all right let's get started uh hello everyone I'm Jin professor of city and transportation at MIT welcome to the MIT Mobility Forum designed by the MIT Mobility initiative today I'm really excited to have Professor Fred maning come join us talk about highway safety particular the Contemporary issues and the forthcoming age of auton vehicles right highway safety problem have really persisted uh I would say even getting worse in recent recent years right traffic fatality in the United States actua
lly increased 10% from 2020 to 2021 here maybe I just show a quick chart to get some context here so this is showing the total death the the gr line over the century here uh it picked around 1960s going down until 19 229 2009 10 Etc but then it goes up in the last few years so we are actually not going to persist we're actually going to the wrong Direct I would say and this actually happened at the same time where we had all sort of Innovations technology policies instrument try to reduce safety
here right so these persistant as Professor man will argue that is really because the interaction between the safety intervention and the human behavior the human responses to this Improvement right highway safety remained a perplexing problem that both engineers and also social scientists try to resolve but yet uh to succeed in this I see there's no better person than Professor Fred manin who can really shed light on this puzzle as he wrote in the abstract I shared with everyone unfortunately
currently the the safety related application of say artificial intelligence machine learning are making the same mistakes this are likely to have a disappointing safety consequences for the forth coming years where our trans vehicle will be interacting with many of the human driving human driven vehicles for I think a decades to come but if these data analytical pitfalls of the past are recognized and avoided then there is indeed a huge potential for AI and ml to vastly improve the safety partic
ularly in this forthcoming interactive uh uh regime of Auto autonomous and human driven Vehicles right before I formally introduce Professor manding I do want to warm up the audience here by invi everybody to type your organization location and time so that uh man know who the audience are so I'll see MIT uh Cambridge uh to also I try to reinforce the norm of this forum which is I hope everybody to contribute an idea right either as a question or a comment to Professor manen's presentation uh Fr
ed uh Manor is Al long MIT graduate from PhD in 1983 uh he's a professor in Department of Civil and environmental engineering at the University of South Florida where he runs the Center for Urban Transportation research before then he was the head of the Department of Civil Engineering in Purdue University also previously in washing University of Washington uh his b research had been cited over 37,000 time in Google Scholar uh and he founded and current edit of chief of the journal andal me acci
dent research which really ranked the the top one among Transportation journals and before that he also the editorinchief of Transportation research Part B methodological uh he received a numerous award and recognitions including the web of science highly cited researchers for many many years the counil of University Transportation Center Lifetime Achievement Award in 2021 and most cited also the 50 years history of the Journal of accident analysis and prevention right and last I want to mention
the annual Foundation top 10 transation leaders Academia right without further Ado let me pass the Flor on to Fred please great so let me share my screen get started so today I'm going to talk about is uh perspectives of Highway Safety and as was noted in the introduction we have seen an increase in the number of highway deaths and pedestrian desk and I'm going to give some insight into what might be causing those increases so let's look about you know what has been happening in highway safety
we've had incredible advances in vehicle safety Technologies we multiple airbags autonomous braking Lane departure warning I mean just countless safety Innovations and we've also had drunk driving policies that have tried to affect some of the behavior avior we've had texting laws red light cameras so many many Innovations so with all of this we would expect that Highway fatalities and pedestrian fatalities on highways and bicyclist P vehicle interfaces we would expect all these to be declining
but they haven't declined and this give let me this gives a little explanation of of what's been happening that this is a shorter time period than was presented initially but if we look at from 2011 to uh 2021 we can see it's like pretty flat even though these incredible vehicle Technologies are are being implemented and then of course we do have the spike uh in 2020 to 2021 and you I should the fatalities in the blue include pedestrian and vehicle fatalities any Highway related Fatality and the
n the dotted line is fatality per 100 million vehicle miles travel so it's not just oh there's more cars we expect more fatalities you know the rate is actually increasing so let's look at what are some of the disturbing highway safety Trends one is that bicyclist fatalities have gone up 55% from their 2010 low motorcycle fatalities again these are like vulnerable Road users have reached all-time highs including a 21% increase from 2019 to 20121 and pedestrian fatalities have increased 80% from
their 2009 low so what I'm going to argue is like what's going on here I would I'm going to point to four I mean there's many things going on obviously it's a very complex engineering human behavior interaction but I'm going to point it look at four things and argue that we've not accounted for the effects of safety Innovation on human behavior we put in a technology but we don't not think about how people are going to respond to that technology behaviorally we've not considered the externalitie
