By Lance Eliot, the AI Trends Insider

For those of you that have lived through multiple generations of computers, you likely know that the amount of memory available on an off-the-shelf computer has steadily increased.  It is nearly humorous today to look back at the Radio Shack TRS-80 microcomputer of the mid-1980s and realize that it came with 24KB RAM, which at the time was thought to be a whopping amount of memory. You might recall that when 8- inch floppy drives first appeared in the mid-1970s that people were ecstatic that those gargantuan floppies could hold 800KB. Next along came 5.25 inch floppies that held 1.2MB, and then the 3.5 inch floppies that had 1.44MB. Of course, in modern times we look at a 1TB USB memory stick and seem to think it’s not much memory at all.

There is a famous or perhaps infamous quote attributed to Bill Gates in which he allegedly said that “no one will need more than 637K of memory for a personal computer, 640K ought to be enough for anybody.” I don’t want to be accused of spreading fake news, and so please be aware that Bill Gates has refuted that he said such a thing. There are a number of myth busters that have tried to show that he never said the 640K statement and that it was falsely attributed to him, while others now appear to believe there’s an elitist led cover-up underway and he really did say it – a cover-up of the same significance as to maybe we faked the moon landing and perhaps there was a grand conspiracy plot to assassinate JFK.

I’m not going to get mired into that whole debate, and instead my emphasis is that we’ve seen the amount of computer memory increase greatly over time. And, equally important to my discussion here, we’ve all seemed to use that memory up. In other words, each time we’ve gotten more computer memory available, we all seem to find ways to use it. You might at first assume that with the increase in available memory that a lot of it therefore just sits idle and unused. Certainly, there are many that do allow the available memory to be unused, but inch by inch we all seem to eventually use it up

Indeed, there is a popular credo known among most software developers that says this: “Any given program will expand to fill all the available memory.” Or, namely that we’ll fill computer memory with either programs or data, and every time we get more computer memory, we once again ultimately fill it up. It’s axiomatic.

Another way to think of this phenomena is to say that it is induced demand.

Induced demand is an economic theory that postulates there are situations that can possibly spark demand for a resource (it is “induced” to occur). It involves demand and supply. In particular, if the supply of something is increased, then in fact one belief is that more of that supply is consumed, or in other words demand will rise (versus assuming that the existing demand will remain the same, and thus there would then be an “excess” of the now added supply).

One example of the potential impact of induced demand relates to roadways.

Suppose that you have a local freeway that has been getting clogged with traffic. The assumption by most local politicians and even transportation planners is that if you were to expand the local freeway that it would then alleviate the freeway traffic congestion. This seems to make intuitive sense. If a pipe is not able to handle the flow of something (in this case, traffic flow), if you simply enlarge the pipe it seems logical to assume that the traffic will flow more readily. We all know this from our everyday experience of interacting with water flowing in pipes — get a bigger pipe and the water flows more readily. Obviously!

But, hey, not so fast on that assumption.

New York City is famous for having built at great expense the Triborough Bridge to alleviate traffic congestion that was flummoxing the Queensborough Bridge. At first, it seemed like the new bridge did the trick and the flow on the old bridge improved. Gradually, inexorably, the flow on the old bridge returned to its congested mess, and meanwhile even the new bridge also became a congested mess. In theory, that’s not what should have happened, at least in the eyes of the politicians, the transportation planners, and the driving public.

How Latent Demand Can Lead to Increased Traffic

How could this have occurred? Some would argue that it was an example of induced demand, or referred to as “traffic generation” or “induced traffic” or sometimes also described as an example of latent demand. The logic of this argument is that there was demand that wanted to use the bridges but was not doing so, and once the new capacity was added, people perceived that they should go ahead and start driving on the bridges. This then continued until once again the bridge capacities reached their congestion limit, and so any additional latent demand then becomes suppressed and awaits any new supply (more bridges) that might be built.

I can assure you that in Southern California this seems to happen with a frequent and blatantly observable basis. We keep adding additional lanes to our freeways, either by sacrificing an emergency lane or by narrowing existing lanes, and at first we get a moment of congestion relief. In short order, things seem to get gummed up again. This is especially irksome because usually there are promises that once we deal with the agony and delays of the construction to add lanes, it will be worth it since the traffic will improve. Some cynical drivers would say that it wasn’t worth the construction woes and there’s just no point to these nutty and misguided construction projects, other than falsely appeasing the public and putting construction workers to work.

