Skip to main content

Introduction by Professor Mark Hoffman, Dean, UNSW Engineering

Welcome ladies and gentlemen to today's Learn@Lunch, which is a bite size lecture series for people such as yourself to hear from some of UNSW leading researchers. My name is Mark Hoffman and I am Dean of Engineering at UNSW. Before we get started I'd like to acknowledge the Gadigal People of the Eora Nation, the traditional custodians of this land. I'd like to pay my respects to the elders past and present. In the 70 years since UNSW started, we've continued to refine and grow our commitment to providing lifelong learning opportunities for university alumni and friends and so we're delighted that you can join us here today. I'd also like to thank those in attendance from our scientist circle, our supporters who enabled UNSW to advance knowledge through research and education by giving in their will. Now a few housekeeping notes before we get started, before I introduce the speaker, today's session will be recorded and available for podcasting on the UNSW alumni website. So also please have your mobile phones switched off, and I'll do that when I sit down.

Now today we have a very exciting talk with the subtitle, Why Commuters Should Exit The Driver Seat. At our fingertips now, we have the means to avoid traffic congestion with Google maps with ways we can choose public transport with less walking through the UNSW transport trip planner. But everybody has this information, so how is it going to change a travel behaviour? To take us through this, it's my pleasure to introduce today's speaker. Professor Travis Waller is the advisee and chair of Transport Innovation and also Deputy Dean research in the Faculty of Engineering, and the executive director for the Research Centre for Integrated Transport Innovation known as rCITI at The University of New South Wales. Professor Waller has received extensive accolades and awards for research and work, including a 100 Top Innovations under 35, and the US Transportation Research Board's Fred Burggraf, the Hojjat Adeli award for Innovations in Computing, and most recently this year was awarded the Pyke Johnson award from the Transportation Research Board, a division of the US National Academies.

In the past 18 years, professor Waller has secured over $26 million in total research funding. He's published over 300 peer reviewed papers, supervised 24 PhD students to completion and conducted over 80 funded research proposals with 30 distinct different sponsors worldwide. And that I think is what's really important about UNSW and Travis. Please join me in welcoming professor Travis Waller.

3:05 Learn@Lunch presentation by Professor S Travis Waller, Advisian Chair of Transport Innovation, UNSW Engineering

Thank you Mark, I really appreciate that. Let's move on from this photo, I don't think that's a terribly accurate representation. Is there a trick to this? Oh, there we go. It's come alive. So, I want to talk about three disruptions today. Pricing, information and autonomy. You can think of autonomy or automation and there are differences, but I'll just sort of group them up for the purposes of today, and I want to talk about these three disruptions and what they're doing to our society's mobility systems. And how to think about that and maybe start to get to the point of how do we deal with that. Now to do that, let's start very quickly one level up and think about the shifting landscape for mobility and planning for it and society's mobility and there are really three key drivers, and these are impacting society across the board in many different domains.

Technology quite obviously accelerating all of the time, very much so for mobility as well. Electric vehicles, automated vehicles, high speed rail, and many other communication technologies and then data, the rise of data both in terms of characterising human activity. There's so much data on us in what we do and the system as a whole and being able to understand it. And then very critically, behaviour, which is the human response to those things. As technology and data and machine learning and AI and everything else that falls in those categories is changing, the human behaviour response changes as well, both as individuals but then collectively as well, and the collective behaviour might not correspond to the individual behaviour and is that distinction that is very characteristic of mobilities and many other systems and I really want to zoom in on that today.

Now, when dealing with the disruption, it is useful to think for a moment, what is it doing? What does the disruption disrupt? Why is it special? It's special because it has the capacity to potentially sweep away all of the structures that we have built on foundational principles. And because of that, to understand the potential implications of disruption, we must return to foundational principles, first principles. Now when doing that as a professor, I run a risk, and the risk is that you think that I'm ivory tower in theoretical, and I am, but I also try to be useful because I think engineers need to be useful. We solve problems, otherwise I shouldn't call myself an engineer. So it is theoretical, and I'll talk about some very simplified examples today and some high level theoretical concepts. But I would argue that's necessary in the of disruption. But our overall aim is something that's useful and practical eventually and linking those two up is a challenge.