s of the vehicle technologies that we've implemented the vehicle safety Technologies we've not accounted for the evolution of human behavior and I'm going to argue that human behavior is changing in fundamental ways not just because of the safety Innovations although that's one reason but there's just a natural evolution of uh attitudes and behavioral preferences and then we've relied on flawed data analytics I mean we've we have many possible data analytic methods but there are fundamental elem
ents related to data analysis that we're not fully have not fully considered so let's look at point one the effects of safety Innovation on driver Behavior so let's say we have the introduction of safety in I will argue that the introduction of safety Innovation changes behavior in fundamental and perhaps unexpected ways if you don't think about it let's consider an example let's consider an example of a technological innovation that improves safety what will happen so I have this little diagram
from uh one of the papers that I published on this subject that looks at the possibility of compensating Behavior so we have some initial conditions um you know on the horizontal axis we have a selected normal driving speed but it doesn't necessarily have to be just speed you could drive distracted and things of that nature and on the vertical axis I have the probability of an injury accident now each person has some equilibrium point point a on the initial conditions curve here where they have
a car or whatever and they have some they like to have some probability of avoiding an injury accident so they might have like oh when I drive I like to have a one in 10,000 chance of being in an injury accident everyone will have a different point a depending on the risk profiles and so on but let's put in the safety Innovation which moves the curve down from the initial conditions curve to the safety Innovations curve so the curve slips down and we expect as Engineers that people will keep dr
iving the same speed and they'll move to point B so the probability of an injury accident will decrease accordingly and we'll move from point A to B but really what happens is some people might say well you've made this car safer I still like to have a one in 10,000 chance of having a crash so I'm going to move from point A to C that's complete compensating Behavior but realistically some people are not going to be driving at the same speeds they'll probably Drive between somewhere between B and
C so for example at Point D in the diagram here where there's some compensating Behavior but there's still some net effect now if there's complete compensating behavior that doesn't mean the safety Innovation doesn't have a benefit to society because driving faster or driving more distracted still has a utility benefit to the driver so there's still benefit to society it's just not a safety benefit and there's also the possibility that people overestimate the effectiveness of the safety feature
and end up at some point e where they drive faster and have a higher probability but that's less likely so if we ignore compensating behavior and we have done this a lot in the profession this leads to an overestimation of the safety Innovation Effectiveness and unexpected consequences which should have been expected so I would argue a lot of the reasons why we're not seeing the great benefits of the safety features in cars is compensating behavior is one of the many factors that is mitigating
the potential safety benefits of these vehicles then there's externalities that sort of relate to this so if there's compensating Behavior one way to think about this if there's compensating Behavior going on people are driving faster or more distracted because they feel their vehicles are safety or are saf that's going to be an externality they're making their utility maximizing decisions but if they're driving faster uh vulnerable Road users pedestrians bicyclists motorcyclists don't have that
injury pre those injury dep uh prevention devices so there's a much higher probability of them being involved in crashes so as an example of externalities of compensating Behavior safety Innovations often result in faster and more distracted d drivers and the consequences as I mentioned vulnerable Road users and drivers without the advanced safety features you know the multiple airbags and so on are be exposed to increased risk of injuries and that's a and if we look at the vulnerable Road user
increases uh in terms of fatalities and injuries you know pedestrians motorcyclist bicyclist much of this can be explained by this compensating Behavior next is the evolution of human behavior you know how people's uh values and behaviors change over time now there's you know there's a lot of literature that explains that human behavior is continually continuously involving you know economics cognitive science everybody knows that people's uh behavior is evolving but these behavioral shifts can
be accelerated due to technological innovations changes in Risk perception changes in societal norms and so on so this is you know the fact that our if you think about it if you're estimating a statistical model not only do your X's change you know your explanatory variables change over time but your betas change over time is how the effect of the explanatory variable on the decisions that you're considering and there's Apple ample evidence in the safety literature that shows that model statist
ical model parameters change significantly over time and if we ignore these temporal shifts this is going to lead to erroneous INF inferences and predictions and also a misinterpretation of how effective the safety Innovations are going to be the next one I'm going to talk about is flawed data analytics so what have you know we've been using statistical and econometric methods more recently we have data driven methods machine learning and so on but there's fundamental elements that we've not ful
ly considered and you know if if you're an econometrics person these would be elements related to selectivity endogenity and identification and I'll give some examples in a second but the above issues are systematic in are endemic to highway safety data and the consequences of ignoring them are not giving them some consideration is we'll have biased safety predictions and we'll be unable to identify underlying causality so let's give an identification example suppose you're looking at motorcycle
