Moral robots and the ethics of self-driving cars
Are all lives equally valuable? Before we tackle the general question, first notice that liberals and conservatives disagree on what constitutes a life. Conservatives think that life begins at conception, and thus, abortion is murder and should be a crime (albeit, an unpunished crime). Liberals think that a mother should have full choice over whether to keep or end her pregnancy. These two views can not be reconciled.
The moral questions get interesting, though, if you think about edge cases. Neither liberals or conservatives are really consistent. Suppose that a man assaults a pregnant woman, and her fetus dies as a result. Should the attacker be charged with murder or assault? I think most liberals would call this murder, but it’s inconsistent, then, that an abortion is not murder.
The conservative viewpoint gets muddled, when you consider In Vitro Fertilization. IVF clinics combine sperm and eggs to make embryos and implant these to help infertile couples have children. Typically, the clinics create many embryos and don’t implant all of them. If life begins at conception, then the clinic is creating many souls and murdering any embryos that get discarded. Should IVF be illegal? How much is an embryo worth? A few questions to sort this out:
Imagine that someone maliciously enters a clinic and destroys a container with 1,000 embryos in it. Should that person be charged with a property crime or mass murder?
Imagine that you are in a fertility clinic that’s burning down. As you run out of the building, you can either save one screaming infant or a refrigerated container full of 1,000 embryos. Which do you choose?
Is a fetus more valuable than an embryo? Suppose you’re driving, your car loses control, and you’re forced to choose to swerve into one of two groups of pedestrians, either three men or two pregnant women. Are we comparing 2 lives against 3? Or 3 against 4? Which is the better moral choice to make?
My wife suggested that two women are always worth more than three men, whether the women are pregnant or not. That opinion alarmed me a bit — I assumed that most people value adult lives equally. I spent some time researching whether this was a common view.
II.
The trolley problem is a classic test of how people think about utilitarian morality.
There’s a train speeding up to a junction. If it goes straight, it will kill five people standing on the tracks. There’s a lever you can pull, to divert the train onto another track, and there’s only one person standing on those other tracks.
Most people agree that it’s better to pull the lever, to save five lives at the expense of losing one.
Differences in how the question is posed can change the outcome, though. An alternative version of the thought experiment suggests that you are on a bridge, watching an oncoming vehicle approaching 5 people. There’s a fat man next to you, and you have the chance to push him over the edge, in front of the vehicle, to stop it. Most people surveyed agree that this is immoral, as you’re killing the man that you push. The moral calculus of lives saved is the same, but the act of physically murdering someone registers as worse than pulling a lever (Note how little attention America’s ongoing drone strikes get).
Thought experiments also don’t reflect how people would really act, in a crisis. How do people actually react to the Trolley problem, if you simulate it?
Well, first off, here’s how a 2 year old solves the problem:
Basically, toddlers are psychopaths.
Mindfield made a convincing simulation of the trolley problem, putting adult test subjects into a switching room with video cameras of the tracks and a train switching lever. Test subjects were tricked into thinking that they were in charge, after the rail company engineer left the room to take a phone call. Watch a few minutes to see how subjects actually responded to the dilemna:
Of the seven test subjects, only two made the choice to save five lives at the expense of one. The other five test subjects got anxious but did not pull the switch. Each gave various rationalizations for the choice, but it seems that the primary reason is that they froze in indecision or fear.
While these questions seem highly abstract, some moral decisions of the future will be solved by artificial intelligence. Self driving cars of the future will occasionally be forced to choose one life over another, if the car is headed for a crash.
An experiment at MIT, the “Moral machine”, tried to quantify which choices humans think that self driving cars should make, by giving a series of questions about where a car should go, in the event of an unavoidable crash. Should a self driving car choose to hit a man or a woman? A child or an adult? A jaywalker or someone crossing the street legally? Should the car crash itself and harm the driver to avoid hitting a larger group of people?
You can take a moment to try the experiment for yourself, to probe your preferences.
We generally agree that saving more people is the right choice, but beyond that, people will pick and choose victims. #AllLivesMatter, but women and children matter a bit more, a criminal is worth more than a cat but less than a dog, being old is worse than being homeless which is worse than being obese.
Beyond the basics, preferences vary by culture. People in Asia care less about saving the young, and give more value to the elderly. Both men and women in all countries agree that women’s lives should be saved sooner than men’s (there’s that patriarchy, oppressing women yet again) but Latin Americans value saving women even more than other countries do. Latin Americans also value high status people more than low status individuals.
After being given many questions by the moral machine, I personally decided that one case trumped all the others — I decided that the car should always drive where pedestrians were jaywalking, instead of where they were legally crossing the street.
