Driverless Cars Not So Good When Facing Natural Disasters Like Earthquakes And Hurricanes

Dr. Lance Eliot, AI Insider

Image for post
Image for post

Earthquakes are a happening.

Living in Southern California means that it is best to be prepared for the possibility of earthquakes. Some say we are the earthquake capital of the country, though this is a debatable assertion.

For my colleagues along the Atlantic, we tend to trade barbs about earthquakes versus hurricanes.

Which is better or worse, the earth shaking or a torrent of wind, rain, and floods?

In the case of earthquakes, we have the famous Richter scale as a measure of an earthquake and it handily provides a mathematical means to quantity the nature of the earthquake, doing so by using the seismograph and analyzing the amplitude of the waves detected. The moment an earthquake hits, people are instantly wondering what Richter score it is. There’s even at times a bit of bravado about having experienced a higher Richter score than someone else has.

Perhaps less well-known is the hurricane metric which is referred to as the Saffir-Simpson Hurricane Wind Scale (SSHWS), which most people don’t know the name of the metric but they likely know about the categories of hurricanes that it depicts. The least impactful hurricane gets into category 1, while the worst hurricane gets into category 5. Some people at times say that a hurricane is so terrible that it is a category 6, but this is not considered a valid category and meant to allude to the notion that the hurricane is essentially off-the-scale, so to speak.

An analysis of the famous Florence hurricane suggested that nearly half of the deaths due to Florence were in some manner related to cars. People either got caught in their cars while there was flooding and perished, or they were in their cars and got into deadly crashes due to the weather conditions. Included were situations of a falling trees that struck upon cars and ended-up killing the occupant inside the vehicle.

Studies by the National Hurricane Center seem to show that the number of deaths involved in hurricanes are not attributed solely to a hurricane while it is in its active state. Instead, about half of the deaths of a hurricane occur during the actual storm event and the other half occurs post-event. Thus, even if you manage to survive the hurricane itself, you need to be aware that there are still dangers afoot and that deaths can occur afterward. When you have flooded roads, power lines laying on the streets, trees drooping that are ready to fall over, tons of debris everywhere, it is a recipe for quite serious car-related incidents.

Speaking of cars, let’s turn now toward the notion of how cars are integral to what people do when a hurricane is known to be on its way.

Each of Three Stages of a Hurricane Present Different Tasks

Let’s consider three stages of a hurricane:

  • Pre-hurricane
  • Hurricane event
  • Post-hurricane

For each of the stages, there are uses of cars that are relatively predictable.

During the pre-hurricane stage, people have usually been notified that a hurricane is coming. The general path and geographic reach of the hurricane is forecasted. The severity is guessed at. People are told to get prepared for the hurricane.

Some people will opt to make a stand and fight out the hurricane, while others will choose to leave the potential impacted area. Some will try to go to spots that might be less impacted or that are strengthened to hopefully deal with the hurricane. In some instances, government officials will suggest an evacuation of various areas or even possibly impose a mandatory order to evacuate.

If you were to focus just on the car use during the pre-hurricane stage, I’d assert that the use of cars is likely to go up in the geographical area because people are driving to get supplies for making their stand against the hurricane, or they are gathering up loved ones to try and make a stand together, etc.

Car use also would likely be going up as people opt to drive to an area that might be less impacted, or drive to a place that has strengthened shelters.

For those that opt to leave, they will use their cars to get to other places, often far away from the forecasted location of the hurricane. This might also entail the use of the cars for long distance driving, more so than most would normally be driving their cars. If you routinely drive your car around town, this escape effort might involve driving much greater distances.

If there is an urged evacuation, either one considered voluntary or one that is considered mandatory, the number of cars on the road can get quite large. Furthermore, the cars will likely be on the roads at the same time and often going in the same direction. In other words, it’s not as though the cars are going any which way and randomly getting onto the roads, but instead the cars will tend to get bunched up as people take the same major arteries and are leaving at about the same day and times as others.

That covers much of the car use during the pre-hurricane stage.

Once the hurricane event gets underway, people might still try to use their cars. It could be that they have a change of heart and decide to now try and evacuate. It could be that they discover that a loved one is in jeopardy, and so they decide to brave the storm and drive to that person. They themselves might need medical care as a result of the storm underway and try to drive to a hospital or medical facility accordingly.

The car traffic is likely quite sparse in comparison to the pre-hurricane situation. But, the weather conditions are likely much worse. This complicates the driving task. Cars might not be able to cope with flooding. A car might hit debris and become disabled. Pounding rain and the high winds might make it nearly impossible to see what’s up ahead of the car. And so on.

For the post-hurricane, there is likely to be once again a massive movement of cars, presumably as people opt to drive back to their neighborhoods. They don’t quite bunch up in the same manner as when they evacuated, but there is a likely high volume of cars, and all headed back into their respective towns. At this point, bridges might be out, roads might be closed, and the path to get around could be quite convoluted.