At times, there is a counter argument that the increase in traffic that ends-up consuming the added supply of roadway is actually traffic being now diverted or switching from some other nearby path. For example, suppose you have a congested freeway and roughly parallel to it there is a highway. Suppose that the traffic on the highway decides to switch over to use the freeway, once they see that the freeway has added capacity. Thus, the reason that the new supply gets consumed is that other traffic has opted to now switch into using the new supply.

If you buy into that argument, you would then say that it wasn’t latent demand per se, and instead it was existing demand that has now been diverted over to the new supply. This would also imply that the highway should then exhibit less traffic than it did before, since presumably that traffic is now on the freeway. Well, a number of studies suggest that in most cases the roadway congestion on added roadway capacity cannot be attributed to this switchover traffic. They have not been able to find a commensurate reduction in other traffic that would relate to the newly added capacity of roadway and therefore conclude that the theory of traffic being diverted or switched is not the culprit for the new congestion.

This then brings us back to the theory that it is latent demand. What kind of latent demand would there be, you might ask? Let’s suppose that for a freeway that is currently congested that there are drivers that know about the congestion and are not caught off-guard about the congestion. Suppose further that those drivers know that the congestion is primarily from 7 a.m. in the morning until 9:30 a.m. in the morning. This would be the so-called morning rush hour traffic.

Drivers that know about the morning rush hour traffic might decide that they will wait to go to work until after 9:30 a.m. They somehow make a deal with their employer for this. So, they sit at home, waiting until 9:30 a.m., and then they get into their cars and head onto the freeway. This would be considered an example of latent demand with respect to the morning rush hour. Drivers are not driving during the rush hour because they are trying to avoid it. Once the freeway capacity is increased, those drivers might decide that now they are willing to drive during the morning rush hour and so they switch their work starting time to say 8 a.m. and get onto the freeway around 7 a.m.

Furthermore, suppose there are drivers that would like to go and get some breakfast and then return home. But, suppose they know that the freeway is too crowded and so they instead buy cereal on the weekends and eat just some cereal for their breakfasts. But, once the freeway capacity is increased, and if it appears that the congestion has disappeared, they now decide they want a hot breakfast and so go ahead and drive to get it. This is more latent demand. The added supply is prompting drivers to go ahead and use the added supply.

These examples of latent demand are suggestive that the roadway supply is eventually consumed and that this induced travel or traffic generation is the basis for the eventual re-congestion.

Let’s return to our economics foundations. The drivers that were in the latent demand camp were perceiving that the “price” of driving during the morning rush hour was too high (they therefore avoided driving during the morning rush hour). Once the supply was increased, the perceived price then dropped. Once the perceived price dropped, the demand to use the supply increased. Eventually, the increase in demand consumed the available supply. Ergo, congestion reappeared.

Other forms of potential latent demand for this situation include:

• Drivers that were using alternatives such as bikes or buses, now decide to start using their cars due to the added supply and lessened price.
• Drivers decide to make longer driving trips and so use more of the supply than they had driven before.
• Drivers decide to make more frequent trips, doing several hops, whereas before they might have only been willing to bear a single longer trip on the roadway that was congested.
• And so on.

You might be tempted to conclude that we should never increase roadway capacity because all that will happen is that it will be entirely consumed and we’ll be back to a congested state. That’s not quite the right kind of thinking here. There are many circumstances whereby increasing the roadway capacity might not lead to furtherance of congestion. All that this is pointing out is that we cannot assume that any added roadway will guarantee a reduction in traffic congestion. We need to be aware of and be contemplating the potential for latent demand, which might or might not happen, and might be small or might be large, depending on the specific situation. Trying to make sweeping generalizations that any added roadway capacity will always lead to more congestion is rather nonsensical and certainly misleading.

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

At the Cybernetic Self-Driving Car Institute, we are running simulations about the future of traffic when there are AI self-driving cars on the roadways. This is important for many technological, social, economic, and political reasons to be given due and serious consideration.

