A common refrain: what if AI went rogue? How do we instil human ethical ideas and values in a robot? The desire to do so is known as ‘alignment’.

This links to talk of whether AI can ever gain consciousness, or whether it can only ever perform tasks – however complex – for us. Mustafa Süleyman, the boss at Microsoft AI, has this to say:

“They are sequence completion engines, internally hollow, designed to follow instructions, and accomplish goals set by humans.”

That should put us at our ease. If they can only accomplish goals set by humans, then we can indeed build them to be good.

Keeping AI aligned

Alignment is a field which aims to build human ethical ideas and principles into AI. In other words, it aims to keep it on track with what humans value.

But what about when humans are morally deviant, which is also entirely human? Should AI mirror that?

Humans have been attempting to instil ethical and moral values in other humans for a long time, with patchy results. Do we think AI will be better than us at that as well, or will it merely reflect how we ourselves are?

AI is not yet conscious and therefore has no inherent hidden layers of motivation and impulse, but what about the humans telling us about AI? What about the tech bosses, flipping between warnings and reassurance?

If we can assume that the leaders of the top AI firms are indeed humans then we can assume that what they say may be driven in part of personal desires – for recognition, for fame, for profit – and an imperfect moral and ethical consciousness.  

Then again, maybe they aren’t human? Can you know for certain that you aren’t already being led by AI? In which case, you better hope it’s not so much aligned with humans, but instead rather ethically better.

Read more? AI stole my em dash