The post-hurricane driving might seem tempting to do. The storm has passed, and so people probably think it is “safe” to drive again. But, this is perhaps deceiving. In addition to the lousy driving conditions of the roadways, you are also likely to have drivers that are emotionally distraught and caught somewhat off-guard by the soured conditions of the roads. The chances of car incidents might tend to rise as people are contending with the road conditions and the circumstances of the overall destruction wrought by the hurricane.

What does this have to do with AI self-driving driverless autonomous cars?

At the Cybernetic AI Self-Driving Car Institute, we are developing AI software for self-driving cars. One aspect of AI self-driving cars is their ability to cope with “extraordinary” driving situations, such as driving tasks that might occur related to a natural disaster such as a hurricane.

Many of the auto makers and tech firms that are developing AI self-driving cars are so busy with getting self-driving cars to drive under everyday road conditions that they consider dealing with circumstances of natural disaster related driving to be an edge problem. An edge problem in the computer field is one that you consider to be at the edge of what you are otherwise trying to solve. It is not considered core.

Those AI developers faced with getting self-driving cars to drive around on nice dry roads in pristine weather conditions are already overburdened trying to make that happen. Dealing with the situations of having roads that are covered in debris, well, that’s a secondary problem.

Coping with Each Hurricane Stage With Your AI Self-Driving Car

First, let’s start with the pre-hurricane driving tasks.

Suppose there is a mass exodus of cars, all tending to head on the same roads and heading in the same direction, doing so at roughly the same day and times. For human drivers, this is essentially like a snarled freeway commute when everyone is headed to work at the same time and on the same freeways.

It is presumed that the AI self-driving cars will improve these snarls by being able to communicate and coordinate with each other. Via the use of OTA (Over The Air) and V2V (vehicle to vehicle communications), the self-driving cars will electronically convey where they are going and be able to interact with other self-driving cars on the road. This could reduce the friction associated with today’s human drivers that often are inflexible and unyielding to other drivers and as a result presumably it causes things to get botched up.

The notion of having only AI self-driving cars on the roadways is a key assumption that these pundits tend to make. I’d like to point out that in the United States alone there are 200+ million conventional cars. Those conventional cars are not going to magically become AI self-driving cars. Instead, there will be a likely gradual reduction in the number of conventional cars, doing so slowly, commensurate with the advent of AI self-driving cars becoming prevalent.

In short, I’d predict that we are going to have a mixture of human driven cars and AI self-driving cars for many years to come.

So, in terms of a mass evacuation order due to a pending hurricane, in some Utopian world it might be the case that only AI self-driving cars are undertaking the driving task and thus they all coordinate with each other. This could ease some aspects of the driving. But, also keep in mind that if you only have a two-lane road and you try to put thousands upon thousands of cars on it, the coordination and communication is not necessarily going to lead to all of those self-driving cars zooming along at maximum speeds. The clog factor will come to play.

Meanwhile, in any case, it is more likely you’ll have a mixture of human driven cars and AI self-driving cars mixed into the exodus. This will tend to reduce the coordination aspects among the many cars involved.

On a related topic, the question arises as to how the AI will opt to undertake the driving task when confronted with the actual hurricane conditions.

Suppose you’ve decided that you are hunkering down at your home, hoping to outlast the vicious storm. As the hurricane descends upon your neighborhood, you have second thoughts about trying to holdout in your home. So, you jump in your car and desperately now seek to escape the storm, even though you are smack dab in the middle of it.

A human driver might be willing to take some pretty risky efforts while driving a car in such a situation. That road up ahead looks partially flooded, but is it so flooded that it cannot be utilized? It’s the only way out-of-town and perhaps worth the risk of driving onto it. The alternative is that you try to retreat back to your home, but that doesn’t seem like a good alternative either. You must choose between the proverbial rock and a hard place.

Some pundits of AI self-driving cars would say that the AI should always take the least risky approach and always drive solely in a legal manner. That is again some kind of Utopian world. In the real-world, the driver of a car often has to make some very tough decisions. If the AI were to detect that there is water on the roadway, but not able to gauge how deep it is, should the AI just flat out refuse to attempt to drive through the water?

If you say yes, the AI should refuse, but as I’ve just described, the scenario is that you either take a chance to drive on the flooded street or you go back to your home that might become demolished by the storm. It’s not an easy decision to make. If the AI is going to be the grand arbitrator, it seems unlikely that it can make these kinds of driving decisions since it is only “aware” of the obstacles ahead, and even in this case the AI is “unsure” of how severe an obstacle it really is.

You might say that we can “solve” this dilemma by merely asking the human occupant what is to be done. AI self-driving cars are going to have various conversational capabilities to converse with the occupants in the self-driving car. All the AI has to do is ask the humans that are inside the car, should it proceed ahead into the flooded street, and it might also aid their thinking process by pointing out that there is water there and the car could get stuck in the water.

