Picture this: There is a trolley coming towards five stuck on the tracks. You are in the position where you can easily reach the lever to reroute the trolley to the sidetrack and save their lives. But: in doing so, you will kill one person that is, by ill luck, stuck on the sidetracks. What do you do? In nowadays world, Artificial Intelligence is faced with the same problem when dealing with autonomous driving for instance. How ethical is it to let AI decide on human survival?
Picture the following situation: There is a trolley coming towards five tied-up people who are stuck on the tracks. You are standing in a place where you can easily reach the lever to change the direction of the trolley and reroute it to the sidetrack. If you do so, you will save the lives of five people. However, there is one person on the sidetrack who is also tied-up and can’t move. By redirecting the trolley, you will have to deal with killing that individual. What do you choose to do?
Now picture the same situation with an AI in the driver's seat: How would their decision look like? And how ethical is Artificial Intelligence?
If you are familiar with philosophical theories, you can probably recall this thought experiment which has recently come back to the public interest with the rise of AI, especially in connection with autonomous driving. Why would you go back into analyzing this one more time if it has been covered by so many authors in ethics and philosophy?
The reality is, that this experiment is not as abstract as we think it is. There have been many real-life examples when a person had to decide on a similar outcome with human lives at threat - and there was no possibility to save everyone involved. When the decision is made by a human, it is based on the quick assessment of the situation but is also based on emotion, level of stress, and moral judgment. One may argue that is not a decision but more precisely a reaction to the given situation. But how to judge which person should be saved during an accident caused by an autonomous car when the car itself lacks the moral judgment of the human being, basing that decision on a computational algorithm?
How ethical can Artificial Intelligence be?
Our last wattx insight session was revolving around the consciousness of AI. Tobias, a wattx Business Development intern and Psychology Master student, has presented the differences between human and artificial processing. As humans, we learn from experience and through the comparison of past reactions and past consequences. We compare the way we reacted with the consequences we got.
This is very similar to artificial processing, where we enter data points and answers to a specific subject into the machine which analyzes them and makes the next decision based on that learning. Coming back to our Trolley Experiment and autonomous driving example, we can simply narrow down the problem to leaving the decision to AI to trust: Should we trust enough that the algorithm will be fair to everyone? Or should we trust that it will be fair towards us in particular? How can we make sure that the AI makes an ethical decision?
The supporters of AI argue that it is designed in a way that creates a safer environment by eliminating human mistakes. However, we are not able to get rid of random accidents which will put the machine in charge of a decision. Imagine you are in a car that is involved in an accident. You are in no control of the car, but the car’s algorithm itself decides whether to save your life or the life of the people on the next line. On what facts and data will that decision be based? Are you going to be saved because it is your car, or maybe the information about you and the other passengers will be processed in a way to make that decision? How do we avoid discrimination in setting up this algorithm? What decision criteria should decisions be based on? And most importantly; who is supposed to make that final decision - you, the scientist, or the government?
So far, we have only focused on autonomous cars. But there are far more use cases and therefore potential threats than only that in the topic of AI. It comes across almost every aspect of human life - safety, risk of malicious use, issue of automation and job loss, civil rights, its impact on human interactions with each other, etc. It is quite certain that in a short time, AI will be used to improve our lives in almost every aspect we can think of. That’s why it’s even more important to talk about its safety. We need to simply make sure that AI is not allowing or causing any harm.
Another aspect to consider is human adaptation. It is part of our nature and comes with years of evolution. We simply know how to adjust to our environment. So we can also think about AI as something that evolves and most likely won’t stop evolving. In that scenario, humans will learn how to live around and with AI applications the same way they learned how to live around any trains, vacuums, or live according to rules of the law in the country they choose to live in. The only concern is to find that silver lining.