This OP is inspired by another thread started by substantivalism in which I considered the potential for humans being supplanted by ethical androids (i.e. androids with substantial moral beliefs): there is already enough anxiety around having sentient robots walking around making choices, but to attempt to develop these intelligences to make moral decisions in place of humans - even with a human in the loop - has even more baggage. And the easiest way of making it work gives rise to some serious moral concerns, regardless of how sophisticated our morality gets.

Imagine we have an android programmed to adhere to a set of morals without exception, and, against all odds, it is indeed capable of executing its limited, semi-permanent programming without running into any moral quandaries or acting in erratic or unpredictable ways when presented with difficult or novel situations. It remains that the morality of this android could always be wrong; a lack of internal contradictions says little about whether or not the general beliefs, although actionable, are correct. This is to say that, in a practical way, the android might be programmed effectively, but there are greater moral obligations than just avoiding contradiction with limited programming.

That is the best situation. A more realistic one is that the android inevitably gets hung up on something and either cannot decide what to do or acts counterintuitively, poorly, or unpredictably. The ideal case and the more realistic one both indicate that some sort of protocol that changes the programming in response to stimuli is necessary.

Thus, it appears we would need to program significant self-reflection protocols into any such android so it could adapt to a changing moral landscape. But can you program truly moral self-reflection? Can a robot learn to be empathetic or compassionate (which I would argue are the two most important markers of morality)?

Alternatively, and more feasibly, we might constantly program and reprogram these ethical androids. We might even wipe their memories over and over again if these memories would interfere with them executing their tasks in such a way that it is safe for them to continually exist - or even replace these androids if that wouldn’t work. If this sounds like a reasonable avenue of action, allow me to pose the following question:

Why not do it with humans? Why not just set up re-education camps to make people have the "correct" moral beliefs? Why not partially wipe someone’s memories if one of their loved ones dies and it interferes with their work at the construction site manning the crane? Why not get rid of someone if their internal mechanisms fail and they can no longer execute tasks that promote the wellbeing of the public? What if they are a danger to the public? To commit to this method is to commit to the demise of any intuitive notion of human morality. And since these actions are now on the table, all any of them would need in terms of justification is that they produce more good than harm.

So, if we want truly ethical androids, we might have to commit to giving them status equivalent to humans, or we face a horrific double standard. But, if we do that, we risk quite a bit if we cannot produce a mode of self-reflection for them. Thus, I think the best option is to just not create truly ethical androids, but maybe someone can show that I'm wrong.