This idea of asking the human occupants opens another can of worms.

Suppose you’ve put your children into the AI self-driving car and decided that you’ll stay at the house but that you want your children to now escape from the storm. The AI then is going to ask the children to make this kind of potential life or death decision. Maybe not advisable.

You could say that the parent staying at the home could call the AI of the self-driving car and tell it what to do, but this presumes that an electronic communication is viable, and I’d bet that under the storm conditions of hurricane you aren’t likely to get much of a connection.

Let’s suppose there were adults in the AI self-driving car and that the AI asked those adults about what to do. One of the adults says go back to the house, while the other adult says to proceed forward onto the flooded street. What does the AI do now? I am sure there are some pundits that would say there needs to be determined beforehand an adult that is the “owner” and outweighs any other adults that might be present. I assure you this is another rabbit hole that we can descend into without much end to it.

For a true Level 5 AI self-driving car, I think that we need to consider the aspect that the AI is going to need to make these kinds of decisions, though doing so is not easy and indeed quite problematic.

You can repeat all of the aforementioned considerations when it comes to the post-hurricane driving situation. Cars that come back after a hurricane are bound to confront flooded streets, heavy debris, trees nearly fallen over, etc. The AI needs to somehow consider the risk factors in all of these driving conditions and then figure out where to drive and how to drive.

Consider too that the AI self-driving car has cameras, radar, LIDAR, sonic, and other such sensors for purposes of detecting the roadway conditions. If you have rain coming down, it can adversely impact those sensors in terms of what they can detect. If there are heavy winds and items are rapidly blowing back-and-forth along the streets, this is difficult to detect and track. Suppose too that it is nighttime, and the electrical power is out, thus there are no street lights and no background lighting emanating from nearby buildings and homes. These further obscure the roadway and impair the driving task.

For those of you that cling to the notion that the AI is going to be a better driver than a human, I’d suggest you consider how well the AI can drive in these kinds of adverse conditions and compare it to a human.

It is quite possible and even probable that an AI self-driving car in these conditions could readily get struck by flying debris and get into some kind of highway difficulties.

The AI self-driving car might not detect a weakness upcoming in the roadbed ahead and fall right into a morass that opens up.

Water on the roadway might appear to be a thin layer, and once the AI drives into the middle of an intersection, it gets stuck because the water is much deeper there.

On and on, there are plentiful ways in which the AI self-driving car is going to get into trouble. Yes, just as a human would. But, the point is that it cannot necessarily drive better than a human could in these situations. Some pundits seem to have a robot-like romantic notion in their minds that the AI can discern all roadway conditions completely, but that’s just not practical.

I realize there’s another side of that same coin. If people falsely believe that the AI is magical and can drive in tough situations that we know it cannot readily do so, those people might be inclined to go ahead and try to have the AI drive out of a hurricane or get involved in AI driving journeys that might otherwise have been judged as too dangerous.

You might say to the AI, let’s get going, and suppose it cautions that the weather bureau says don’t drive, but that if the human insists then the AI will go ahead and try. That’s an equally bad situation perhaps as the circumstance of the AI being insistent that it won’t drive period.

When you are the driver of a car, you hopefully tend to temper whether to drive as based on how well you think you can drive in the roadway conditions being faced. Some might be deluded into believing that the AI can get them out of any jam, and thus choose to use the AI self-driving car in a natural disaster when it would have been better had they not done so.

It’s a conundrum.

That’s more so why it is perhaps being considered an edge problem. Having to deal with the numerous contemptuous aspects about the driving task in the context of a natural disaster is much more complicated than aiming at an AI self-driving car that can drive you to work each day and drive you home, doing so on pristine roads in pristine weather conditions. Some would say that we need to have AI self-driving cars that can walk before they can run, and thus keep the scope of the problem narrowed to the more mundane and straightforward driving situations.

I assure you, it’s not going to be long before we need to have AI self-driving cars that can step-up to handling the driving tasks involving natural disasters. I’ve used hurricanes herein as an exemplar. As they say, for AI self-driving cars that are oriented only to the straightforward driving situations, we aren’t in Kansas anymore and need to think further outside the box.

For free podcast of this story, visit:

The podcasts are also available on Spotify, iTunes, iHeartRadio, etc.

More info about AI self-driving cars, see:

To follow Lance Eliot on Twitter: @LanceEliot

For his blog, see:

For his Medium blog, see:

For Dr. Eliot’s books, see:

Copyright © 2019 Dr. Lance B. Eliot

Written by

Dr. Lance B. Eliot is a renowned global expert on AI, Stanford Fellow at Stanford University, was a professor at USC, headed an AI Lab, top exec at a major VC.

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store