How should humans really be ‘in the loop’ in cities under autonomous mobility?
What questions lie amidst the wreckage of Joshua Brown’s Tesla Model S?
“A car crash harnesses elements of eroticism, aggression, desire, speed, drama, kinesthetic factors, the stylizing of motion, consumer goods, status — all these in one event.”
—JG Ballard, interviewed in Penthouse, September 1970
On May 7th 2016, 40 year-old Joshua Brown died in a car crash near Williston, Florida. So many people die in car crashes that this in itself is hardly remarkable, something that does not reflect terribly well on us as a species. Yet this one was remarkable as Brown was driving a Tesla Model S in ‘Autopilot mode’ at the time.
And so in fact the Tesla was driving Brown, as it collided with a tractor trailer turning left. The large white side of the trailer against a brightly lit sky apparently rendered the truck ‘invisible’ to the Tesla’s systems, and so the brake was not applied. (It later transpires that Brown was speeding; other rumours suggest he was also watching Harry Potter at the time.)
This is hardly the first time a human has been killed by a robot, or some other less obvious form of autonomous machinery. Hundreds, thousands, must have done. But it is perhaps the most high profile death, partly due to the spotlight shone so intensely on Tesla by Elon Musk—its shy, retiring, almost reclusive founder—but also due to the massive systemic change suggested by autonomous vehicles (AVs). As Ballard’s quote above suggests, such events can pull focus onto the pivotal issues. So if, paraphrasing Cedric Price, autonomous vehicles are the answer then what are the questions hidden amidst the wreckage of Joshua Brown’s Model S?
At core, there’s that key 21st century puzzle: what jobs are best done by people, and what are best done by code? And indeed, how should they be done?
Much systems design orthodoxy recommends keeping “the human in the loop”, just as a contemporary airliners have pilots despite them being largely flown by autonomous systems. It seems reassuring to us that a human can take control. But we know in reality this is increasingly a fallacy. Humans perhaps too easily place great faith in systems, to the extent that it’s difficult to grab the wheel, as it were, when control may be suddenly handed back: this is the so-called “handoff problem”. Further, it could be that peoples’ skill and awareness atrophy as more agency is lent to code; several airliner crashes can apparently be attributed to pilots failing to recover the aircraft after a fly-by-wire system collapse. This is called the “automation paradox”.
More simply, we may be keeping the human in the loop in order to have someone to blame when something goes wrong — in this case, the speeding Joshua Brown — as the blame is simply distributed too diffusely otherwise. This is human as “liability sponge”, in the words of Madeleine Clare Elish and Tim Hwang.
So should driving be a joint effort, between human and code — just as the best chess players are a combination of human and computer, sometimes known as ‘centaurs’? Could this be what I’ve called a ‘shepherd, sheepdog’ relationship? The human ‘shepherd’ guides a suite of autonomous ‘sheepdogs’ — different species but clear communication, common goal (well, more or less) but with people in control.
Or is driving substantially different, with code increasingly so much better at driving that people can only make it worse? This seems to be the view of Musk, Google et al, whereas the traditional car manufacturers, perhaps not surprisingly, see things a little differently.
It could be the best approach is a much clearer distinction in terms of agency and autonomy. I’ve long argued for greater human agency in many interactions which we are unthinkingly outsourcing to code: the active, smart citizens model I discussed four years ago, using Monderman’s shared space as analogy, somewhat ironically, it turns out. Whereas many other interactions and transactions could be best fulfilled by code – in this case, much mobility and logistics. Where both interact, the shepherd/sheepdog model can be observed, with clear, holistic thinking about who, or what, does what. (Addressing that essay’s brief allusion to the embedded human rights issues, too.)
In this sense, Monderman’s shared space becomes an even more appropriate model, meaning we could design out traffic lights, pedestrian crossings, bollards, barriers, even bike lanes, potentially, as we uncover a more fluid, human and engaged approach to streets, with the active modes of walking and cycling atop a mobility hierarchy propped on various grid/non-grid forms of autonomous public transport (Ed. More on that later.)
This all requires a firm hand on the urban policy tiller, for it may be that driving itself is the least of the issues. A broad shift to AVs may mean the disappearance of multiple categories of jobs — taxi drivers, truck drivers, deliverers and couriers, motel and service station workers, gas station attendants, mechanics, insurers, car dealerships, and more besides. This change could be navigated, if the aggregate loss of individual private value i.e. all those jobs, is replaced with a creation of broad public value.
That “if” is about the size of Texas, however, and depends on whether we can ensure that the value created by new mobility services is publicly owned and managed, recognising a resurgence in public sector innovation, and balancing global with local, grid with non-grid …
So far, however, the debates around Uber, a harbinger of what’s to follow, indicate little appetite for addressing these concerns. Radical counter-balancing ideas to that potential loss, such as a universal basic income, will need sophisticated strategic presentation if they are to be considered affordable or palatable, in order to achieve a more equitable city. An Overton window will need prising open.
Leaving aside the impact on jobs, the impact on the places we live will also need careful calibrating via bold policy design and implementation. For example, the most transformative approach to autonomous mobility with cities in mind would be a fully autonomous shared system. Research by ETH in Zürich suggests that such a shared autonomous on-demand fleet of ‘taxis’ could remove as much as 80–90% of the private vehicles from many of our cities, if we wanted. Imagine the hugely positive impact on city space and street life in almost every dimension, never mind broader returns on safety, health, environmental qualities, economic livelihood and equality of access. (This year, my team at Arup has done detailed design schemes of streets under conditions of autonomous mobility, including understanding their broad strategic impact as well as what these streets might feel like, what might they be used for. They are potentially wonderfully open, intensely green, malleable public spaces – if we design the conditions to enable that.)
