to imagine that a self-driving car could properly operate in the real-world if you didn’t have cameras on it and weren’t doing vision processing of the images. You could maybe decide you’ll only use cameras, but that’s a potential drawback since there are going to be situations where vision alone won’t provide a sufficient ability to sense the real-world around the self-driving car. Thus, you’d likely want to add at least radar. Now, with the cameras and radar, you have a fighting chance of being able to have a self-driving car that can operate in the real-world. Adding sonar would help further.
What about LIDAR? Well, if you only had LIDAR, you’d probably not have much of an operational self-driving car, so you’d likely want to add cameras too. Now, with LIDAR and cameras, you have a fighting chance. If you also add radar, you’ve further increased the abilities. Add sonic sensors and you’ve got even more going for you.
Indeed, you might say to yourself, hey, I want my self-driving car to have as many kinds of sensors that will increase the capabilities of the self-driving car to the maximum possible. Therefore, if you already had cameras, radar, and sonar, you’d likely be inclined to add LIDAR. That being said, you also need to be aware that nothing in life is free. If you add LIDAR, you are adding the costs associated with the LIDAR sensor. You are also increasing the nature of the AI programming required to be able to collect the LIDAR data and analyze it.
There are these major stages of processing for self-driving cars:
o Sensor data collection and interpretation
o Sensor fusion
o Virtual model updating
o AI action plan updating
o Car controls commands issuance
See my framework about AI self-driving cars: https://aitrends.com/selfdrivingcars/framework-ai-self-driving-driverless-cars-big-picture/
If you add LIDAR to the set of sensors for your self-driving car, you also presumably need to add the software needed to do the sensor data collection and interpretation of the LIDAR. You also presumably need to boost the sensor fusion to be able to handle trying to figure out how to reconcile the LIDAR results, the radar results, the camera vision processing results, and the sonar results. Some would say that makes sense because it’s like reconciling your sense of smell, sense of sight, sense of touch, sense of hearing, and that if you lacked one of those senses you’d have a lesser ability to sense the world. You would likely argue that the overhead of doing the sensor fusion is worth what you’d gain.
Nearly all of the auto makers and tech firms would agree that LIDAR is essential to achieving a true AI self-driving car. A true AI self-driving car is considered by industry standards to be a self-driving car of a Level 5. There are levels less than 5 that are self-driving cars requiring a human driver. These involve co-sharing of the driving task with a human driver. For a Level 5 self-driving car, the idea is that the self-driving car is driven only by the AI, and there is no need for a human driver. The Level 5 self-driving car even is likely to omit entirely any driving controls for humans, and the Level 5 is expected to be able to drive the car as a human would (in terms of being able to handle any driving task to the same degree a human could do so).
For my article about the levels of AI self-driving cars, see: https://aitrends.com/selfdrivingcars/richter-scale-levels-self-driving-cars/
Tesla Foregoes LIDAR
It might then seem obvious that of course all self-driving cars would use LIDAR. Not so for Tesla. Tesla and Elon Musk have opted to go without LIDAR. One of Elon Musk’s most famous quotes for those in the self-driving car field is this one:
“In my view, it’s a crutch that will drive companies to a local maximum that they will find very hard to get out of. Perhaps I am wrong, and I will look like a fool. But I am quite certain that I am not.”
This is the crossing of the Rubicon for Tesla.
Right now and for the foreseeable future, they are not making use of LIDAR. It could be that they’ve made a good bet and everyone else will later on realize they’ve needlessly deployed LIDAR. Or, maybe there’s more than one way to skin a cat, and it will turn out that Tesla was right about being able to forego LIDAR, while the other firms were right to not forego it. Perhaps both such approaches will achieve the same ends of getting us to a Level 5 self-driving car.