safety and you estimate a statistical model and you find that the if it's raining you have a higher probability of injury severity so there's two possible explanations to this result one is that rain does increase the in injury severity you know the nutonian physics involved reduce coefficient of frictions and so on but the second one is that only the riskiest Riders ride in the rain and they'll have more severe crashes whether it's raining or not so the effect of the rain variable could be pur
e selectivity and the actual rain itself as minimal effect we've been as analysts in safety data we have been unable to untangle these elements we are not sure how to you know what is the actual effect of Rin and what is the actual effect of the selectivity that might be caused by rain and the consequence of that is you could come up with fundamentally different policies so for example if you thought it was a nutonian physics explanation it's like well we should have motorcycle tires with higher
uh friction but if you think it's the selectivity uh element we should you develop a advertising Campa campaign that discourages motorcyclists from riding in uh adverse conditions the next one and this is an example that I actually wrote a paper on is the effectiveness of vehicle safety features and I wrote a paper on this in the early 2000s that looked at the effect of side impact airbags so if you look at the de development implementation of side impact airbags and vehicles they're found to r
educe fatalities relative to Vehicles without them but side impact airbags seem to become less effective over time so what's going on here one explanation is there's a self- selectivity the people that buy the cars with uh most safety features are a self- selected group of the safest drivers this is why any safety Innovation that we develop with cars if you look at the vehicles with the safety features and vehicles without the safety features initially you're going to find that they're much safe
r than they appear because the safest drivers buy these cars there's an and self- selectivity going on and if we look at the side impact airbag Effectiveness the Insurance Institute for Highway Safety reported them 45% effective in reducing fatalities in side impact crashes then in 2006 as 37 2008 31 201025 what's going on here is most likely the initial 2004 the safest drivers are owning these cars they're going to be in situations that are less safety critical by 2010 you're starting to get th
e real effect and eventually when you do like 2020 which almost all of the new cars have in side impact airbags you get a much better uh interpretation of the actual effectiveness of side impact airbags so the thing to be cautious here when you hear a new safety feature that oh this is 90% effective we should do this think about the fundamental self selectivity they probably uh when you get those numbers the most common thing is you look at the vehicles with the safety features and without and t
hat's exactly what you don't want to do because of the self- selectivity you're always going to overestimate the effectiveness so the future of Highway Safety in the autonomous human uh driven vehicle environment now we can think about we're putting autonomous vehicles into the vehicle Fleet now and this is going to create all sorts of things that we have to think about in advance so what we really need to consider are the thing you know the elements I've talked about selectivity temporally shif
ting Behavior compensating Behavior and the resulting externalities so if we look at Future possibilities how one way as an example to minimize the compensating behavior is to develop something like a you might call an AI assisted driver driving and what this does is it adjust Vehicle Safety Fe Fe systems after learning driver responses and to changing conditions and you can imagine that you could develop a an AI system in a car that adjusts the safety features to mitigate risk compensating Beha
vior one way to think about this is you would make the car seem more dangerous than it actually is so for example if you have a really good braking system you probably apply your brakes much later and you know have roughly the same probability of getting into an accident but you could have develop an intelligent system for in within the car so that it detects when you're applying your brakes and then thus mitigating potential risk compensating behavior when we think of autonomous vehicles and th
is is something that's really important as we start to introduce autonomous vehicles into the traffic stream the current approach is design autonomous vehicles to avoid collisions in response to human driven vehicle trajectories but I would argue that one of the most important things uh that we have to think about in autonomous vehicles is how they affect the behavior of human driven vehicles around them and I would argue the second the alternative approach is we want to Des design autonomous ve
hicles to interact with human vehicles to alter crash critical vehicle trajectories before they happen so for example you could develop an autonomous vehicle that detects vehicle you know close following distance behind it you know flashes brake lights does all sorts of interaction maybe Maneuvers uh accelerates decelerates to affect the behavior of the human drivers around the aut autonomous vehicle and the likely consequence if we ignore the human response to autonomous vehicles what we're goi
ng to find and this is what we're doing now right we're just throwing the autonomous vehicles in and see what happens what's going to happen is human driving behavior will change around the predictable driving of autonomous vehicles and there's already antidotal stories where uh you an autonomous vehicle behaves in a very predictable manner so uh drivers tailgate autonomous vehicles cut in front autonomous vehicles but unfortunately those behaviors they develop around these predictable uh behavi
