人工智能与自动驾驶汽车
目录 (Table of Contents)
The New Decision Makers
新决策者
Quick Fix
快速解决
Problem Perspective
问题视角
Braking to the Rescue
急救
The Path Forward
前进的道路
新决策者 (The New Decision Makers)
We make decisions everyday. We, as humans, make decisions regarding our lives, which may impact others, every single day. But what about when the decisions made by machines have the ability to impact human lives more drastically than ever before? Some of the world’s biggest tech companies now have self-driving car programmes. These companies argue that the vehicles could improve road safety, ease traffic and improve fuel efficiency, and their claims do have a scientific foothold to stand on.
我们每天都会做出决定。 作为人类，我们每天都会做出有关生活的决定，这可能会影响他人。 但是，当机器做出的决策比以往任何时候都能够对人类的生活产生更大的影响时，又该怎么办呢？ 现在，一些世界上最大的科技公司都有自动驾驶汽车计划。 这些公司认为，这些车辆可以提高道路安全性，减轻交通负担并提高燃油效率，他们的主张确实有立足之本。
But others say the cars raise ethical issues, and could have unintended consequences for public safety. This in turn kicks off a growing moral debate on autonomous cars’ decision making process in the event of a situation that could endanger lives. When a human is put in a life threatening situation, any way they react will be construed as just that — a reaction. While it is true that self driving cars will ultimately reduce accidents by removing human error from the equation, they cannot totally avoid them. But if self driving vehicles are programmed beforehand on how to react, this can be construed as premeditated violence, and subject to bias introduced by the programmers or policy makers.
但是其他人说，汽车引发了道德问题，并可能对公共安全产生意想不到的后果。 反过来，在可能危及生命的情况下，这也引发了关于自动驾驶汽车决策过程的日益激烈的道德辩论。 当一个人处于危及生命的境地时，他们做出的任何React都将被解释为React。 尽管自动驾驶汽车确实可以通过消除方程式中的人为错误来最终减少事故，但它们无法完全避免此类事故。 但是，如果事先为自动驾驶汽车编程了如何React的方法，则可以将其解释为有预谋的暴力行为，并容易受到程序员或政策制定者的偏见。
The debate about whether ethics are universal or vary between cultures is an old one, and now the problem in programming self-driving cars has reinvigorated it. Decisions made by humans can differ based on their social and economic conditions. But autonomous vehicles do not belong to any nationality, nor do they have reactions. They made decisions based on the rules programmed in them, and it is our responsibility to prevent these decisions from having any bias that could have crept in. Or is it better to make these decisions random, in order to eliminate the possibility of bias, at the risk of increasing accidents through unpredictable outcomes?
关于伦理学是普遍的还是不同的文化之间的争论是一个古老的争论，现在无人驾驶汽车的编程问题使它重新焕发了活力。 人类做出的决定可能会因其社会和经济状况而异。 但是自动驾驶汽车不属于任何国籍，也没有React。 他们根据其中制定的规则制定决策，因此我们有责任防止这些决策产生任何可能引起的偏见。或者，最好是随机做出这些决策，以消除出现偏见的可能性。通过不可预测的结果增加事故的风险？
快速解决 (Quick Fix)
An easy solution would be to hand over control to the human driver of the car, in the event of a collision, whose reactions would be on the spot and not premeditated. But unfortunately, this is not a fail safe, as in most situations, the human brain is not fast enough to process and make a decision on the spot.
一个简单的解决方案是在发生碰撞的情况下将控制权移交给汽车的驾驶员，驾驶员的React将是现场的，而不是有预谋的。 但是不幸的是，这并不是一种故障安全措施，因为在大多数情况下，人脑的速度不足以现场处理并做出决定。
问题视角 (The Problem Perspective)
Most ethical problems have different types of people on either side of the road, such as a young girl and an old woman, and the problem has somehow turned into deciding which classes of people deserve to live over others. It is far from clear that a self-driving car will actually be able to make distinctions of this type, and the moral repercussions of making these decisions are huge, as we are stuck in a moral bog of deciding which people to kill. Under the above assumptions, we describe a situation where in there are people on either side of the road, and a situation causes the vehicle to get into an unavoidable accident. Now the vehicle has 3 options-1. Swerve left and injure/kill the people on the left2. Swerve right and injure/kill the people on the left. 3. Continue on path and injure/kill the people in front (assuming the vehicle in front is within the vehicle’s stopping distance).
