IEEE Technology and Society Magazine - September 2020 - 18

practice. And it does not address the
miscategorization of many experimental vehicles: these vehicles may
be aimed at Level Four driving in the
future, but they can be no higher
than Level Three in practice, by SAE
definition, if a backup driver is necessary to maintain safe operation.

Moving from Levels of
Automation to Conditions for
Operation
The SAE levels, as originally developed and subsequently invoked,
have contributed to a particular narrative of self-driving vehicles. From
this view, there is a clear problem to
be solved and a race to achieve the
solution first. It's a view that suits

An innovation schema pointing
one way, but saying nothing about
the desirability of the destination,
makes a poor roadmap.
some technology developers, but
does little to help societies make
good decisions about technology.
As Pittinsky suggests elsewhere in
this issue, the responsible development of algorithmic technologies
demands consultation with diverse
perspectives [23]. Regulators, consumers, transport planners, and citizens need new ways to talk and
think about the possibilities and limits of self-driving vehicles.
First, new typologies should start
with the recognition that, if they
come from authoritative sources,
they are devices for communication as well as analysis. There is
therefore a responsibility to publicly clarify the limits as well as the
possibilities of particular systems.

This demands a greater focus on
the operational design domain and
varied options for human-machine
collaboration and interaction, and
a downplaying of autonomy and the
direct replacement of human beings
with machines. To the extent that
the driving task is a focus of a new
framework, it must be open to new
arrangements of shared humanmachine control and collaboration.
And developers should avoid using
a numbered levels structure that
implicitly orders the categories in
terms of difficulty and value. But if
AVs are going to change the world,
we also need to know more about
technologies' relationships with
their contexts. Policymakers and the
public need clearer information about the conditions in
which particular automated
devices can operate, and
the additiona l cha nges
that might be required in
order for such systems to
be safe, equit a ble, a nd
effective. This means less
focus on the "driving task"
a nd more a t t e n t i o n t o
place, infrastructure, and
road rules.
On infrastructure, we might
look to the Infrastructure Support
levels for Automated Driving (ISAD)
recently proposed by the INFRAMIX
project [24], which seeks to categorize parts of roads according to
their connectivity. To this, we could
add consideration of physical as
well as digital infrastructures. Material arrangements of roads and road
furniture are harder to standardize than digital systems, and vary
more from place to place. In addition to varying road types, places
are defined by various cultures and
patterns of road use, which might
make some AV systems inadequate
or wholly inappropriate.
Social factors on the roadway,
neglected by the levels today, are
IEEE TECHNOLOGY AND SOCIETY MAGAZINE

therefore of similar importance
to physical and digital infrastructure. The SAE levels have little to
say about the behavior of other
road users, but AV developers are
now starting to admit that the success of their systems may depend
upon more predictable patterns of
behavior from cyclists, pedestrians,
and others [25]. New typologies for
AVs should therefore be explicit
about what else is required for the
systems to function as designed.
The real benefits of "autonomous"
vehicles will come when they are
embedded in and able to work with
whole systems, including other road
users and physical and digital infrastructures. We need ways to evaluate such systems, and ask old but
important technology assessment
questions: Who pays? Who benefits?
Who decides?
Finally, policymakers need clearer ways to talk about technologies being tested and technologies
being deployed. The widespread
use of safety drivers as a fallback
for prototype self-driving cars may
be necessary for their safe development, but it means that testing for
Level Four is happening, in effect, at
Level Three. This means that these
vehicles potentially come with all of
the hazards of mixed-mode operation including mode confusion,
automation complacency and problems of handovers. Notional future
readiness for Level Four operation
should not be allowed to confuse or
distract from the real problems and
risks that arise in testing and development, especially since the levels
do not represent linear increases in
capability. New typologies should
aim for clarity about the conditions
for testing as well as the conditions
of use, and ensure that developers
are being realistic about what will
be necessary for their systems to
be safe and effective in reaching
stated goals.
∕

SEPTEMBER 2020

IEEE Technology and Society Magazine - September 2020

Table of Contents for the Digital Edition of IEEE Technology and Society Magazine - September 2020