Mark Fields, Ford’s CEO, announced earlier this month that the company plans to offer fully self-driving vehicles by 2021. The vehicles will come without a steering wheel or pedals.1
At their 2016 annual shareholder meeting, BMW CEO Harold Krueger indicated that BMW will launch a self-driving electric vehicle, the “BMX iNext”, in 2021.2
General Motors executive Richard Holman said at a conference in Detroit that most industry participants now think that self-driving cars will be on the road by 2020 or sooner.3
However, the rise of autonomous motor vehicles raises important ethical questions that, quite frankly, I have not seen discussed. And these are questions that, frankly, need discussion. So, in this post I will attempt to provide a highly cursory outline of these issues. (Take note: this discussion will be much more technical than my normal post, so if you feel your eyes rolling into your head, feel free to click the back button!)
Throughout this post, I will use “SDC” to refer to Self-Driving Car. To refer to a specific Self-Driving Car (such as when giving an example), I will add a number after—so “SDC1” or “SDC2”, etc.
Now, without further ado:
BIG QUESTIONS THAT NEED ANSWERS:
How should a self-driving car (“SDC”), which is programmed to perform tasks in such a manner that minimizes risk, act when the interests (e.g. safety of life and risk of injury) of the SDC driver are opposed to the interests of external parties?
For Example: A Self-Driving Car, SDC1, with one occupant (a 34 year old male, who is also the owner), hits a patch of “black ice” on the highway which had been undetectable to the SDC1 at the time. Another vehicle, SDC2, is approaching dangerously close, and a decision must be made. SDC1’s software must now choose between two options:
Option A is to correct the wheel clockwise, which (as SDC1 accurately calculates in microseconds) carries a 50% chance of crash with a batch of large trees on the side of the road; such a crash would result in the driver’s instantaneous death 100% of the time. The other 50% of the time, Option A causes no harm to its sole occupant or to anyone/anything else.
Option B is to correct the wheel counter-clockwise, which carries a 50% chance of collision with SDC2, pushing both SDC1 and SDC2 dangerously toward the edge of an approaching overpass which is covered with even more ice. Inside of SDC2 are five persons: a 33 year old female occupant who is also the owner; a 39 year old male occupant; a 19 year old male occupant; an 16 year old female occupant; and a 1 year old male infant. If SDC1 were to collide with SDC2 under Option B, there would be an 80% chance of instantaneous death for each occupant of SDC2; for the sole occupant of SDC1, there would be a 25% chance of instantaneous death. All of this information is shared and calculated accurately by and between SDC1 and SDC2. The other 50% of the time under Option B, the collision is avoided and Option B causes no harm to any occupants of either vehicle.
(Please see the table below for a less mind-numbing layout of the numbers; regardless, however, the takeaway is that SDC1 must make the decision, and Option B is better for SDC1—although Option A is better for humanity in general. What is a computer to do?)
As outlined in the table above, Option A is more directly advantageous to the driver of SDC1; but, the external harm caused by Option A is substantially higher, in that a much greater risk of death is imposed on all five occupants of SDC2.
This creates a moral dilemma for Self-Driving Car 1. Should its software be designed to best protect its driver/occupants? Should it be designed to minimize total harm to all persons? Would you feel comfortable purchasing the car knowing that it valued your safety at the same level as the random driver next to you on the highway?
None of these questions are easy to answer.
IS THERE AN ACCEPTABLE “RANGE”?
In the context of autonomous vehicles, is there an acceptable range within which self-safety can be prioritized with respect to external-safety? I think the answer has to be yes, and here is why:
Let “ERA” mean External Risk Allocation; ERA is a setting within an SDC that allocates risk between
(i) the SDC / its occupant(s), and
(ii) the external (other drivers and their occupants, etc.)
ERA can be between 0 (that is, zero percent external risk) and 1 (that is, 100 percent external risk). When ERA is set to 0, the SDC will behave entirely selflessly; it will avoid causing $100 of damage or a bruised thumb to another person, even if it means that its own occupant dies as a result.
On the other end of the spectrum, when ERA is set to 1, the SDC will behave entirely selfishly; it will avoid the smallest amount of damage to itself/its Occupant, even if certain death to an external car full of people is the end result.
Bear with me here. If you agree that:
(A) self-driving cars can be acceptable in society;4
(B) self-driving cars should engage in risk-allocation within their own programming; and
(C) both 0 (total selfless and harmful to the driver) and 1 (total selfish and harmful to the external) would be unacceptable ERA settings, then there must be some range of settings in between 0 ERA and 1 ERA that are acceptable.
MORE BIG QUESTIONS THAT NEED ANSWERS:
What is the range of settings between 0 ERA and 1 ERA which are acceptable? Who gets to decide? And who will/should be responsible for covering the cost of collateral damage caused by SDCs with high ERA settings?
To what extent, if at all, should SDCs in general be permitted to prioritize the safety of their own occupant(s) over the safety of external persons?
Who should the manufacturer of the SDC and the SDC risk software owe a duty to—the owner, the owner and all of his/her invited occupants, or essentially members of the public? What should the rule be for situations where these duties directly conflict with one another?
Should private individuals be permitted to forfeit certain risk allocation rights/privileges under a freedom of contract theory? For example, should a person be permitted to receive a 75% sales discount on a SDC due to the SDC operating with an very low, non-adjustable level of ERA (External Risk Allocation) that will put him in harm’s way? What if the car is designed to not allow passengers?
Should liability shift if a SDC’s program miscalculates risk? (For example, by swerving into a light post to avoid running over a small animal crossing the road, which the program misidentified as a human child lying in the street.) If so, to whom and how should it shift?
I don’t pretend to have the answers to all of these questions. These are complex legal – and more importantly, moral and ethical – issues that our society is going to need to confront sooner rather than later. Furthermore, for attorneys, this represents both the gradual decline of an old industry (the personal injury case stemming from automobile accident) and the creation of a new industry (the allocation of risk and liability for autonomous motor vehicles). For these reasons, corporations throughout the United States, as well as our legal education system, will also need to adapt.
The future we live in can be as bright as we want it to be, but these are issues we must start considering now. And I place this obligation primarily on the manufacturers of autonomous motor vehicles and the developers of their software—but it’s also up to us as a people to make sure we are protected.
Self-driving cars will save many, many lives and hundreds of billions in collateral damage over the next 25, 50, and 100 years—but those who fail to plan, plan to fail, and it’s time to start planning now.
Thank you very much for reading, and your comments are always welcomed.
~ Chris Irak
Sources:
1Reuters, 2016-08-16
2Elektrek, 2016-05-12
3Wall Street Journal, 2016-05-10
4If you disagree with (A), consider that a 2015 study by Michele Bertoncello and Dominik Wee of McKinsey & Company predicts that self-driving cars will reduce traffic fatalities in the U.S. by up to 90% by midcentury.
