Is AI Ethics Special?

If you want to look at questions in ethics, you need to look closely at the world. Although ethics may refer to abstract values or principles, it’s essentially an applied and practical subject. It needs to make a difference to the world. You need to understand the subject matter in hand. There are currently many claims that AI is presenting us with unique, or particularly urgent, ethical questions, including the claim that owing to AI, we face an existential threat. Our project has been aiming to understand, as closely as possible, what are the particular ethical issues that AI presents.

Initial attempts to focus on this question were presented at IJCAI-16 in New York in a paper entitle The Distinctiveness of AI Ethics, and Implications for Ethical Codes, published here as a blog post on this site.

We argue that a good approach to this question is to consider that there are various ethical challenges which are distinctive to AI, even if not necessarily unique. It’s also vital to remember that one feature of AI is, indeed, its very broad scope; so that different areas of AI will produce their own ethical questions, which may be very different to other areas. For instance, interactions between robots and humans may produce particular challenges, the use of autonomous trading agents give rise to quite other questions, and likewise for the use of AI-assisted medical diagnosis, and so on.

And there are other issues in AI which are shared with other rapidly developing technology. Here are some examples:

For instance … the very fact that AI is a rapidly developing technology means that it’s hard to predict what will occur even in the near future, and hard to anticipate ethical questions in advance, and to produce codes and regulation as fast as the AI is developing. But this is the case in other areas of science and technology as well.

Some issues especially in particular areas of AI concern the use and manipulation of masses of data – Big Data. But there has also been work for some time now on ethics of data management in, for example, genomics research, where the fact that humans each have 3 billion points of data on their DNA which can be analysed and shared pretty cheaply, has given genomics researchers many ethical challenges.

AI often involves highly complex systems and the effects of the operation of AI may be some distance from producers – as when machine learning produces algorithms which have (sometimes hidden) effects some way off. Among other issues, this has led to questions about how responsibility works in such large networks. But precisely this question is also already being asked of the large networks of scientists collaborating around the globe on projects involving sensitive data about humans.

Nonetheless, there are some characteristic challenges in AI. As well as the issues already mentioned, these include:

The characteristic hype about AI, which can bring with it some difficulties, as we discuss here, together with some brief suggestions about how to combat this.

The ways in which the use of AI interfaces with wider social and human factors. The study of the social and cultural impacts of technology is a well-established area of research. One lesson we need to bear in mind in considering ethical issues, is just how subtle, yet pervasive, social, cultural and economic impacts there may be from the use of AI. And how hard it may be to uncover these.

Professional ethics typically assumes that professionals know and can control the products or services they provide. A fuller discussion of professional ethics, with an account of the distinctive problems posed for AI, can be found here.

AI’s typical use, which is to enhance, supplement, or replace the work of humans. This then tends to give rise to questions which we see of particular significance to ethics, and to our values more broadly. One issue of particular interest is how in thinking about this, we have a tendency to idealise both the agency of humans, and the agency of machines.




We would like to thank the Future of Life Institute for their generous sponsorship of our programme of research.