Both models face the problem that there appears to be an epistemic gap between having an intention and performing a certain intentional action. In each case, we have no difficulty with the idea that agents know their intentions without observation (by introspection), but the conclusion falls short of (AK), the claim that agents know their intentional actions without observation. (p.13)
[T]he gap might consist in the fact that the intention itself is conceived of as something that falls short of the whole intentional action. Here the intention would stand to an action as a sense-datum does to an object of perception. (p.13)
I have suggested that we could learn to live with a metaphysical gap between intentions and intentional actions by adopting a more fallibilist conception of agent’s knowledge. […] [A]wareness of one’s plan provide fallible evidence for thinking that one is acting as planned. (p.19)
Newstead, A. (2006) “Knowledge by Intention? On the Possibility of Agent’s Knowledge”
Anscombe believes that we, as agents, have non-observational knowledge of what we are doing. And such practical knowledge (or: agent’s knowledge, in Newstead’s terminology) is not just our knowledge of our own mental state, i.e., our intentions. It has long been debated how such kind of non-observational knowledge about our intentional actions is possible.
Newstead appears to think that the crux of the problem lies in the infallibilist conception of evidence. She seems to think that Anscombe et al. feel that there is a difficulty in making sense of practical knowledge only because our knowledge of our intention does not necessitate any action. (That is, a skeptical scenario is always possible.)
This diagnosis of the debate is mistaken, I think. The Anscombian worry is not about fallibility, which she embraces. The issue is about the non-observational nature of practical knowledge. That we know about p based on certain mental state m (that we have introspective access to) as defeasible evidence does not make our knowledge about p non-observational (otherwise our perceptual knowledge is non-observational).
It is also worth pointing out that our intention has no assertive force (unlike perception), so it is not straightforward that introspective knowledge about our own intentions can be evidence for anything unless we also know that our intentions normally produce corresponding actions. But then, first of all, it is unclear that we know the latter non-observationally. Secondly, even if that is non-observational, it is a predictive knowledge that is qualitatively different from my non-observational knowledge of my current action.