And I would argue that one way of highlighting that this is even if it seems theoretical, even if it seems simple and abstract, it is impactful and one way of highlighting that is I always like to think past industry sponsors. So now over the last 20 years, it's been over 40 different industries and governmental agencies that have funded the research I'm about to present. And I would say someone somewhere around the world has, I have been able to convince that it is practical even at first if it seems theoretical. So being a researcher, we love questions and if we're thinking foundational, what are three foundational questions for mobility? What traveller cost? What is the cost to travellers? What do they pay or what costs are they incurring in terms of externalities that they may or may not even be paying? Regardless, what is the cost? What will be known? What will people know when they're making the choices that they will make? And mobility is characterised by choices.

It's human choice drives that dynamic and what will they know? And for much of the history of transport, mobility, it's primarily been historic personal experience. Two weeks ago I took the train, and it took me 28 minutes. Today it took me 23. I took this... I drove this route and took me 15 minutes, now it took me 32, and this experience that we build up informs our choices and what tech will be used and this, if you look over the last several hundred years has changed radically as well. From horse, to automobile to maybe electric to maybe automated and rail and so on. All of these are being disrupted. So traveller costs, tailored pricing, the whole rise of mobility as a service, a return maybe to the concepts of congestion pricing as a more tailored approach. This potentially is being disrupted. What will be known with connected information, the historic personal experience that is now gone and very disrupted as well and what tech will be used. Obviously autonomous vehicles and that's only one example of the technology.

So all of these being disrupted at the same time, any one disruption has the ability to transform, to revolutionise, and it's being disrupted in multiple dimensions simultaneously. And all of those come to a point when it comes again to behaviour, individual behaviour and system. Now I've used this word behaviour multiple times and at the risk of seeming academic, I want to do a little example, and some have seen this, and I apologise for repetition, or some have seen it not from me, but I'll do it and I'll do it quickly. But I'll start with this. So consider a city network. I intentionally try to pick one that most people in the room probably don't have a lot of familiarity with and that was intentional because the examples will be hypothetical. Don't worry about the specific street or experience just to put the setting.

But I want to explain that while we zoom in and look at one little tiny thing, that concept and those methods have to scale up to this. We have to come up with tools to quantify, but that work at scale and this is the smallest of the scales that will do. But we'll zoom in. So I'll just pick by random a place, and we zoom in, and we think about this, and the goal is through these examples to get some insight into quantified methods that simultaneously account for individual choice, quantifying a human beings individual choice making capacity, mathematically and the resulting congestion and other costs that will be created by that choice. So representing choice and the impact of that choice and that requires mathematical representation and some degree of approximation. So zooming in, the simplest sort of system that would make any sense whatsoever.

Two paths connecting two points. I'm at A down at the bottom, and I want to get to D and there are two paths. One that goes through B and one through C. Now mathematically, the way we think about this, and practically if we're out, and I'm going to use road examples, but it's equally applicable to multimodal, to freight to other forms of networks. I do road because that has a nice intuitive feel for many people. Now the choice is, which path? Through B or C? Discrete choice, simple choice, B or C? The impact is congestion. The more people that make a choice, the greater the cost. You can think of a roadway that's empty at 3:00 AM. If you travel, there's a certain travel time. If two people travel, the travel times greater. If 20, 50, 100 people the travel time goes up. The travel time is a function of the number of people that make that choice. So it's a function. A function looks something roughly like this, it's increasing. So cost, time, money, whatever increases with flow. Easy peasy.

So, simple. These can be much more complicated functions, but it's not I want to keep it as simple as possible, and we'll say on this road segment from A to B, that's a particular road segment, and the cost is X divided by 10. X is the number of people. So if a hundred people choose that, the travel time, let's say is 10. 10 what? 10 minutes, 10 hours, $10. We'll say minutes. It's 10 minutes. If a hundred people choose it, divide by 10 that's the travel time. For simplicity so we can do it in our head I've kept simple functions, but don't let that hang you up. These can be much more complicated functions and everything would hold. Now to keep it simple. Also, I'll fix the cost on this link to be seven, it could be a function, but again, to keep it simple, I'll fix it. The examples still holds, so seven minutes no matter what. And the bottom is the same but switched from A to C seven, from C to D, X over ten. So the number of people divided by 10. So that's the cost response. So we have two paths, ABD, ACD, and the costs are functions of flow.