I didn’t think of this for moral reasons, exactly, I was just trying to imagine a future that I could safely live in. If I know that I’m generally safe while crossing at a protected light, then I can plan my day so as not to get hit. Otherwise, I’m at the mercy of whatever algorithm Waymo or Uber is using to value my life (maybe Waymo hired too many intersectional feminists and their car has made the decision to run me over to erase my white male privilege).
People in organized societies are generally more likely to think that the legality of who’s crossing the road is important in making moral decisions. People in poorer countries aren’t familiar with the idea that laws are consistent and trustable, so they don’t judge things this way.
One important question that’s left out of the trolley problem is whether you’d be willing to sacrifice your own life to save five strangers.
Most people agree that it’s ethical for a car to crash itself, risking the life of the occupants, if this would save more pedestrians. Paradoxically, most people say that they would not buy a self driving car which was programmed to make this choice.
When self driving cars become legal, in capitalist countries, I predict that each car will be programmed to save the lives of its occupants at the expense of possibly hitting multiple pedestrians.
In socialist countries, there’s a desire to make all people’s outcomes equal. Historically, this has been a difficult problem, as some people work harder than others, some are more talented than others, some are born with better luck or privilege. I expect that the self driving socialist car will spend its day crushing all pedestrians in sight, to ensure absolute equality.
III.
Why does anyone sacrifice their life for others? Where does altruism come from? Typically, evolution should only allow selfish organisms to reproduce. We care about saving ourselves and our children, and all other behavior should be a penalty. Genes can propagate, though, if they help distant relatives. If your siblings share 50% of your DNA and your cousins share 12.5%, it would still make sense to give your life to save 2 brothers or 8 cousins.
Prairie dogs and similar mammals will stand up and chirp when a predator is near, to alert the colony. This behavior puts the chirping animal’s life at more risk from the predator, but helps the colony as a whole stay safer.
Human societies are selected not just genetically, but also by cultural competition. A culture with some self sacrificing tendencies may survive over a selfish one. In particular, a patriotic state that sends its sons to war will defeat a selfish culture where individuals choose not to fight. This willingness to fight varies by culture.
The NFL kneeling controversy was a big “choose a side” debate between respecting “those who have died fighting for our country” and “those kneeling to protest police violence against blacks”. At the time, I wanted to quip that this was funny, since black people are overrepresented in our military. Aren’t we respecting black victims either way?
I looked into the numbers, and found out that line of thought was wrong. Black people are overrepresented in the military, making up about 20% of the army and the navy, even though they make up about 11% of the civilian population.
If you actually look at casualty numbers, though, the numbers are closer to the demographics of America. White people are actually a bit overrepresented in casualties and blacks underrepresented.
This is because the units that see the most combat have a different racial mix. The infantry is disproportionately white, and blacks are evenly represented in the marines. One commentator suggests that white and black people join the military for different reasons — blacks are looking for job skills and opportunities, so they join combat support units, whites are looking for adventure and they join combat units.
The overall demographics of the military are also thrown off because black people tend to stay enlisted for longer, possibly finding more opportunity there than they would outside of the military. Recruitment numbers for each year are closer to the average racial mix in the US.
I would have guessed that poor people join the military more often than richer ones. It turns out that’s not true, either — the poorest demographics are underrepresented, probably because the military aims to mostly recruit high school graduates, and the poorest people in America don’t graduate as often. That study shows increased recruitment the richer a family is. I feel like that trend must break down at some point, I doubt that many millionaires send their kids into the military, but the data isn’t really good enough to see that, since wealth is estimated by neighborhood, not by family.
There are also strong regional variations in terms of who is willing to fight:
And there are differences in ideology. 60% of veterans voted for Trump, 34% for Clinton. Rural, white Americans dislike the NFL protests because they espouse a more militant, patriotic attitude.
Cultures, like animal colonies, thrive through their collective behavior and their collective ability at group defense and conservative cultures are more focused on survival values.
This should select for the most militant cultures possible and, historically, it sometimes has. In modern times, technology has reduced this competition, as a wealthy country can defend itself with less soldiers and less lives at risk. The collectivist soviet union did a better job than western countries at amassing a large military and should have been able to defeat NATO in a conventional war for control of Europe. Nuclear weapons made this style of conflict obsolete, though, and allowed individualistic countries to thrive.
In the future, a patriotic, militant ethic may also become unnecessary as we replace human soldiers with killing robots. If people disagree about how best to program a car to drive morally, then imagine how hard it will be to make robotic soldiers that minimize collateral damage.
IV.
Utilitarianism tries to reduce morality to simple math. In life or death cases, an action is decided based on the net number of lives saved. A hostage rescue is a success if killing one hostage saves five. Shooting a probable criminal is moral if the odds are that at least one murder was prevented.
