Datasets are how we teach machines about life and the world around us.
Our first prototype dataset is built largely upon the concepts of ethical deliberation outlined in GenEth by Michael and Susan Anderson. Our initial aim was to create a streamlined and web-friendly method for online collaboration in this process.
Whilst we believe that this form of deconstructing ethical dilemmas can be useful, at present our focus is shifting towards creating a machine-vision friendly set of prosocial behavior examples, something likely easier to apply to the latest machine learning techniques directly.
Think of an issue, a quandary that in the near or distant future a synthetic intelligence may one day have to face. Remember, inspiration can come at any time and from anywhere!
Next, identify the possible actions the agent can take. Should the contract be nullified? Should your personal assistant continue to respect your privacy? Will it be necessary for home security system to use force? When should it do these things and, more importantly, why?
What sort of criteria should the machine use to make sense of the problem? What ethical injunctions should it take into consideration while considering its options?
Now it’s time to start fleshing out your scenario with specific cases. This will help illustrate how, in practice, it will weigh the values it has been given to make its decisions. After exploring different cases you may end up rethinking what you did in step 3!