Let's say that we've done a demand study, so that's a different field of mathematics. That's statistics land use surveys Let's say we've done a study, and we've determined there will be 50 people at A that want to go to D. That's demand, 50. Then our question is what choice will they make? What's the rational market economic choice and the choice, the economic choices well they'll do what's best for themselves. They'll make the choice that self-optimising, and they'll play a game. It's a Nash equilibrium actually. They'll play a game, and the market will only stabilise when no one could unilaterally switch and do better. The non-coordination. We don't phone each other in the morning right now and say if you take this path, and I take this path, it'll be better for us both. We each make our own independent self-optimising choice and that leads to an equilibrium.

So what is that? Well, this is easy because the cost structure is symmetric. So they'll split in half, 25 of the people will go through B, 25 through C. So if 25 people are on this link, 25 divided by 10 is 2.5. 2.5 plus seven, 9.5. The bottom is the same in reverse. Everyone in the system will have a travel time of 9.5 minutes. That's an equilibrium, the market stabilises, that's the solution. And you can do this as a forecast. This is my 30 year, 50 year model off into the future. Now, let's say we look at this intuitively, and we say each path has a slow link, and a fast link. There's a 2.5 and a seven on the top and there's a seven and a 2.5 on the bottom. The sevens are inefficient. We should connect the two fast links. We should invest money, do a PPP or do a bond or something and get money and build a road connecting B to C, connecting the two fast links. So we build a business case. We can make this much more complicated and hold, but I'm doing it at one, so we can do it in our head.

How will the choice change? Well, this adds a new path ABCD, and at first that path is very attractive. Right now there's 25 people up top, 25 at the bottom. On the day this opens, this is 2.5 plus one, plus 2.5 that's six. Six is much better than 9.5, people will switch, I'll switch and someone else switches and someone else switches. That's a better option. What happens once we all switch? Well, now it's 50 over 10, plus one plus 50 over 10, that's 11. That's worse than the 9.5. Well, maybe I should switch back. We look at the numbers. The other paths are both worse because it's 50 over 10 plus seven that's 12. Five plus seven, so the other two paths are worse. We've raised money, we've built infrastructure because of human behaviour they move around, they're all worse off, and they won't move themselves back into the old option that was better.

This can happen. This is the unintended consequences of human behaviour. Networks are complicated, we cannot rely on common sense intuition or 20, 30, 40 years of experience to know how a network will respond. Mathematics and characterization are the only things we can use. And this is only one tiny little example. And this was formalised all the way back in 1956, but it's used to this day in treasuries around the world to decide how trillions of dollars of infrastructure get invested because human beings cannot just do it intuitively. And so while that was a tiny example, the formulation is used to code algorithms, which are then embedded in software and used for massive, very large networks around the world. And that's only one choice. It's only route choice. It's only one particular little choice in the big spectrum of things we have to account for.

Housing choices, land use, trip making. Do I make the trip or not? What time do I depart? My departure time choice, my mode choice. Am I using shared mobility? That's a new one that's a reason that we have to characterise. Tow usage, do I use a tow or not? Mobility is a service, and many others are all these choices, and most of them can have this counterintuitive behaviour. And in fact, to highlight this, there was research on route and mode choice that led to the theory of multinomial logit modelling, which is a discrete choice and is used across fields of economics now and econometrics. And it was actually the road mode choice research that led to the Nobel prize in economics in 2000 for professor Daniel McFadden. So it's a very serious research that's required to illuminate and characterise this behaviour. And without it we can get some very paradoxical outcomes when we face these new disruptions.

And over the past 60 years there has been an advance, the model that I showed ignored very many things, time evolution. It was an average steady state sort of model. As we all know, transport one minute to the next is very different, there's a time dynamics. The old treasury models are not accounting for that. Stochasticity inherently, information, the fact that people do not have perfect information. Again, missing from any of the models and so on and so on. And so it's advancing our understanding of the underlying behaviour to be able to account for these technological and sometimes policy based new realities. So that was mainly just, so we can get a bit of an insight. What am I saying when I say behaviour? That's what I'm talking about. So back to pricing. So disruption number one. This price, is pricing really a disruption? It's an old... we've always had it, we've been talking about it for in some cases centuries why now? One is most infrastructure around the world, transporting infrastructure is from petrol taxes.