Outside of life and death, the goal is to maximize happiness or well being for as many people as possible. As a useful example, the Copenhagen Consensus tries to prioritize humanitarian interventions in terms of return on investment, asking where a charity can produce the most good for the least cost. The winning interventions tend to be in terms of eradicating common diseases or supplementing micronutrients (say, vitamins or iodine) for young children to grow healthier, smarter kids.
What’s left out of these calculations is quantity of life, not just quality. If a medical intervention saves 50,000 lives, would it be just as effective to encourage people to have 50,000 more children? It’s easy to say that we should save as many lives as possible, but how many lives should we create?
Imagine two worlds. One has a billion people who each have an income of $100,000 per year. Another has ten billion people, who each have an income of $10,000 per year. Which world would you rather live in? Which world do you think is a happier place?
Another philosophical argument, the mere addition paradox, suggests that having more people in the world is always better, even if some people end up less happy as a result. The argument is typically made in 4 steps:
Group A isn’t any less happy if you add a separate, second group (A+) that’s less well off. B- is happier than A+, one group has lost a bit, but on average both are doing better. B is no worse than B-, merging two equal groups shouldn’t change anything. Group B must be happier than group A.
The argument is even stronger if you consider that happiness doesn’t vary that much, by country, in the world today. A person that is twice as rich is not twice as happy. If twice as many people share the world’s wealth, the average level of happiness will go up.
Taking this argument to its logical end (the repugnant conclusion) suggests that we’re better off with a huge number of people living in some just bearable level of poverty rather than with a smaller number of wealthy people.
Is the optimal future a Malthusian nightmare where we push up against the planet’s resource limits? Or is it optimal to create high quality of life for a smaller number of people?
The real problem, I think, is that we have no idea what the planet’s resource limits are. Human population has been growing exponentially for hundreds of years. In nature, exponential populations don’t approach limits and then stabilize, they exceed bounds and then crash. The snowshoe hare, common in Alaska and Canada, breeds very rapidly, but rather than seeing long-term exponential growth, its population goes in cycles.
After a few years of rapid growth, the rabbits eat too much vegetation and start to starve. Meanwhile, lynx numbers rise as the cats start eating the plentiful rabbits. Eventually, the rabbits starve, then the cats starve, and the cycle repeats every ten years or so.
In the rat utopia experiments, overpopulation in a group of caged rats lead the population to subsequently crash all the way to zero.
Humans have no predators, but at our unprecedented rate of population and consumption growth, we may overuse whatever resources we need. There’s a time lag between consumption and depletion, so we won’t know that we’ve exceeded limits until it’s too late.
The key thing is to make sure that the future lasts for as long as possible, the key goal is to find stability and avoid existential risks to civilization.
The same logic applies with animal cruelty. Slatestarcodex attempts to quantify how intelligent a given species is, and then suggests possibly using this as a metric for which animals we should and should not eat. Some findings are unexpected. Lobsters are incredibly dumb and probably don’t suffer much. Eating beef is better than chicken for animal suffering because it kills less animals — a cow is smarter than a chicken, but yields vastly more meat.
Likewise, with animals, I think we shouldn’t worry much about short term suffering but instead worry about a major long term problem — extinction. Eating beef is not a terrible crime, as long as cows are free to live reasonably good lives. Cutting down forests on Borneo, on the other hand, might drive Orangutans to extinction, and the world will be permanently impoverished by that loss.
V.
There’s a problem here that utilitarianism doesn’t solve. If we know the carrying capacity of the planet, then the happiness optimizing goal should be to reach that population limit and keep society at that point. Now, suppose that people still want to have more children. Who gets to fill the limited number of available spaces? Mathematical reasoning doesn’t help here, all we’re left with is competition between people and groups.
Our world is more complicated than one single equation, though. At the moment, the world faces simultaneous trends with growing populations and with shrinking populations.
Couples in wealthy countries are generally having less than 2 children and such countries will shrink, barring significant immigration:
Meanwhile, poorer countries in Africa are growing without much end in sight, adding 3 billion people in the next century:
Bill Gates warns that Africa’s population growth is fast enough that it will prevent progress at fixing poverty there. In a utilitarian sense, Africa will be a happier place, but a vastly less stable one.
The drive to share wealth between small, rich places and the rest of the world will be strong. Immigration struggles and nationalist movements in the 21st century will likely dwarf anything we’ve seen to date.
Utilitarianism doesn’t give us an answer as to what to do here, but other values might. Conservative values suggest only national borders and group conflict. Liberal values promote equality for women, which is the single best way to slow population growth.
