But a lot of useful Science can be done by observing, documenting, cataloging and explaining the effects of AI being already inflicted on us.
I think if there were more transparency concerning these systems and their affiliated organizations, that would obviate the need to add another layer of reasoning, which would reduce the number of unknowns, rather than increase.
Of course, that requires cooperation and trust. Which brings one back to the question of intents. Which can conflate intents meant to 'inflict' (harm?) or intents meant to grow, develop, inspire, progress.
I think this is less about AGI/ full AI than it is about learning systems. https://en.wikipedia.org/wiki/Artificial_general_intelligenc...
That said, the impact of this stuff especially in the context of the web is both profound and evolving.
Psychology and sociology are 'soft' sciences because you can't test people psychologically and sociologically in 'hard' ways without seriously crossing some serious ethical boundaries.
It could be argued that a lot of the decisions being made that back the design in contemporary infrastructures already are engaged in this sort of testing and behavioral modification, in ways that seem innocent from the onset, but yet still propagate across social systems in ways that are far too chaotic and complex to forecast.