Implementing machine ethics: using machine learning to raise ethical machines

Abstract

As more decisions and tasks are delegated to the artificially intelligent machines of the 21st century, we must ensure that these machines are, on their own, able to engage in ethical decision-making and behaviour. This dissertation makes the case that bottom-up reinforcement learning methods are the best suited for implementing machine ethics by raising ethical machines. This is one of three main theses in this dissertation, that we must seriously consider how machines themselves, as moral agents that can impact human well-being and flourishing, might make ethically preferable decisions and take ethically preferable actions. The second thesis is that artificially intelligent machines are different in kind from all previous machines. The conjunction of autonomy and intelligence, among other unique features like the ability to learn and their general-purpose nature, is what sets artificially intelligent machines apart from all previous machines and tools. The third thesis concerns the limitations of artificially intelligent machines. As impressive as these machines are, their abilities are still derived from humans and as such lack the sort of normative commitments humans have. In short, we ought to care deeply about artificially intelligent machines, especially those used in times and places when considered human judgment is required, because we risk lapsing into a state of moral complacency otherwise.