Right now, there are self-driving car pundits that are saying we will have zero traffic congestion once we have AI self-driving cars. They paint a rather rosy picture of how wonderful the roads will be. No more bumper to bumper traffic. No more morning or afternoon rush hour clogs. It will be a beautiful sight of self-driving cars flowing at maximum speeds and it will be a breeze to get from one part of town to the other.

This painted picture is appealing to everyone. Politicians are motivated to support self-driving cars. Transportation planners love it. The public is excited to see this all happen. Let’s not wait one minute more. Get those congestion freeing AI self-driving cars on the roads, and like so much Drano will relieve the existing horrors of traffic congestion.

But, hey, not so fast!

Remember the whole discussion herein about latent demand and how it impacts added roadway supply. Let’s consider how that might apply.

Part of the reason that some are so excited about self-driving cars is because it will allow those that otherwise cannot readily drive a car to be able to driven around in a car. This might be elderly that aren’t able to drive, this might be those that are disabled that aren’t able to drive, and so on. You could even say that this includes children – there are many that believe we’ll be putting our children into AI self-driving cars and sending them to school via that means, rather than having to go with them as an adult and drive them to school.

We could consider all of those existing non-drivers as latent demand. Once the AI self-driving car is readily available, there will presumably be a ton of latent demand that will come out of the woodwork.

Will our existing roadways be able to handle this?

Might all of these AI self-driving cars radically increase the number of cars on the roadways, and the number of miles driven, and the number of trips taken? It sure seems like it will. Essentially, we are “reducing” the “price” of driving in the sense that those that could not otherwise afford to be driven before (due to the cost of say hiring a chauffeur), will now be able to “afford” the cost to be driven around. This decrease in pricing will increase the demand, which will soak up the supply of roadways.

You might say that ridesharing has already started us down this same path. By having available Uber and Lyft and other ridesharing services, the cost to be driven around has been reduced, at least in terms of the access to ridesharing versus the prior taxi-led approach which had all sorts of frictional costs involved.

AI Self-Driving Cars Seen Boosting Ridesharing

AI self-driving cars will presumably boost ridesharing exponentially.

Anyone that owns an AI self-driving car is possibly going to be using their self-driving car as their own personal ridesharing service for others to pay to use. You don’t have to be an Uber or Lyft driver. You just somehow advertise that your AI self-driving car is available for ridesharing and voila you can make money off your AI self-driving car.

This is why Uber and Lyft are furiously trying to get into self-driving cars themselves, since they can see the handwriting on the wall that their existing business model of having human drivers is going to eventually go away and they could therefore be another example of a disrupted industry by a new technology (in this case, the AI self-driving car).  Will such ridesharing services simply be the conduit to connect those that own AI self-driving cars that want to make available their self-driving car for ridesharing with those that need a lift? Couldn’t a Facebook do this instead? That’s what scares all of the existing ridesharing services.

Think also about other ways that AI self-driving cars might increase driving.

Some are saying that you could move out of the downtown city area and live in the suburbs, because with an AI self-driving car you can have it pick you up in the morning, you sleep on the way to work, and no need to worry about that morning commute. Our whole pattern of where we live and where we work could change. Distance between us and work, or us and the mall (if malls still exist!), and so on, won’t really matter since we have an AI self-driving car that will take us wherever we want to go. You don’t need to know how to drive. You don’t need to stay awake and be able to drive for ten hours straight. You won’t need to find a back-up driver so you can switch during long trips. It’s all driving being done by the AI.

You might decide to ditch taking the bus. You might not ride your bike. You might instead decide to go in that cool AI self-driving car instead. The amount of latent demand is potentially enormous. The AI self-driving car might become the most traffic inducing, traffic generator of all time.

We should be careful in assuming that AI self-driving cars will get us to the vaunted zero congestion.

No matter how well the AI self-driving cars drive, and how well coordinated they are, volume is still volume.