or of an autonomous vehicle will propagate to the behavior around other human vehicles so changes in Behavior around autonomous vehicles will create unsafe Behavior around human driven Vehicles increasing the human- to human driven vehicle crash rates now this is most certain you know this is my great prediction here this is certainly going to happen unless we develop autonomous vehicles that can mitigate this because we talking at the beginning where we you know one of the arguments is that dri
ver behavior during covid uh is now propagating in to uh the postco years so the bad habits we may have developed during Co when traffic volumes were low police enforcement was low you know there's probably some propagation of that this sort of propagation will also happen with driving around autonomous vehicles and then human vehicles so the future of Highway Safety the fundamentals of human behavior are still not being fully addressed when we develop a new technology we're just expecting that
it'll be effective based on primarily in the engineering aspects of it we're not considering the possibility of exam for example of compensating behaviors that people may start driving faster more distracted more recklessly to mitigate per perhaps totally the effectiveness of these features now there are some safety features such as uh shoulder belts that you can argue argue that have not we've not been able to compensate to overcome their true Effectiveness but the safety features that we've de
veloped since those initial highly effective safety features of uh occupant restraint are not quite as obvious and not quite as powerful as that so the compensating element can easily mitigate what the potential increase in safety is now there's a lot of AIML that opens up new possibilities in data analytics but unfortunately the AIML perhaps even worse than the statistics and econometrics the people that are applying AIML to safety data are not considering the fundamentals of human behavior con
cepts of selectivity endogeneity uh changing behavior all of those things are not being fully considered so the current approach as I uh mentioned previously is we introduce a technology and we simply hope that the technology advances overwhelm the mitigating human behavior and I should mention as I conclude here and we get into the exciting part with questions and interaction um I've written uh several papers with my colleagues on Concepts presented uh in this uh in this talk so feel free to lo
ok some of those up and get very detail statistical and econometric Analysis of what's going on with these data and I think that's I I'm not sure how we're close to 20 minutes so great thank you so much BR you open so much discussion points here right so we're jump into directly one is you throughout the end you mentioned this the aggressiveness of human Visa V AV May propagate into a human to human interaction context here right so actually have an example I last year went to cartin with my son
and right after when I drive home I indeed find I actually drive much more aggressively I couldn't help control myself actually that's first time I had first ever accident I ever had on this right but then back to the a discussion so does that mean that we shouldn't design a to be too conservative so meaning building some degree of aggressiveness like a s on par with a human may be better for the system perspective right right and I you know first of all to get to your uh compensating example I
should tell you think it's already done it's already done and gone what go ahead okay so I I should tell you that um I you know I own like eight cars and they go from the 1960s you know because I collect some 1960 1970 all the way up to 2020 now I should tell you that my newer cars have the full level one you know with the Collision Andoid and say autonomous braking adaptive cruise control and to your point when you notice how you're driving behavior changes since you know when I start driving
a one of my level one Cars my behavior changes quite a bit you know because now I have I can I know the car is safer I start messing with my apple carplay and all of this distracted stuff when I get into one of my you know 1960 1970 cars you know your life is in danger and it becomes much more obvious they don't break as well so but we do have to be careful because this is also a case where we have self selectivity if you look at the effectiveness of level one safety Innovations you find they're
pretty excess uh successful but you only observe the safest drivers driving these cars I mean if you actually you know as my little anecdotal exact if you actually have these cars you have an L1 car and a way non- L1 car you know my 1960s car was like the first year with the shoulder belts and you know no safety features whatsoever it's like it was a survival of the fittest right but um so that you know it's clear that that compensating Behavior has gone on but to your point it is it's not the
way we design autonomous vehicles we don't want them to design them so they have a higher risk of accidents but there are ways we can keep it same very low accident risk but still communicate to the surrounding human drivers that you know this car can still behave erratically while not actually increasing the probability of a crash so for an example if you are being if an autonomous vehicle is being tailgated you could algorithm just starts flashing the brake lights sort of an indication that an
d you don't even actually have to apply the brakes right you can just communicate with the drivers around it so that that I think is going to be a critical critical safety uh element of that relate to the statement you made that make the situation seemingly more dangerous right right right exactly that's both in terms of your own vehicle but also signant to the other vehicles about yes yeah so here then you mention four reasons that we may make M mistake the first one is a compensating Behavior
right for that assume that the degree of compensation will vary uh either by different technology or by different group of people can we give some example where certain technology some somehow people compensate a lot while certain technology people don't compensate much what drives these different degrees Yeah I think the you know the uh the areas where they compensate the most are the ones that you would consider active safety features so for example uh a passive safety feature would be like an