The humans in this kind of loop, a rekindled circular relationship with our streets, would benefit hugely, certainly more so than by being able to grab a steering wheel every now and then.
Yet it’s worth noting that we could achieve these numbers without AVs if we wanted to — look at the centre of any old Italian city — just as we can achieve safer streets through Hans Monderman’s ‘shared space’ principles, road-tested for decades in Dutch towns now. Do we need AVs to pull focus on these principles? Just how did all those cars get there in the first place? Are we so incapable of making good decisions about urbanism that we need to use these autonomous chariots as Trojan Horses?
We are already beginning to share vehicles in a way unthinkable a decade ago. Already 50% of all journeys on Uber and Lyft in San Francisco are shared journeys. San Francisco is not typical in this sense, so further questions are hidden here: how will sharing play out in the likes of Sheffield, Sapporo, Stockholm, Soweto, Siena, Shajah, Surat, Santiago? All have subtly different sharing cultures.
From my conversations with automobile manufacturers, it seems even the appropriate size of a shared vehicle is a known unknown. A promising lead might be to explore a S, M, L, XL approach to such vehicles, breaking down the large units that currently dominate our streets — garbage trucks, fire engines, buses — into networked distributions of smaller, quieter units, almost a packet-switched approach to transit, in terms of dynamics. With integrated thinking and practice, fire mitigation and garbage recycling, for example, can be better baked into buildings and places—this means smaller fire engines need carry only a few firefighters, smaller garbage trucks need only shift radically reduced trash. Simply making large trucks into autonomous large trucks would be missing a trick, an opportunity to minimise, to re-think the form. Other smaller units might include pods, four-seaters, e-bikes and all points in-between. Others may be more like shared minibuses, shuttles, taxis, ferries, or small vans, depending on need. The bus would far more transformative than the car.
Managing this Richard Scarry-like fleet will take careful yet decisive urban policy; imagine the impact of Uber’s plans to deliver anyone and anything anywhere, and then scale that up. To Uber, everything looks like a car delivering something, just as with General Motors and their own ‘a hammer only sees nails’ approaches. Without urban policy shaping what hits our streets, there is a strong chance that individually-owned autonomous vehicles would substantially increase congestion, constantly and absurdly circling around to avoid parking, just as Uber et al are vastly increasing traffic on the road at the expense of far more effective modes of transport.
Yet a shared autonomous fleet — which we could simply call public transport, and run as such, so we can calibrate it— could genuinely free our cities. If shared and public, we could achieve that volume reduction in overall traffic and all its attendant benefits.
Thus the core question there is not one of technology as such, but one of ownership. Would shared autonomous shuttles, on-demand mini-buses constantly scurrying around in the gaps left by ‘grid-model’ mass transit, genuinely be public transport, in which value is generated and retained locally, spread equitably, politically accountable and capable of integrating alongside other services and plans? Or is it something else?
And if conventional cars do continue to exist — and as cultural objects, they are about more than simply getting from A to B — are they accessed via services for pure driving pleasure rather than owned? That way you might enjoy a top-down tour of the Dolomites at the weekend, in a car of your choice, whether a bugged-out Beetle or a ’72 Ferrari Daytona. Just like the car adverts, in fact. Or do conventional, 20th century-style cars still have a place in the everyday, in the utilitarian? Does actual ownership of individual cars become an increasingly rare pursuit, akin to owning a horse i.e. something for rich people, or something for the weekend. (The rest of us occasionally ride horses for fun, at purpose-built centres for doing so.) The horse used to be our primary vehicle in cities and elsewhere. Now it isn’t.
Perhaps the biggest question of all posed by Joshua Brown’s Tesla, its top peeled off like a tin can, is how we do this? The introduction of cars, on a mass scale, was one of the worst non-decisions we’ve taken — in terms of deaths and injuries, carbon, air quality, obesity, social fabric, equality, and general deterioration of urban space and city life. We must reconcile ourselves with the true effects of that un-made decision as we face a decision we now have to collectively make. In Crash, Ballard wrote of wanting to “rub the human face in its own vomit and force it to look in the mirror.” Of course, he wasn’t writing solely about cars, but it’s no accident he used the car crash to pull focus.
There’s a saying: ‘when you invent the motor car, you invent the traffic jam.’ Yet we didn’t look at the traffic jam then, and we should pause to consider whether we’re looking at its equivalent now.
We are on the verge of a revolution equivalent in scale and impact to the introduction of 20th century cars. Yet this time it is intentional. Indeed, it must be intentional, as this new technology has to prove that it is substantially improved on an vastly widespread, highly regulated, existing model.
So, we have a choice as to how it happens. AVs have huge potential for transforming how we live, work and play, and the places in which we do those things, for the better.
But are we any better prepared for this broader debate about our cities? Is Tesla driving towards this future too fast? Are they the equivalent of General Motors leaning on Los Angeles to remove its tramways, technology driving decision-making? Conversely, are legacy car manufacturers too resistant, the chandlers ignoring the transformative potential of electric light on our cities? Are policymakers better practiced in making decisions about good urban outcomes either way? Design’s role in framing questions here may be more useful than styling a chassis, but in shaping a broader understanding of what’s at stake, and what agency we actually have.
Could we take a more active decision this time, and with two key outcomes in mind in particular? Firstly, can we achieve this largely positive shift, but with equitable civic gain in mind as well as individual? Secondly, can we actively look to prevent people dying in the process, without humans being crash test dummies and “liability sponges”?
Could poor old Joshua Brown be the first and last casualty of this transformation?
A shorter edit of this piece was first published in Disegno #12, 19 September 2016. It was written during July 2016.