If you read the Microsoft AI Code of Conduct you might think it is unlikely. It’s an interesting document, with Mustafa Suleyman’s hand prints all over it, clearly positioning Microsoft against Anthropic’s model welfare orientated approach to ethics. This is the point which stood out to me most of all:
Our AI should make you sharper, not dependent.
“MAI Models should discourage patterns of interaction that cause excessive reliance or emotional dependence.”
“Users should retain ownership of their goals, choices, actions, and decision
The problem is that if enterprise AI is deployed to reduce headcount, it’s pretty inevitable that this principle will be violated. The problem is the organisational context of implementation rather than the technology being implemented. If you read Microsoft’s advertising materials for Copilot, the language of productivity features constantly: doing more with less, working more quickly, getting more done. There is nothing inherent to this which means headcount will be reduced. It is indeed possible to introduce this in order to free people up for ‘high value work’.
But in a financially constrained context what justifies the investment necessary to implement these systems? Training, governance and security are all resource intensive undertakings which risk matching whatever near term productivity gains can be realised. If the costs of these systems will rise with time (and I honestly struggle to see any scenario in which they don’t*) then it increases the pressure to realise gains elsewhere. The nature of the productivity savings themselves (chaotically realised and heavily individualised) also lends itself to a reorganisation of work in which coping with the epistemic fallout of enterprise AI requires restructuring work around checking and filtering workslop.
If you look at actually existing organisations in the actual conditions of the late 2020s the commercial appeal of these platforms is going to be about reducing headcount, as well as reorganising labour to make that sustained headcount sustainable given the affordances and constraints of enterprise AI. In these circumstances it is extremely likely that enterprise AI will erode professional autonomy even if Microsoft don’t want it to.
*Even if there was a sustained fall in inference costs, the degree of lock-in from enterprise governance means switching costs will remain high.