For Tesla, if they are betting wrong, it would imply that they will be unable to achieve a Level 5 self-driving car. And if that’s the case, and the only way to get there is to add LIDAR, they would then need to add it to their self-driving cars. This would be a likely costly endeavor to retrofit and might or might not be viable. They might then opt to redesign future designs and write-off the prior models as unalterable, but at that point will be behind other auto makers, and will need to after-the-fact figure out how to integrate it into everything else. Either way, it’s going to be costly and could cause significant delays and a falling behind of the rest of the marketplace.
It would also cause Tesla to have to eat crow, as it were, since they’ve all along advertised that your Tesla has “Full Self-Driving Hardware on All Cars” – which might even get them caught in lawsuits by Tesla owners that argue they were ripped-off and did not actually get all the hardware truly needed for a self-driving car. This could lead to class action lawsuits. It could drain the company of money and focus. It would likely cause the stock to drop like a rock.
For my article about product liability for AI self-driving cars, see: https://aitrends.com/selfdrivingcars/product-liability-self-driving-cars-looming-cloud-ahead/
For my article about class action lawsuits against AI self-driving car makers see: https://aitrends.com/selfdrivingcars/first-salvo-class-action-lawsuits-defective-self-driving-cars/
This does not mean that Tesla couldn’t re-cross the Rubicon and opt to add LIDAR, but it just shows that when you’ve made the decision to cross the Rubicon, going back is often somewhat infeasible or going to be darned hard to do.
Perhaps Elon Musk had uttered “alea iacta est” when he made this rather monumental decision.
Another potential crossing of the Rubicon involves deciding whether to get to Level 5 by going straight to it, or instead to get there by progressing via Level 3 and Level 4 first.
Some believe that you need to crawl before you walk, and walk before your run, in order to progress in this world. For self-driving cars, this translates into achieving Level 3 self-driving cars first. Then, after maturing with Level 3, move into Level 4. After maturing with Level 4, move into Level 5. This is the proverbial “baby steps” at a time kind of approach.
Others assert that there’s no need to do this progressively. You can skip past the intermediary levels. Just aim directly to get to Level 5. Some would say it is a waste of time to do the intermediary levels. Others would claim you’ll not get to Level 5 if you don’t cut your teeth first on the lower levels. No one knows for sure.
Meanwhile, Waymo has pretty much made a bet that you can get straight to Level 5 and there’s no need to do the intermediaries. They rather blatantly eschew the intermediary steps approach. They have taken the bold route of get to the moon or bust. No need to land elsewhere beforehand. Will they be right? Suppose their approach falls flat and it turns out those that got to Level 4 are able to make the leap to Level 5, meanwhile maybe the efforts underway on Level 5 aren’t able to be finalized.
For more about the notion that Level 5 is like a moonshot, see my article: https://aitrends.com/selfdrivingcars/self-driving-car-mother-ai-projects-moonshot/
Does this mean that Waymo cannot re-cross the Rubicon and opt to first settle for a Level 4. As with all of these crossings, they could certainly back-down, though it would likely involve added effort, costs, and so on.
When developing the AI for self-driving cars, by-and-large it involves making use of various machine learning models. Tough choices are made about which kinds of neural networks to craft and what forms of learning algorithms to employ. Decisions, decisions, decisions.
Trying to later on change these decisions can be difficult and costly. It’s another crossing of the Rubicon.
For my article about machine learning and AI self-driving cars, see: https://aitrends.com/ai-insider/machine-learning-benchmarks-and-ai-self-driving-cars/
At the crux of most AI self-driving car systems there is a virtual world model. It is used to bring together all of the information and interpretations about the world surrounding the self-driving car. It embodies the latest status gleaned from the sensors and the sensor fusion. It is used for the creation of AI action plans. It is crucial for doing what-if scenarios in real-time for the AI to try and anticipate what might happen next.