airbag yeah you know it's there and you might do some compensation but it an active safety feature that might be you autonomous vehicle uh braking or Lane departure warning that is something that I think you have a lot of compensating behavior and I you know to give an example um some of my and again anecdotal some of my level one cars have heads up display so you just start focusing on that heads up display that has the little radar indicators that tell you if there's cars beside you and stuff
and I realize you just don't look in your side view mirrors that much anymore because everything is in the heads up display but when you go back and use your non-level one non headsup display vehicle it's very dangerous you know there's an adaptation period so so I think the active safety features the ones that are providing feedback you know an exam another example would be if you have a vehicle that has really good braking performance almost invariably you'll start applying your brakes later
and sort of mitigating some of the effectiveness of the braking system so the active safety features almost certainly have compensating Behavior whereas the passive ones may have less of less so I see so then to your second point about the externality right if at the individual level You could argue that people just have a different degree of tolerance so it's my decision to choose where do I balance between the two but the the thing is if you choose a high risk balance you also impose the inati
ve for other user who didn't choose to do so right that's problematic so for that like see in congestion part when we has congestion anality we have the device like congestion pricing try to address that right even though politically difficult but we we somehow successfully Implement some of this in terms of safety externality what are the meaningful policy interventions that we can internalize this externality now that's an interesting point and I'd have to think about that I've been thinking u
h that the you know the most obvious solution uh to that sort of compensating Behavior would be to develop a safety system in the car that mitigated the risk compensating behavior I'm not sure if there's actually a policy that you could develop that sort of uh and again instead of looking there's a whole field of Transportation safety that instead of looking at observed crashes that looks at uh vehicle trajectories and near misses right so so that might be you know I'll give you an example of a
study that I did in um is a sort of a real time adjustment uh we did video analytics of an inter section where we look at NE near misses and if we start getting a high number of near misses then we change the S the signal timing gets changed um to avoid these you so you don't have to wait for a crash to occur these things can be adapted in real time but I I think in this case you know perhaps on like congestion it's hard I mean we've we've made over the years many policy safety related policies
but I don't see an obious policy uh to mitigate this compensating Behavior because it it goes on and a lot of people don't even realize what they're doing um so there's a there's an awareness it's not like uh you know for drunk driving yeah you have to quit drinking but for how do you uh your risk mitigating I guess you could develop a a system in the car that is tell you know determines when you're applying your brakes and gives you a warning late break late break application or something but i
t looks like instead of a sort of a it has to be a policy that is centered around the specific technology right well thank you yeah so the third one about the the temporarily changing nature of human behavior here right particular you mentioned that uh some of the behavior Chang due to the safety technology itself but some of the human behavior Evolution are totally exogenous is outside reasons that right can you give us some example of that also what are the cause for such Evolution so you know
as an example you know the one mention we say well that behavior changes in response to technology but there's a very large body of literature that has determined that behavior changes because of macroeconomic conditions and you may notice um during the Great Recession uh of you know in the late 2000s there was a significant reduction in the number of crashes and historically what we have found if you look at the aggregate data during recessions there are fewer fatalities per mild driven and du
ring good Economic Times uh there are many fatalities per mild driven so if you're looking about how's our our current economy is doing great because we have a lot of fatalities right you know maybe that's so but really the people that have looked theoretically at what's going going on here is there's a a lot of history that macroeconomic conditions change in fundamental ways VI people's risk profiles and again there's a whole body of research that sort of look at how this affect another element
that inherently makes human behavior unstable is there's a difference between attitudes and behavior so you know we have an attitude that oh I would love to save the plan planet and so but then we have the behavior is I drive an SUV that gets 15 miles to the gallon so how do you reconcile this to and I think as as we go through life there's always a difference between our attitudes and our behavior and there's a cognitive dissonance there sort of the Leon feser thing for the early 60s and we're
constantly reconciling that dissonance and that constant uh reconcil ation of the dissonance between your attitudes and your behavior inherently make your behavior sort of unstable so you're adjusting your attitudes and then you're adjusting your behavior so there's a lot going on and if you look at the fields of Economics I mean there's a work of coupan in the 1960s there have been many people have looked at fundamental ways why people's tastes are CH tastes and behavior are changing over time
