According to the BSI, the best way to make sure people are accountable for what their robots do is to make sure AI design is transparent. That might be a lot harder than it sounds, though. Even if the code governing robots is freely accessible, that doesn't guarantee we can ever know why they do what they do.
In the case of neural networks, the outputs and decisions are the product of deep learning. There's nothing in the network you can point to that governs a certain outcome like you can with programmatic code. If a deep learning AI used in law enforcement started displaying racist behavior, it might not be a easy to figure out why. You'd just have to retrain it.
Going beyond the design of AI, the BSI report speculates on larger ideas like forming emotional bonds with robots. Is it okay to love a robot? There's no good answer to that one, but it's definitely going to be an issue we face. And what should happen if we become too dependent on AI? The BSI urges AI designers not to cut humans out altogether. If we come to rely on AI to get a job done, we might not notice when its behavior or priorities start delivering sub-optimal results -- or when it starts stockpiling weapons to exterminate humanity.
Now read: IBM’s resistive computing could massively accelerate AI — and get us closer to Asimov’s Positronic Brain