If electric vehicles take off in a very big way, that will require a fundamental shift in the way we fund infrastructure, which is already coming under strain. Not just building but maintaining even. So maintaining, managing and so on. A system benefits, understanding that the current... current pricing approaches are very aggregate. For instance, if you're driving, petrol taxes is the main way that we price mobility for the driving portion anyway and if you're driving on the Sydney Harbour Bridge at 9:00 AM on a Friday, you pay the same petrol tax as if you're driving on a rural road at 3:00 AM on a Tuesday. That is inherently unfair because the externalities that you're causing society are very different in those two situations. Now, congestion pricing has been actively looked at for a very long time. There've been some approaches, Singapore, London and otherwise in a very aggregate way and an argument that is often used is, well, we're not doing more there because it's politically difficult.

It is very politically difficult, but I would argue that's not the reason we've not done it. The reason we've not done it is because it has been to date technically impossible to do. Why? Because, market economics will show efficiently priced market. The prices must change just as quickly as the market. If they don't, that introduces inefficiencies. How quickly to transport conditions change second to second. So the prices would have to change second to second, to have it truly be an efficiently priced market. That has two main issues. One, we need to have perfect observability of the system to even know what the price is while simultaneously communicating that price back out. That would require a completely different information infrastructure, which is one of the three disruptions is ICT. The other is the human brain's not capable of accounting for cheap prices that change every second. That's the third disruption on autonomy, the car, the computer, replacing the human brain. So the disruptions are sweeping away what have been the barriers to doing this properly.

Yes, still politically difficult, but at least now it might be a fight worth having because it's technologically possible to envision. And finally mobility as a service. Mobility as a service is the market response. It is moving in to fill the gap left by these inefficiencies. And I view mobility as a service primarily as a pricing innovation. It's providing a sought after level of service at a better price. And that's through personalised options, on demand journey planning, bundle packages, and identifying gaps in overall mobility services. But again, I view personally that as a pricing innovation in the same way that the Amazon cloud really is a pricing innovation if you're wanting computer power. You can buy a large server, or you can buy exactly the cycles you need. It's a similar concept here. You can buy exactly the mobility you need instead of buying a whole car.

So this leads us back to a very old thing. So AC Pigou was a very famous economist from a hundred years ago who actually articulated out completely the concept of congestion pricing and the fundamental first principles. And so when I say return to first principles, this is part of it. Sweep away the structures that we've built over the last hundred years with the automobile and return to first principles of how people actually behave and re-envision how the systems of structures should be based on fundamental economic behaviour. That's how to address a disruption and this is laid out. This is understood economically. So that was disruption one. How are we on time? Okay. Disruption two, information. So information is exploding on every front. So, between Google, telecommunications, phone apps that are collecting data, we are able to characterise the activity of human beings to a shocking degree.

And this serves as a potential replacement for the traditional processes of urban planning, which are mainly household surveys, loop detectors buried under the roadway. These things I'm of the opinion will die off. We know increasingly so much, but what's lacking... the research I've been working on, what are the methodologies to quantify this in a reliable way for planning. That's where the research comes in. Social media in addition to Google and then also financial. We had a partnership with MasterCard, and we're getting the data, you can tell massive amount of human behaviour, mobility and otherwise from credit card transactions made anonymous. But someone bought a coffee at this Starbucks at 10:12 AM, so they're here at this time, and they got lunch over here at this time, and you could really tell about a person's activity pattern, travel pattern. So exploding. This is some work by colleague of mine, associate professor Taha Rashidi in 2017 using Twitter data to characterise Sydney Vivid.

So no other transport sort of data, just Twitter data and able to sort of describe the human mobility around Sydney Vivid. Google is a very active player in this space. So, Mark mentioned our city earlier. So our city is first the entity outside of America to have the Google maps outreach grant. It was an in kind grant, most likely valued at a few million because it was complete access to all Google travel times, and some other data that they have. Every city, every roadway, everywhere around the world, every five minutes for a year. And so we actually archived that, used quite a lot of space so that we can start to come up with mathematical ways to replace their traditional planning process and researchers in the centre have been working now and using that, de-validate some models that I will talk about in a moment, but then also look at new initiatives on traffic management and leapfrogging physical infrastructure with some partner countries.

