Patrick Lin: The ethical dilemma of self-driving carsThis is a thought experiment.
这是一个思想实验。
Let's say at some point in the not so distant future, you're barreling down the highway in your self-driving car, and you find yourself boxed in on all sides by other cars.
假设在不久的将来的某个时刻，你正乘坐自动驾驶汽车在高速公路上疾驰，发现自己被其他汽车四面围住。
Suddenly, a large, heavy object falls off the truck in front of you.
突然，一个又大又重的物体从你前面的卡车上掉下来。
Your car can't stop in time to avoid the collision, so it needs to make a decision: go straight and hit the object, swerve left into an SUV, or swerve right into a motorcycle.
你的车无法及时停下以避免碰撞，所以它需要做出决定：直行撞上物体，向左急转弯撞上一辆SUV，或者向右急转弯撞上一辆摩托车。
Should it prioritize your safety by hitting the motorcycle, minimize danger to others by not swerving, even if it means hitting the large object and sacrificing your life, or take the middle ground by hitting the SUV, which has a high passenger safety rating?
它应该通过撞击摩托车来优先考虑你的安全吗？或者通过不转弯来最小化对他人的危险，即使这意味着撞上大物体并牺牲你的生命？还是采取折中方案，撞向那辆乘客安全评级很高的SUV？
So what should the self-driving car do?
那么，自动驾驶汽车应该怎么做？
If we were driving that boxed in car in manual mode, whichever way we'd react would be understood as just that, a reaction, not a deliberate decision.
如果我们手动驾驶那辆被围住的车，无论我们如何反应，都会被理解为只是一种反应，而非深思熟虑的决定。
It would be an instinctual panicked move with no forethought or malice.
那将是一种本能的恐慌举动，没有预谋或恶意。
But if a programmer were to instruct the car to make the same move, given conditions it may sense in the future, well, that looks more like premeditated homicide.
但如果程序员指示汽车做出同样的举动，考虑到它未来可能感知到的条件，那么，这看起来更像是有预谋的杀人。
Now, to be fair, self-driving cars are predicted to dramatically reduce traffic accidents and fatalities by removing human error from the driving equation.
公平地说，预测自动驾驶汽车将通过消除驾驶中的人为错误，从而大幅减少交通事故和死亡。
Plus, there may be all sorts of other benefits: eased road congestion, decreased harmful emissions, and minimized unproductive and stressful driving time.
此外，可能还有各种各样的其他好处：缓解道路拥堵，减少有害排放，并最大限度地减少无益且充满压力的驾驶时间。
But accidents can and will still happen, and when they do, their outcomes may be determined months or years in advance by programmers or policy makers.
但事故可能且仍然会发生，而当它们发生时，其结果可能在数月或数年前就由程序员或政策制定者决定了。
And they'll have some difficult decisions to make.
他们将需要做出一些艰难的决定。
It's tempting to offer up general decision-making principles, like minimize harm, but even that quickly leads to morally murky decisions.
人们很容易提出一般性的决策原则，比如“伤害最小化”，但即便是这样，也会迅速导致道德上模糊不清的决定。
For example, let's say we have the same initial set up, but now there's a motorcyclist wearing a helmet to your left and another one without a helmet to your right.
例如，假设我们有相同的初始设置，但现在你左边有一个戴头盔的摩托车手，右边有一个没戴头盔的。
Which one should your robot car crash into?
你的机器人汽车应该撞向哪一个？
If you say the biker with the helmet because she's more likely to survive, then aren't you penalizing the responsible motorist?
如果你说撞戴头盔的车手，因为她更有可能幸存，那么你岂不是在惩罚负责任的驾驶者？
If, instead, you save the biker without the helmet because he's acting irresponsibly, then you've gone way beyond the initial design principle about minimizing harm, and the robot car is now meting out street justice.
相反，如果你救那个没戴头盔的车手，因为他的行为不负责任，那么你就远远超出了最初关于“伤害最小化”的设计原则，机器人汽车现在是在执行街头正义了。
The ethical considerations get more complicated here.