but it's a you know there's many causes of that thank you yeah uh so probably the last question from me so I encourage the audience again put your comments so question to it then I'll give it to John to moderate that uh so last question is more on the analytical part right F you contribute to actually significant to both the kind of a Classics statistics eometric based methodology as well as the the kind of recent machine learning based methodology give us sense of what how do they compare part
icularly in the sense of falling into the victim of the four problem just us illustrate sure sure so when we think of I think a lot of people that use the datadriven methods you know the machine learning their primary focus is on prediction okay but in addition to prediction there's causality because a lot of times just being able to predict something is usually short term and doesn't give you any insight into what is going on so when you think of traditional statistics and econometrics there's
certainly a prediction element but there's also a search for the underlying causality so and that because it's the causality that will determine what a an effective policy might be and I think that's one of the uh you know major problems with the application of machine learning in safety data is that we're just focus on prediction I mean you know for example the in the journal that I edit I get many papers that say oh look this model predicts better than the econometrics model you know of course
it predicts better but we're we want to know not only prediction we need to know causality because it's the causality that is going to determine the policy that actually affects um you know the that actually affects the in this case the probability of fatality or whatever looking at in the safety area so that's the fundamental difference between the two and both have Advantage I mean there's an advant if you focus totally on causality your prediction may not be so good right if you to totally o
n prediction you don't know what's going on so there's a balance between those two that we have not yet struck in the field right you know a recent talk I I just watched you you categorize this application three bucket right the first bucket is this this area that AI will be just beautifully applied and nothing wrong just go with it the second category is AI with at least moderate level of caution on this and the third he said is AI with Extreme Caution that you should apply on this give some se
nse about the three categories right so you I mean there are some applications that you know have to be you know realtime applications you don't have time to think about the causality you're collecting data and instantly providing a prediction um you know in that case uh you're not so much interested in uh the causality because it's like an instantaneous react so that's an obvious application of machine learning right because this can do the prediction instantaneously but when you get to a longe
r term and think about how it might affect in the future then you really have to think about the causal effects great thank you so now let me keep the form to jel for the audience okay super thanks so much Fred um just before we we kick off with the questions from the audience I wonder if you know one of the main points that you made was about the introduction of uh safety technology actually sort of changing people's behavior and causing them to embrace riskier behavior and I you know just for
for those of us who don't read you know econometrics today at night if you were explaining this to a Smart 14-year-old how would you explain your your evidence for that I think it was great that you articulated the difference between active safety features on vehicles and passive safety but I think it's worth just returning to that uh that if you could explain that a little bit more what's your evidence yeah you know I would instead of getting into all of the econometrics that actually uncovers
that evidence you know if I was explaining it to the way I would explain it to one of my children or something although now both of my children have PhD so they can understand the comment but if I was explaining it to them when they were younger you know I think using the anecdotal explanation is perhaps the best is you know when you know for example in my case when I'm driving my level one autonomous vehicle car uh you can notice that I'm noticeably less uh attentive to the roadway than I am dr
iving one of my you know 1970s cars but and there's a lot of you can some of it you know we've developed these safety features but we also have things like apple carplay you know so we've developed we have all these safety features but we have all these additional distractions um I remember one of my 1960s car uh this was an mg if any of you you know remember the English car makers but they were so concerned about distraction that the owner would not provide a radio as standard equipment they th
ought the radio was too distractive to driving but if we look at cars today just think of what we have beyond the radio I mean you have all the you know some of it is voice activated but I can tell you my Apple carplay to try to get play Amazon music you know I'm trying to have it play one of my band CDs or something it's you know I'm continuously distracted so in some sense we've uh with all the advanced safety features we've had uh We've also added a lot more distractions within the vehicle an
d that and we have the people have a greater opportunity to risk compensate from the distraction perspective because now uh and that and that is an important thing that we really haven't considered because we developed the you developed the safety in One path and you develop the amenities the Apple carplay in another path we not thinking about how those two interact yeah and that in fact I'm glad you raised distraction because that's exactly what I was driving at was if it's even possible to sep
arate these two effects the effect of increasing safety technology features which as you argue is driving in fact riskier behavior on the one hand and then you know distraction which of course has been a major focus of you know Road Safety agencies around the world for for some time now just how to disaggregate those too and you know I should mention that even though my diagram had speed on the horizontal axis it's really speed slash distraction right because I find when I drive uh My Level One