But all of that's on analytics and information is bi-directional. Analytics is at most half of the problem or opportunity depending on how you're looking at it. Information also transfers out. Yes, we can talk about analytics, observe the system, get the information in. But at the same time, information is flowing out, changing the behaviour of the system you're monitoring. And if you build your analytics based on old assumptions about behaviour, the analytics will actually not be very useful. It's bit like Quantum theory, you can't observe something without disturbing it. So it can't just be analytics. It's bi-directional. Do this again, very quickly and I'll not do it to depth, but so let's think of an individual now. So let's think only selfishly where one person, and we're travelling from A to D and this is the original Google map sort of thing. How do I get from A to D? And so as a single person, I won't care about the cost impact of what I'm doing.

It's just what's my choice, self-optimising. And so I can treat it as fixed cost, and will say one, one, one, three, four. I'll just say these are the travel times what's my shortest path? Well, ABD is one four that's five. ACD is three one that's four. If I get A to B to C to D one, one, one, that's three. My shortest path is ABCD, cost of three. That's what I'll do, that's my shortest path. Well, let's say now that I actually link BD is not really four, it's a very dangerous link. Half the time it's a one, half the time it's a seven. Now, this is from some earlier stochastic shortest path work I did with tenacity scalpels back in 2002. So it's an adaptive shortest path because half the time it's one half seven, so how do we deal with that? Well, if I don't have information, it's the same as before because I don't know. If it's one or seven half the time it's one half seven. You can have a few different behavioural assumptions around that, but one is I do the average, the average is four, and I avoid the link.

But if I have information, well the way that we solved it is we made something called a hyper path, which is a conditional path. Instead of thinking of paths now you think of paths based on the information states that you might learn as you're moving. And there will be five strategies because if I go ACD, well that stays the same because there's no information, there's no choice. If I go AB, at B I learn, and I can adapt. And so I condition my strategies based on what I learn. And there are two States. One or seven. So state one, state two. I can learn that that link is in state one or in state seven one or two. And the strategy... the optimal strategy is now hybrid. I go AB and if at B and I learn that BD is in state one, I do link BD and I get a cost of one, and my overall cost will be two with probability 0.5. If I learn it's in state two, I go on to C and D one, one, one I get three. So half the time I'll get two half three. My overall cost is now on average 2.5. So through information and adaptivity, I've changed from an average cost of three to an average of 2.5 that's a benefit to me.

I benefit from information in ITS. The challenge, you know, and we introduced some algorithms for this. There are three different cases I won't go more into that, but the issue with this is it only works for fixed costs. As we saw in an earlier example, costs are not fixed. They change with flow. So it's not useful to use that viewpoint when considering system behaviour. So we introduced a new model. This was with a former PhD student of mine in 2006, which we called user equilibrium with resource, its adaptive equilibrium. And we, it's a Beckmann style formulation with a conditional probability matrix and flow matrix, but it has the same equilibrium conditions, and it has all of the properties of the Beckmann model. And the core idea is if we re-introduce functions, so this link is X, this is X , and this middle link is our uncertain link. I won't go through these details that much, but what I'll say is without information, if we just take the average, and we assume people don't know, and they play the earlier equilibrium game, it'll will calibrate to a cost of 16.

Without information people will adopt a certain strategy and the cost of 16. If we provide information, so 12 travellers let's say, there's five hyper paths and people will equilibrate over the hyper paths, and we can't step through that. This one, you actually need a computer to help with the optimal strategy being, you know, the equilibrium strategy being four people follow hyper path one, no one follows strategies two, three. Three people follow hyper path strategy four, five follow hyper path strategy five, and they each have a cost of 18. So we give information to everyone. They make a perfectly rational economic choice. Everyone gets worse off. And that may seem counterintuitive, but the reason is the same as the first example. People optimise themselves they're not optimising the system. It's a tragedy of the common sort of outcome.

There's no guarantee by building infrastructure or even giving information, there's no guarantee that the system will do better. Now, the question from there becomes, but do people really behave this way? This is a beautiful economic model, built on some first principles. It's relying on very safe economic assumptions people are self-optimising, but do people really behave that way? Well, for that, since introducing this model, we've turned to a few things. One is the field of experimental economics. Another colleague of mine, professor Vinayak Dixit is very good experimental economics. I'm asking the questions how to real people play this game. And we've looked with polling and incentivized games driving lab experiments and more recently looking at pervasive data like the Google data that we have archived and there's a long history of publications using experimental economics but relatively recently last decade or two for these traffic transport mobility behaviour questions, and it's been viewed as a very useful tool.

