Cybersecurity as a political and ethical value in autonomous
vehicles
The work is supported by theNational Science FoundationGrant: SES - 1650589 and the Simons Institute for the Theory of Computing at UC Berkeley
Analyzing ethical and political dimensions of control handoffs
Based on existing and forthcoming work
Outline
• Handoff analytic
• Competing visions for autonomous vehicle futures
• Business models and technical configurations
• Open source / proprietary approaches
• The handoff model is designed to explore the Values implications of the transfer of responsibility for a Function within a given System
between Components (particularly human and machine)
• When a system control in a system is handed off from a human to a mechanical system, or from a mechanical to a computational system, along side functional performance, the handoff analytic suggests the new configuration of actors and Modes of acting has consequences for values (like privacy, security, autonomy, responsibility etc)
• Decomposing how systems work rather than simply what they do (at the correct level of abstraction), before and after the control handoff, makes political and ethical consequences apparent.
• Requires mapping the introduction of new components in a system (human, physical or material, computational).
• Considering how they accomplish their various functions.
• Interrogating the mode of acting – by force, by affordance, by nudging, by transmission/communication etc…
• Identifying who owns/controls the various components, and what their agenda (and triggers) might be.
• The goal is to assist analysts / policy makers identify what is at stake in control transitions.
• A bit like Actor Network Theory deployed as a tool for practical ethics / values in design analysis.
Common now
Tesla and other vehicle ADAS
Waymo One
Fantasy / military Thinking only in terms of
functional performance
Autonomous Driver Assist Connected
Autonomous Cars
All sensing and computation on-board
$$$$$$$
Nissan ‘SAM”
Phantom Auto
Non-autonomous driving control system
Input route
Teleoperate
Select destination / Price Support
Feedback Support
Start/Stop Map / Status
Destination Route Start/Stop Update
Sensor data Training
Sensor data Gather data
Ride-Hailing Business Model
Instruct vehicle
Waymo One Interface determines
‘identity’ of occupant Tracking user autonomy through affordances of interface
Driver Assist Cars
Private ownership business model
Human input control
Vehicle communicate status Control transition (affordance)
Telemetry data Compliance data
Driver monitoring
Instruct vehicle
Make sensor data human readable Maybe a decision maker
Determine control transitions
Insurance (i.e. Tesla)
Augmented reality windshield
Connected cars
Connected cars
Input destination Control input
Control input
Control input
Control input
Control input Control input Control input
Control input Sensor data
Sensor data Sensor data
Sensor data V2P
V2I
V2V
Autonomy when humans cannot supervise vehicle control?
Input destination Control input
Control input
Control input
Control input
Control input Control input Control input
Control input Sensor data
Sensor data Sensor data
Sensor data V2P
V2I
V2V
Open source / proprietary software in autonomous vehicles
Private ownership business model Use of tech company driving
control platforms in OEM manufactured cars
Fig – Developer Economics
Currently 23% open source Push for more open source
• Reactive vs proactive?
• Influences on preferable approaches:
• Consequences
• Attack surface size
• Private ownership / fleet ownership / ride-hailing?
• Autonomous / Driver Assist / Connected
• How many controlling components
• Nature of interaction
• Ownership / Integration of ownership
• Interoperability?
Cybersecurity as a political and ethical value in autonomous
vehicles
The work is supported by theNational Science FoundationGrant: SES - 1650589 and the Simons Institute for the Theory of Computing at UC Berkeley
Analyzing ethical and political dimensions of control handoffs