cars they're not as fast as my non-level one cars so I don't drive them faster but I do drive them a lot more distracted okay I want to come some of the comments in the chat really interesting and I think one of the if not one of the first questions we had was from Burn grush and he basically said have we accounted for increasing vehicle size and weight now couple this with a question from Ben perir and why are collisions decreasing in Europe so there's this is a related set of questions I mean
we um one of our uh senior uh researchers one of our senior fellows at MIT David zipper has been writing a lot about car bloat and you know Vehicles just getting larger and larger and with higher surfaces you know Greater Heights at the at the front end of the vehicles resulting in greater risk and indeed fatalities to pedestrians first what are your thoughts on car bloat and then if that is a specific to the US conversation please um elaborate on that yeah I think the the car bloat is a a very
serious issue and you know one of the to give an example with electric vehicles tend to be much heavier um now I should give you a little antidotal story is that just earlier this week I bought a 1976 Triumph TR6 which is like 2500 lb little British sports car right but the car that was on the trailer that was backed out before it was I some BMW i6 or something but I looked at the the weight of that car is like over 6,000 pounds and the weight of my little Roadster is like 2 200 PBS right so a l
ot of this I think with the uh car blo we are seeing perhaps a greater variance in the weight distribution on cars on the road perhaps more than any time uh in the history of the vehicle mix because now we have cars from like 2500 pounds up to 6,000 pounds and over these are just regular cars you know not including heavy trucks and so on so that rate weight distribu uh I think imposes an externality people with heavy cars are imposing an externality on vehicles with lighter cars now it's not jus
t weight because you have crumple zones and everything uh you can have a car that is lighter and still be pretty safe if you'd have very Innovative crumple zones and so on but the car weight is again I think one of the re reasons and as you mentioned the it seemed like in the 1990s we were designing cars to be uh highly efficient from an air perspective right coefficient of drag was very good and the consequence of having good coefficient of drag is you have a nice slope on the hood it probably
was safe for pedestrians as well but since then we' become less uh concerned about coefficient of drag we have more efficient engin and so on and as you pointed out uh a lot of the larger vehicles have these vertical grills and you know when you think of the injury and even uh small low uh kilometer per hour collisions it can be pretty significant now with regard to Europe it's interesting that and almost all of the developed countries as you know except for the United States um have have notice
d significant improvements over time in fatalities per kilometer driven or whatever and I think I'm not sure if it's a uniquely uh uh America American phenomenon what makes maybe uh people in the US more likely to risk compensate I that would be I again I haven't followed the literature uh comparing you know the behavior of us drivers versus European drivers but you know there's all there's also an adaptation that goes on I'm G give you an example um when you drive in Boston for you know how peo
ple drive in Boston is much different than how they might drive in Tampa for example and you this is and so you can see this dis arities spatial disparities within the us and this got me to the point let's say you had an autonomous vehicle that is actually learning from the vehicles around it which you know you have to be really cautious you know having it learn driving behaviors from be vehicles around it but if you have an AI autonomous vehicle that was trained in Boston and then you you know
started driving it in Indianapolis or something God knows you know probably the first 100 miles you'd have like three wrecks or something so there there's and I think when we think of the US compared to other developed countries there's something there's fundamental change there's fundamental differences I think in behavior and that tends to be reflected in how effective the safety features are from one country to the next yeah yeah exactly I mean it is interesting that your risk profile driving
on a road in the United States is roughly twice that of driving on a road in Canada when you know there's not a huge amount of difference in the sort of uh characteristics of the the two countries from A yeah I think you there's some there must there's some cultural uh you know fundamental cultural element that's going on um but yes there yeah the and this is you I think safety researchers in the US have been perplexed by this because when they look at the the advances in safety that have been
made in the European Union uh and then compare that to what's going on in the US it's a perplexing problem yeah actually we have uh Peter Norton had posed a question that I want to come to and maybe Peter if you're still on with us you can jump in directly but before we get to that I want to uh pose a question that that um Jane Chu offered in the chat and she basically said um you know these do these laws mean physical interv interventions for example traffic calming support Public Safety more s
uccessfully than safety features in individual Vehicles how do we achieve safe designs in the US Transportation Network given not just flaws in individual thinking but what Roger Ruck has called an Institutional flaw with how we educate our transportation Engineers I thought this was a very good question and it was in the back of my mind as well and I'm sure many our of our viewers about you know the question of how we actually design our road Network how how do you think about that impact on sa