So we conducted an experiment in collaboration with French National Research lab, IFSTTAR led by Jean-Luc Ygnace who's a former PhD student of mine now a lecturer at UTS and published in 2017 looking at exactly this question. So 144 participants playing this game iteratively seeing where they would end up. I'm not going to go through all the tiny data except that this is basically the game play over time and people do equilibrate as the models will predict the basic cases. This is sort of the beginning of a statistical analysis thing. Are the results statistically significant? The red being the cost of the new information case, the blue the cost of the information case and with statistical significance as published in plus, yes, we can observe through rational game-playing behaviour over the 144 participants with no information they actually did better than with the cost was lower information induced a collective detriment.

Now what does this all mean? The lesson is this, in the absence of deception, inducement or pricing. The power of information is that it makes us more efficient at being selfish. We are better ruthless self optimizers. It doesn't make us better people. It doesn't make us socially conscious necessarily. It makes us more efficient at being selfish. And sometimes it's good if everybody's more productive that might be good, but there may be situations where that's bad tragedy of outcomes... tragedy of the common outcomes. And we have to acknowledge this, characterise it and plan for it. We cannot pretend that, Oh, if we tell everybody everything, they'll do the right thing. Doesn't work that way. When it comes to many of these sort of questions that has been the typical response. So that's the takeaway is we have to mathematically characterise this behaviour and account for it in our planning else we are guaranteed to get unintended consequences of massive scale, especially when you start talking about trillions of dollars of investment.

That was two disruptions. Third, autonomous driving, automated driving, depending on how you want to think of it quickly now. Our AV's safer says the first. That's usually the selling point. Generally, yes, but even some human factors experiments are saying no, and in some very old known ways, for instance, it's been shown now, there's a study from the University of Michigan showing that when human beings realised they're driving alongside an AV, there's an overwhelming tendency to behave more aggressively because they assume that the technology will compensate. There by leading to an overall less safe situation. That's the same thing as seat belts when they were introduced, people driving more aggressively thinking it would save their life. This is always a concern, especially in the transition period. But there is lots of questions around that safety issue.

There are many more things we can go down that track in terms of the driver is responsible and insurance and things from there. But then also on efficiency. Are they more efficient? We can ask that in terms of an individual organ for society. As an individual, the answer might be very clearly yes. But as we saw from the earlier examples, that doesn't necessarily mean that's true for society. So we've been looking at this actually for a very long time. So this is a project started. It was between myself, and a professor of computer science back at the university of Texas, Austin back in 2007, and we weren't trying to build the car, he worked on building the cars, but that wasn't this project. It wasn't about building the car. It was about understanding how the cars might change our systems. And to my knowledge, it was the first major project in the world looking at the potential impact of AV's on urban congestion.

And we simulated it, we modelled it, we started to understand what's going on and how will they change things. And from there, since the time establishing our city, we've continued that research but also built into consumer behaviour, the uptake of AV's and other technologies, traveller behaviour as I've been talking a lot about today. And then how that traveller behaviour relates into system behaviour. This is some work by a colleague of mine, doctor Miad Saberi who's a senior lecturer in the centre who came in from another university on stability of AV's. So even before you get to the societal questions in terms of the microscopic driving characteristics, they're not necessarily stable. You get many of the same traffic flow... fluid dynamic instabilities that you would from humans unless they're synchronised, unless there is connection. So you need information, connectivity, even if they are autonomous, you optimising entities even at the local level.

At the broader societal level. So some of the most recent research we've been doing, so we've been building a host of models, both in terms of the network models I was talking about earlier, but also in the computable general equilibrium space, which is economic modelling and publications over there on the right and linking them. And I believe first in the world actually linking these two types of modelling so simultaneously we can account for these new technologies while looking at things like income utility and social welfare. So being able to put numbers on these things. And the point is, from quantifying AV's or any other technology, putting a number on what's the change in the transport cost, what's the change in the transport benefit and what's the change in welfare per capita? And it's complicated because if you think of the overall sort of supply chain of behaviour and dealing with these AV's, it's really about removing the trip ends.