There are some that criticize the induced demand theory as somewhat hogwash-like in that the critics claim that when roadways are expanded and become clogged it isn’t the traveling that causes this, but instead there is an economic benefit that people must perceive and so the basis for their traveling. If people get into their cars and are driving on these expanded roadways, it is presumably because they see an economic benefit in doing so. Therefore, the increase in cars on the roadways is a good thing in that people are gaining more economic benefit. People are traveling for a purpose and we need to consider the larger picture of how economically there is a collective benefit involved. I won’t take the space and time here to detail their argument, but want to at least make you aware of it and you can then postulate it and consider researching more about it, if you like.  In short, if indeed AI self-driving cars produces congestion, it in a sense is a reflection that we’ll have a lot more people gaining economic benefit by making those driving trips than otherwise if they didn’t.

Let me give you a personal example of how this induced demand might impact someone’s driving and miles traveled. When my children were in middle school, I used to drive them to school in the morning (this was a delight to do, and I miss it dearly!). Suppose we use the letter M to represent my son, and the letter L to represent my daughter. My son has a friend that we’ll label G, and my daughter has a friend we’ll label S. We had a custom of getting bagels and donuts in the morning, and I’ll label the donut shop as D.

Here was a daily morning commute:

(M+L) + G + S + D

This meant that M and L got into our car, I drove us to pick-up my son’s friend G, we then drove to pick-up my daughter’s friend S, and then we stopped at the D to get some grub. We used my car to do so this, so we have 1 vehicle, or V1. The distance traveled was 3 miles in total.

Let’s represent this as:

V1: (M+L) + G + S + D = 3 miles

Now, somedays, my daughter was behind schedule (time was a key factor in these morning trips!), and it was prudent to go ahead and do this:

V1: M + G + L + S + D = 5 miles

This shows that M got into the car with me, we went and picked up his friend G, we came back to the home to get L, we then went to get S, and then to D.  The total miles is now 5 miles, because of the trip to get G and come back to our house. It was worth the extra distance because we were trying to beat the clock.

Some mornings, I would get up really early, and do this:

V1: D + (M + L) + G + S = 7 miles

I would go get the D grub, then drive back home, and proceed with the rest of the sequence. The number of miles of the morning trip has now risen to 7 miles. It made sense because I then had the D in-hand and this saved us time once the rest of the sequence occurred.

Some mornings, we were running late, and we’d ask the parents of G and of S to drop-off their kids at our house:

V3: [G] + [S] + (M+L) + D = 10 miles

Notice now that we had three cars involved, including the cars of the parents for G and S. This now makes what otherwise would have been a 3 mile trip into a now 10 mile trip.

How does this relate to AI self-driving cars?

The odds are that this 10 mile trip would be the likely candidate once we have AI self-driving cars. We would either send our AI self-driving car to go get G and S and bring them to our house, or maybe the parents would send their kids to our house via their AI self-driving cars.

The point is that with AI self-driving cars it will be much easier to go ahead and have a driving trip undertaken. It seems very likely that the number of trips, the length of trips, and the number of miles traveled are going to go up, probably a lot.

What we don’t yet know is what will be the cost of the AI self-driving cars? Right now, everyone seems to be assuming, either implicitly or explicitly, that the cost of the AI self-driving car is about the same as driving any conventional car. Suppose though that the cost of a self-driving car is a lot higher than conventional cars? This could increase the “price” of making use of self-driving cars and therefore not lead us down the path of the price decrease (which led us to the demand increase and the supply consumption).  When you consider a conventional car, you acknowledge that having a human driver means there’s an additional price or cost associated with the use of the car. Is it the case that the AI self-driving car “driver” will be less than that cost, the same as, or more than that cost?

We’ve been running various simulations, taking a look at mixtures of the price aspects, along with also considering the mix of human driven cars versus AI self-driving cars. I say this because it is unrealistic to assume that suddenly one day we will have instantaneously all self-driving cars on the road. We won’t. We instead will have a mix of both human driven cars and AI driven cars. Eventually, presumably, the number of AI driven cars will gradually overtake the number of human driven cars, and maybe someday it will only be AI driven cars (there’s a lot of controversy on that point!).  Anyway, all of us need to put some sober, serious thought toward the future of our car travel and consider how this will impact our habits, how we live, where we live, the roadways, and the rest. It’s an issue, a large problem to be dealt with, and the platitudes of “zero congestion” need to be carefully scrutinized, and in fact maybe we should substitute instead the more thought provoking phrase of “induced demand.”

This content is originally posted on AI Trends.