fety yes I think uh BEC You know we can think about how the spectrum is going you know at some point maybe you know 60 years from now we'll have all autonomous vehicles and then the problem will resolve itself then it just becomes a software problem um but until then I you know you're probably correct and you know I'll have to make sure I do another addition of my textbook principles of Highway engineering and traffic analysis but we when you think of what could be the most effective safety poli
cy it's probably you know not looking at controlling the individual behavior in uh specific Vehicles although artificial intelligent assisted driving once we get a safety features enough we can start looking at those risk compensating elements but as pointed out in the question I do think probably before we get to that level things like traffic calming and stuff are probably the best ways to address safety in the very short term now it's interesting that there is a demand for safety features in
vehicles right because people certain people want to be safe and they tend to pay significant amounts of money for additional safety features and the market is responding to that right so they're by level one automation so there will be people that are seeking those safety features but I think if you want to make a more to the point raise if you want to make an immediate impact things about designing safer roads is probably def more effective than you know adding more safety features to cars on
unless you add the safety features and then develop the AI to adjust the for the compensating Behavior then you might be on the same uh playing field but yes without I think we probably overestimate always overestimate the effectiveness of Technology right so we're saying oh yeah we'll put this safety feature in it decreases crashes and as I mentioned before these safety features are often Justified oh it is a third % decrease in the probability of a crash that number comes from the safest drive
rs because those are the ones in those cars right it's really not 30% and so we have to think about that self- selectivity uh when we get into looking at individual safety features but yes I think designing safer roads um is probably the best way to go but there's a cost to that too right roads aren't cheap yeah well absolutely and but I think the core of Jane's question is around this concept that you know we have trained Transportation Engineers for Generations at at MIT and other institutions
to focus on this concept of level of service and level of service is a concept that just says you want to move more cars more quickly through the road Network um and so it's interesting that that's the objective rather than you know protecting people's lives and I think that's that's a little bit of where at least in my mind it creates this cognitive dissidence and we need to sort of maybe reink what the larger objectives are do do you have thoughts on that yeah know when we design roads it's n
ot all for a level of service uh a lot of it is the actual sight distance and there is a lot of accident prevention in there uh stopping sight distance so for example uh again this is you know chapters in my text on this chapter three is we designed roads for a an anticipated I height stop expected stopping distance of the vehicle you know so there's it's not just number of lanes right there there is a significant safety element being considered in design of highway but we design a road with so
for example interstates are designed for 70 mph driving speeds but that's based on a stand standard vehicle with specified braking characteristics if you have a high performance car you could probably drive that road with the same level of risk maybe 80 or 90 miles an hour right but for the standard designed vehicle we design them for a certain speed to um you to allow the vehicle to stop in a safe distance so there we do consider that but as you can imagine as was pointed out earlier there's an
incredible variance in vehicles on the road right you have these heavy trucks it's interesting heavy trucks that have terrible braking performance uh are not really the design vehicle because the driver's eye height is so high uh you know when you do vertical and horizontal alignment of design uh that's you know that's much more effective than minimizing the braking distance so we do consider elements of Road design but the high variance makes that a challenge the high variance in the vehicles
that drive on the roads makes it a serious challenge so there's only so much we can do in designing the neonian physics of avoiding crashes right and when we look at it though most crashes are caused by human error right and that's the great promise of autonomous vehicles we get the humans out of the decisionmaking process then it becomes much easier you know if I give this talk 50 years from now it's like well we did it we have the machine learning and autonomous vehicles but to the point is a
lot of people try to correlate um traffic congestion with safety and you know some some people say well if you have high congestion it's inherently safer because you can't get to the speeds that cause fatalities right uh so there's a there's an interesting interaction between level of service and safety um but a lot of it you know there's a lot of elements involve like the speed variants and the vehicles there's a lot of things that have to be considered there excellent we could go on there's so
many great comments that we will share with you uh in the chat but we're out of time so I'm going to turn it back to J great than thank you thank you so much Fred this is really a br and discussion here uh most people we we we see you as the Brea researcher uh but I I do want to let people know that you also have two other very important uh Endeavor the first I'm sharing is f is actually lead guitarist of a band uh vulgaris and also as a a motor racer here so I will say you're probably the most
colorful Transportation professor in the world with all these different variety of interest and hobbies so everybody please join me thank Professor M for the presentation today yeah thank you thanks
Comments