So in freight, very often it's the first mile, last mile problem. And even for our own personal behaviour it's often that, if you own a car where do you park it? Where do you get it in? AV's might break that concern, which creates a new form of externality, something people used to care about, but they might not now need to. And that creates a new externality that might have a societal detriment amongst many others. And I won't go through numbers because they're done. We actually did Sydney is one of the test cases, and it's currently in scientific peer review. So I won't mention the numbers, but I am confident enough to say that the number starts with a negative sign. So ultimately, AV's or information save us. And ultimately my response to that would be no, not on their own. Information makes us efficient. Yes, but it does not make us better. And that has to be embraced and understood. And that means that we have to align costs and pricing and broader platforms to that reality, to societal objectives. So that personal costs and incentives align to the societal impact.

And technology provides us tools of doing that new ways. And for AV's, you know something to keep in mind for AV's because there's a strong tendency to say AV's will fix all this because they'll do what's right. Most governments I've spoken to around the world, we trust the markets and that's appropriate but that implies that the decisions will still primarily lie with the traveller. The AV will be an agent of the traveller, and they will be ruthlessly self-optimising, so we cannot rely upon AV's solving this for us. They will just be travellers on steroids. They will be to the millisecond optimising every small gap of possible utility for their passenger and that has to be acknowledged and accounted for. Less life one way of thinking of this AV's on their own will still be self-interested, self-focused and on top of that they might still be unstable in a localised sense, in a traffic dynamic sense. AV's plus information where you might be able to cure the instability, the fluid dynamic volatility and smooth traffic that may be possible connected vehicles, autonomous vehicles.

But without pricing they'll still be self-focused only by integrating through the technology with the understanding of information and its behaviour impacts with new pricing mechanisms can that actually be orchestrated and coordinated to become societally focused. And with that I am done.

39:50 Q+A session with audience

Mark Hoffman:                 Thank you very much Travis. We've now got some time for some questions. If I could ask people who wish to ask questions to put your hand up. We have some roving microphones and also if you could try and keep your question concise and right to the point of the question, it'd be really good.

Speaker 1:                           You mentioned quite early on that there's a political cost to try to introduce pricing and things like that. And one thing I've heard talked about a lot is that some travellers are unable to afford price increases and that can cause those travellers to stop travelling at all and just remove them from the question. And one of the political things is that the low income earners, if you try and do this kind of pricing for society, they stop being able to work because they can't travel. So I'm wondering how you look at that kind of a problem with all of this?

Travis Waller:                     Absolutely. I try to stray beyond my domain expertise. I have worked with many economists in this space. So for over a decade I collaborated a lot with professor Kara Kockelman who's a professor of Transport Economics back at University of Texas who worked on the idea of equity fairness, credit-based congestion pricing mechanisms and one thing we often stress is that, and I think that maybe the cynicism is well earned maybe it's not, the public feeling whenever we talk about pricing, we're talking about higher prices. That need not necessarily be the case. Tailored pricing does not necessarily mean inflated. That from my point of view, the question of funding is separate from the question of efficiency in fairness and societally aligned. So it might not be the prices go up, but that they are tailored to the activity that people are wanting to do.

And when it comes to socio demographics to consider potentially some credit based systems. I would argue there are multiple mechanisms of trying to address the fairness and getting away from this concept that anytime we talk about pricing, we're just talking about adding onto the price up and up because we want to build some new roadway. It's not that, so tailoring does not mean increase. So it might coincide with decreases in Petrol tax or decreases in some other costs that those people are already paying, but tailoring it to certain behaviours versus others. But on aggregate they might not be paying anything more than what they're paying right now.

Speaker 2:                           Yes. Assume you have a tunnel for traffic using gravity models. And my projections by the way came out true unlike the cross city tunnel, but that's not the story. That work you showed it was non saturated, and we found... this is in the 80s, so a long time ago. All this modelling tends to break down in saturated networks. You've comment about that?