这里的伦理考量变得更加复杂。
In both of our scenarios, the underlying design is functioning as a targeting algorithm of sorts.
在我们的两个场景中，底层设计都起到了一种目标选择算法的作用。
In other words, it's systematically favoring or discriminating against a certain type of object to crash into.
换句话说，它系统性地偏向或歧视某种类型的撞击对象。
And the owners of the target vehicles will suffer the negative consequences of this algorithm through no fault of their own.
而目标车辆的车主将因这个算法承受负面后果，尽管他们自己并无过错。
Our new technologies are opening up many other novel ethical dilemmas.
我们的新技术正在引发许多其他新的伦理困境。
For instance, if you had to choose between a car that would always save as many lives as possible in an accident, or one that would save you at any cost, which would you buy?
例如，如果你必须在一种事故中总是尽可能拯救更多生命的汽车，和一种不惜一切代价拯救你的汽车之间做出选择，你会买哪一种？
What happens if the cars start analyzing and factoring in the passengers of the cars and the particulars of their lives?
如果汽车开始分析并考虑车内乘客及其生活细节，会发生什么？
Could it be the case that a random decision is still better than a predetermined one designed to minimize harm?
有没有可能，随机决策仍然比预先设定的、旨在最小化伤害的决策更好？
And who should be making all of these decisions anyhow?
那么，到底谁应该做出所有这些决定？
Programmers? Companies? Governments?
程序员？公司？政府？
Reality may not play out exactly like our thought experiments, but that's not the point.
现实可能不会完全像我们的思想实验那样展开，但这不是重点。
They're designed to isolate and stress test our intuitions on ethics, just like science experiments do for the physical world.
它们旨在隔离并压力测试我们的伦理直觉，就像科学实验对物理世界所做的那样。
Spotting these moral hairpin turns now will help us maneuver the unfamiliar road of technology ethics, and allow us to cruise confidently and conscientiously into our brave new future.
现在发现这些道德上的急转弯，将帮助我们驾驭技术伦理这条陌生的道路，并让我们能够自信且尽责地驶向我们美丽的新未来。
内容总结
这个思想实验探讨了自动驾驶汽车在不可避免的事故中面临的伦理决策困境，以及由此引发的深层社会与道德问题。
其主要内容如下：
1.核心困境：程序化的“预谋” vs. 人类的“反应”： 当自动驾驶汽车必须在瞬间做出涉及生死的选择（如撞向摩托车、SUV还是大型物体）时，其决策是基于预先编写的算法。这与人类驾驶员在恐慌中的本能反应有本质不同，前者像“有预谋的杀人”，引发了关于责任与道德的尖锐问题。
2.“最小化伤害”原则的局限性： 即便采用“最小化伤害”这一看似普适的伦理原则，也会迅速陷入道德泥潭。例如，在必须选择撞向一名戴头盔或未戴头盔的摩托车手时，无论选择哪一方，都会导致惩罚负责任者或超越原始原则、执行“街头正义”的矛盾后果。
3.系统性的歧视与公平问题： 自动驾驶汽车的决策算法在本质上成为一种目标选择算法，会系统性地偏好或歧视某一类道路使用者（如摩托车 vs. SUV，戴头盔者 vs. 未戴头盔者）。这将导致部分人无故承受算法带来的负面后果，引发严重的公平性质疑。
4.新技术引发的更广泛伦理问题： 实验延伸出更多亟待回答的问题：消费者会购买“利他型”还是“自保型”的自动驾驶汽车？算法是否应该评估乘客的生命价值或生活细节？随机决策是否比精心设计的伦理算法更可取？最终，谁有权（程序员、公司还是政府）来制定这些生死规则？
5.思想实验的目的与意义： 这些极端场景并非预测现实，而是像科学实验一样，用于隔离和压力测试我们的伦理直觉。提前识别和思考这些复杂的“道德急转弯”，对于引导我们安全、负责任地驶向由新技术定义的未来至关重要。