In that sense, it’s like having to decide whether to use a Rubik’s cube or use a Rubik’s snake or a Rubik’s domino. Each has its own merits. Whichever one you pick, everything else gets shaped around it. Thus, if you put at the core a virtual world model structure that is of shape Q, you are going to base the rest of the AI on that structure. It’s no easy thing to then undo and suddenly shift to shape Z. It would be costly and involve gutting much of the AI system you’d already built.
It’s once again a crossing of the Rubicon.
Another tough choice in some cases is which brand/model of car to use as the core car underlying your AI self-driving car. For the auto makers, they are of course going to choose their own brand/model. For the tech firms that are trying to make the AI of the self-driving car, the question arises as to whom do you get into bed with. The AI you craft will be to a certain extent particular to that particular car.
I know that some of you will object and say that the AI, if properly written, should be readily ported over to some other self-driving car. This is much harder than it seems. I assure you it’s not just like re-compiling your code and voila it works on a different kind of car.
Furthermore, many of these tech firms are painting themselves into a corner. They are writing their AI code with magic numbers and other facets that will make porting the AI system nearly impossible. Without good commenting and thinking ahead about generalizing your system, it’s going to be stuck on whatever brand/model you started with. The rush right now to get the stuff to work is more important than making it portable. There are many that will be shocked down the road that they suddenly realize they cannot overnight shift onto some other model car.
See my article about kits for AI self-driving cars: https://aitrends.com/selfdrivingcars/kits-and-ai-self-driving-cars/
See my article about idealism and AI self-driving cars: https://aitrends.com/selfdrivingcars/idealism-and-ai-self-driving-cars/
This last example of crossing the Rubicon has to do with putting AI self-driving cars onto public roadways, perhaps doing so prematurely.
The auto makers and tech firms are eager to put their self-driving cars onto public roadways. It is a sign to the world that there is progress being made. It helps boost stock prices. It helps for the AI itself to gain “experience” from being driven miles upon miles. It helps the AI developers as they tune and fix the AI systems and do so based on real-world encounters by the self-driving car.
That’s all well and good, except for the fact that it is a grand experiment upon the public. If the self-driving cars have problems and get into accidents, it’s not going to be good times for self-driving cars. Indeed, it’s the bad apple in the barrel in that even if only one specific brand of self-driving car gets into trouble, the public will perceive this as the entire barrel is bad.
If the public becomes disenchanted with AI self-driving cars, you can bet that regulators will change their tune and no longer be so supportive of self-driving cars. A backlash will most certainly occur. This could slow down AI self-driving car progress. It could somewhat curtail it, but it seems unlikely to stop it entirely. Right now, we’re playing a game of dice and just hoping that few enough of the AI self-driving cars on the roadways have incidents that it won’t become a nightmare for the whole industry.
For more about this rolling of the dice, see my article about responsibility and AI self-driving cars: https://aitrends.com/ai-insider/responsibility-and-ai-self-driving-cars/
This then is another example of crossing the Rubicon.
Putting AI self-driving cars onto the roadways, which if it turns out premature, might make it difficult to continue forward with self-driving cars, at least not at the pace that it is today.
For the AI self-driving car field, there are a plethora of crossings of the Rubicon. Some decision makers are crossing the Rubicon and doing so like Caesar, fully aware of the chances they are taking, and betting that in the end they’ve made the right choice. There are some decision makers that are blissfully unaware that they have crossed the Rubicon, and only once something untoward happens will they realize that oops, they made decisions earlier that now haunt them. Each of these decisions are not necessarily immutable and undoable per se, it’s more like there is a cost and adverse impact if you’ve made the wrong choice and need to backtrack or redo what you’ve done.
I’d ask that all of you involved in AI self-driving cars make sure to be cognizant of the Rubicon’s you’ve already crossed, and which ones are still up ahead. I’m hoping that by my raising your awareness, in the end you’ll be able to recite the immortal words of Caesar: Veni, vidi, vici (which translates loosely into I came, and I saw, and I conquered).
Copyright 2018 Dr. Lance Eliot
This content is originally posted on AI Trends.