Travis Waller:                     Yeah. A few things as in either low congestion or high. So there's two concepts within the traffic modelling user equilibrium system optimality and there are different sorts of modelling to be done. So traffic dynamics versus latent demand, planning sort of networks. And so for the planning sort of light and demand modelling, you will get results that at low congestion, your system optimising and self-optimising tend to be similar, and your highly congested tend to be similar, and it's the middle area will differ the most. The argument is that even if they start moving closer, there is still a gap and with efficiency gains of technology that creates more of an opportunity for that gap. So suppose it depends on if you're just wanting me to comment broadly on the applicability of the modelling to different scenarios or if this holds under high congestion. This will hold under high congestion because even under very high congestion system objectives do not align to self-objectives and people behave in a self-optimising way.

Speaker 3:                           I'm struggling a little bit because of my background, but coming back to fundamentals. Are you saying that in the long run people are going to use a personal vehicle, probably on the road. Background to my question is, should we therefore in the public sector be contributing more money to building roads and building railways?

Travis Waller:                     No, I'm absolutely not saying that. The core thesis I want to convey is that be it roads or rail or anything else, human beings are self-optimising and acknowledging that reality makes us consider our options more carefully. And so I've used road examples, but in terms of the self... if it is more in interest, if based on someone's personal utility function, if it did make sense for them to use rail, they would, but they would make that decision based on a self-optimization principle versus a system.

Speaker 4:                           All right. Maybe not on quite on the topic, but you've been talking a lot about autonomous vehicles, and I had a very interesting discussion with a chap from Israel about a month ago about autonomy and autonomous vehicles and we had a disagreement about the spread or the introduction, right of introduction of autonomous vehicles. I was feeling that they would come sooner. He was arguing, they would come much later. Just wondering if you have any thoughts about that.

Travis Waller:                     That's a very tricky one. I was at Toyota giving a talk to their employees four, five months ago. The manufacturers are keen, but anything new they like because that provides a marketing bump for them. But the degree of functionality could be questionable. Part of it depends on what do you define as an automated vehicle. Is it one that's going to pick up your kids without you in it and do everything on its own, and you not have to worry about, or is it something that's a few steps beyond cruise control. So I think it's going to be incremental in terms of this technology, and it will from those I speak with. The particular gaps will be those making the most economic sense. Interstate freight when it first really breaks through, it'll likely be in those economic streams. I do think it will happen, but I'm not the most of the optimistic.

Speaker 6:                           Just want to know, do you include government resources provided to subsidise other sort of transport in your equation? 

Travis Waller:                     Some of them yeah, usually scenario based. So we... My personal sort of philosophy and even as a centre we try not to have a strong policy viewpoints, but if there are specific policy options we will quantitatively evaluate them with the modelling. So there are cases where we will evaluate if there are particular pricing structures or schemes or investments, how that will change things.

Speaker 7:                           Travis, it seems like if we all act selfishly the society loses. We have systems in place today like SCAT that try to optimally balance road traffic. It's clear then with autonomous vehicles, government, or maybe not market needs to then step in even more than they currently do because if we all act selfishly, it's going to be perhaps as you said, worse on steroids. How is Uber, and their sort of technological advances going to challenge that, and I'll get to the point that there is a bad scenario where Uber control every car autonomously and intentionally slow down some links to provide a premium service to other links where they're flying faster. So, how will you and the industry go on to ensure the government do continue to play in the societal control space and not just leave it to market?

Travis Waller:                     I think your question is fascinating and exactly the sort question we have to be thinking through. And I would argue even... and again, trying to just be relatively agnostic politically, and I honestly I'm, I care about the numbers and little else. I feel sorry for my family and friends. Even for a market driven person, the specific example you highlighted was the emergent of a natural monopoly, which even in a market principled approach can lead to systematic issues that should be avoided, and a prime role of government is avoiding the existence of monopolies because they could then exhibit exactly the behaviour you're talking about. If they did not... if there were other players that would limit their ability and profit for trying to do such a thing because some other player would be able to exploit that to undercut them. And so even if the... So I guess the short answer to maybe a long potential philosophical discussion is government should be working to ensure that Uber does not evolve into a natural monopoly.

Mark Hoffman:                 Well, if we don't have any further questions, I think we should give a very big thank you to Travis.

Travis Waller:                     Thank you very much.

Mark Hoffman:                 I just want to give a heads up for our next Learn@Lunch, which will be addressing a growing wildfire threat, which will be associate professor Jason Sharples. It will be held on the 13th of November. So, please make sure that's in your calendars and thank you very much for coming today and enjoy the rest of your week. Thank you.

Cover image