This is the logical conclusion of software's cognitive arrogance. Software developers tend to believe they know better than anyone else, and that is the reason so much friction exists in the surface between developers and domain practitioners. All the power struggle played out in the psychodrama of agile, and the apparent inevitability of the big ball of mud curse: it all stems from this cognitive arrogance.
And AI is the ultimate smoke screen. Calling software 'intelligent' is the ultimate expression of this arrogance... it somehow got credible again, after AlphaGO, and received a huge boost with LLMs apparently solving NLP.
Software can never be better than the domain knowledge developers are able to imprint in it. Trying to avoid learning about any domains by brute forcing statistical games and claiming that the resulting software is somehow 'intelligent' is a massive overplay of the hand we've been dealt. It will end up the same way as before: with massive backlash and decades of the label 'AI' becoming toxic again.
What stands out to me is the example of the State of Michigan using AI to detect unemployment fraud. The problem was that 93% of the fraud was completely wrong and ruined people's lives with criminal convictions.
And AI is the ultimate smoke screen. Calling software 'intelligent' is the ultimate expression of this arrogance... it somehow got credible again, after AlphaGO, and received a huge boost with LLMs apparently solving NLP.
Software can never be better than the domain knowledge developers are able to imprint in it. Trying to avoid learning about any domains by brute forcing statistical games and claiming that the resulting software is somehow 'intelligent' is a massive overplay of the hand we've been dealt. It will end up the same way as before: with massive backlash and decades of the label 'AI' becoming toxic again.