大多数道德问题在道路的两边都有不同类型的人，例如一个年轻女孩和一个老妇，并且这个问题已经以某种方式转变为决定哪些阶级的人应该生活在别人之上。 到目前为止，无人驾驶汽车实际上是否能够区分这种类型还很不明确，做出这些决定的道德后果是巨大的，因为我们陷入了决定杀害哪些人的道德困境。 根据上述假设，我们描述了道路两边有人的情况，并且该情况导致车辆陷入不可避免的事故。 现在，该车辆具有3个选项-1.向左倾斜并伤害/杀死左侧的人2。 向右倾斜并伤害/杀死左侧的人。 3.继续前进，并伤害/杀死前面的人(假设前面的车辆在车辆的停车距离之内)。
急救 (Braking to the Rescue)
One interpretation of the absence of braking in most of the cases might be that the brakes are presumed to have failed. This would be a wrong assumption, though, as brake failures on modern vehicles are rare and seldom unpredictable. A self-driving car will have acutely sensitive monitors of the state of its brakes, and will certainly be programmed to insist on being serviced long before brake failure becomes likely. A generic form of the self-driving car dilemma, in keeping with the above points -
在大多数情况下，没有制动的一种解释可能是认为制动发生了故障。 但是，这将是一个错误的假设，因为现代车辆的制动故障很少见，而且很少是不可预测的。 自动驾驶汽车将具有高度灵敏的制动器状态监控器，并且可以通过编程使其坚持在可能发生制动器故障之前很长时间进行维修。 符合以上几点的自动驾驶汽车困境的通用形式：
A properly-serviced self-driving car detects an unavoidable collision and must determine whether or not to swerve while braking — both braking in a straight line and swerving will result in collisions, but with different objects, neither of which is a road going vehicle.
维护良好的自动驾驶汽车会检测到不可避免的碰撞，必须确定在制动时是否要转向-直线制动和转向都会导致碰撞，但碰撞对象却不同，两者都不是上路车辆。
Swerving is a drastic manoeuvre — the stipulation of any self-driving problem, that the objects in the car’s path are within its stopping distance and thus cannot be avoided by simply braking, entails that any steering the car does will be sharp, and thus at significant risk of loss of vehicle control. In any environment sufficiently crowded that all paths available to the car result in collisions, a loss of vehicle control is much more dangerous than a controlled stop. From a practical ethical perspective, simple breaking to minimize potential impact might be a better choice than attempting to make a rational choice of who to kill.
转弯是一种剧烈的动作-任何自动驾驶问题的规定，即汽车路径中的物体都在其停止距离之内，因此无法通过简单的制动来避免，这意味着汽车的任何转向都将变得敏锐，因此在失去车辆控制权的重大风险。 在任何人满为患的环境中，汽车可利用的所有路径均会发生碰撞，失去车辆控制比控制停车要危险得多。 从实用的道德角度来看，比试图合理选择杀死谁的方法，简单地打破可能的潜在影响是更好的选择。
It takes some of the intellectual intrigue out of the problem, but the answer is almost always ‘slam on the brakes’ .
它消除了一些智力上的麻烦，但是答案几乎总是“踩刹车”。
~Andrew Chatham
〜安德鲁·查塔姆
The physics of vehicle dynamics shows that swerving is inherently much riskier than braking in a straight line. For a self-driving car to be able to determine that the increased risk of swerving is compensated for by the reduced risk associated with hitting a different object than the one immediately in front of it, it must have a great deal of detailed information about its environment. Much of this information is practically unavailable, either because of the level of detail required, or the limitations of car-mounted sensors. This entails that it is extremely unlikely that a self-driving car could ever reasonably be required to swerve.
车辆动力学的物理特性表明，转弯固有的危险性比直线制动好。 为了使自动驾驶汽车能够确定由于撞到另一个物体而不是紧挨其前方的物体而降低的风险，可以弥补增加的偏航风险，它必须拥有大量有关其自动驾驶汽车的详细信息。环境。 由于所需的详细程度或车载传感器的局限性，实际上几乎无法获得这些信息。 这就必然要求自动驾驶汽车转弯的可能性极小。
前进的道路 (The Path Forward)
There are two kinds of moral questions about a decision — one about which outcome is good and which bad, and the other, ‘less fundamental’, kind, which concern how a deciding agent should think about risk. Instead of getting bogged down in the moral implications of deciding who should live, which decision should a self driving car make when faced with an accident, we should focus more on how to reduce the risk, i.e., which action will cause the least risk overall, without bringing into consideration the type of people on each side of the road. While this outlook is not an instant fix, I believe it is a step in the right direction while bringing autonomous vehicles onto roads, and the only morally responsible, unbiased path forward.
有关决策的道德问题有两种，一种是关于哪种结果是好的，哪些是不好的，另一种是“基本程度较差”的问题，它涉及决策者应如何考虑风险。 与其沉迷于决定谁应该住，在遇到事故时自动驾驶汽车应该做出哪个决定的道德含义，我们应该更多地集中在如何降低风险上，即哪种行动将使总体风险最小。 ，而不考虑道路两旁的人群类型。 尽管这种前景并非立竿见影，但我相信这是朝着正确方向迈出的一步，同时将自动驾驶汽车推上了道路，这是唯一在道德上负责，无偏见的道路。
翻译自: https://medium.com/swlh/ethical-ai-in-self-driving-cars-b96ad2b9a73
人工智能与自动驾驶汽车