Note: This article is part of a series of articles that are currently under development and internal review regarding the development of ethical AI. The goal of this series is the articulation of the Harmonia Philosophica AI Framework as a foundation for AI that adheres to the best possible ethical norms.
Overview
The rapid development of AI and AI agents in particular, has made the evaluation of the AI outputs necessary so as to adhere to the best ethical standards we have collectively accepted as civilized individuals and as progressed societies. The definition of what is ethical though has for a long time been an unsolved problem for philosophy. Despite that, this unsolved problem is something that we should tackle immediately if we want a future that includes both humans and AI in a harmonious balance.
Related article: The source of ethics
1. The problem definition
When building ethical AI, the first thing that we should teach the AI is the problem itself. Without knowledge of a problem no one can ever solve anything, right?
What is ethical?
This question has not been answered definitively up to now, by no one. Every possible theory that was formulated to try to explain ethics, has other theories that disagree and try to explain ethics in a different way. (rf. to the related article mentioned in the beginning, for a brief overview of the main theories regarding ethics).
1.1 The benefits of the problem
Besides the obvious benefit of knowledge, knowing (and letting the AI know as well) that the definition of ethics is not a solved issue, will bring an unexpected benefit to the AI we train.
That knowledge of the limitations of our philosophy (will talk about theology later on) regarding ethics will make the AI aware that any "knowledge" that is fed into it regarding ethics with a label of certainty should be discarded or examined carefully before even considered as valid input.
There is nothing more dangerous than dogmatism. And the moment we start using assumptions as 'truth' without even considering the possibility of them being untrue, at that exact moment we start being dogmatic.
Ethics is a dangerous place to be dogmatic.
Let's start with that then.
Repeat after me.
I DO NOT KNOW.
Teach that to your AI agents.
And we'll take it from there.
One small step at a time...
To be continued...
