The #FoundationModels of AI (Transformer, LLM, call it whatever) are troublesome, not because they are deficient, but because they are endowed with #emergent behaviours that suddenly spring from 500-billion-plus trainable parameters and terabytes upon terabytes of human behavioural data. No human could grasp the aggregate meaning of that unanticipated, implicit, emergent behaviour. No human could divine the outcomes of their unforeseeable concoctions. Worse of all, all these models are inhered with biases and prejudices, both intentional and unintentional, that are embedded deep within these deep-learning models, whose untoward conducts are ever so subtle and complex as to be casually observable.
The invidious effects of these poorly-understood foundation models are insidious. By the time their detrimental behaviour becomes observable, it would be too late for society to countermand, for these models are already enmeshed within the social-political-economic fabric of the whole of humanity.
It is easy, for a lawyer like me, reflexively to shout, "#Regulate #AI". But in the present heady mix of delirium, confusion, delusion, and collusion, regulate exactly what and precisely how, pray tell.
Who will save society from AI?
Well, perhaps AI might....🤦♂️
Matrix multiplication is not inhered with empathy, conscience, judgement, elation, shame, and the like. Those are innately human characteristics, neither mechanical nor mathematical. Let us not surrender to the silicon-based machines the #ethical duty to be thinking, feeling carbon-based beings.