Well, to be fair, it probably isn’t, but, on the strength of this post over at Larval Subjects, Bryant might just believe that it is. The idea seems to be that representation depends on dynamic and fluid interactions between objects, thus either representation is not what we thought it was (reductionism) or there ain’t such a thing (eliminativism).
Here’s a rough attempt at formalization.
Assumption: Representational/semantic (RS) properties are static.
1) Agency properties (the behaviour of agents) are non-static (i.e. dynamic, fluid, etc.).
2) RS properties supervene (depend on) on Agency properties.
3) All supervenient properties have the same higher-order properties as their subvenient properties.
Conclusion: By (3) RS properties are non-static (contrary to the assumption)
However, this is an unsound argument because 3) is patently false. Supervenient properties don’t get all their higher order properties from their base of subvenient properties. Aesthetic properties plausibly supervene on physical properties (if two things are physically identical, they are aesthetically identical) but physical properties are quantifiable whereas aesthetic properties are not.
So for the argument to work we need to assert either identity between dynamic agency properties and representational/semantic ones (reductionism) at 3, so we can get to the conclusion via the indiscernability of identicals, or eliminativism (there are no RS properties).
So if this argument supports OOO. OOO is committed to reductionism or eliminativism.
To put this argument into context of Levi’s homeostat example:
I’m not contesting the OOO claim regarding the epistemic impenetrability of objects or its claim regarding the non-representational character of our access to them.
However, the considerations adduced in Levi’s post here only establish that our access to objects is non-representational if we make extremely deflationary assumptions about the relationship between knowledge and the dynamic processes in cybernetic systems since the mere dependence of representation on dynamics does not suffice to press his claim.
For example, even if all states of an information processing system S are responsive to changing outputs of objects in S’s world, it doesn’t follow that some of those states are not also responsive to internal states of those objects. S might be armed by a hypothesis-forming device: say, a feedforward neural network with the input layer corresponding to the sensory input from the object’s outputs, while the ‘hidden’ layer might flip outputs into one state if the object being tracked is moving in a phototropic way and into another if it is behaving photophobically. If these behaviors are caused by internal states of the object, then the S could track persistent and causally determinative internal states of the object. In terms of the state space of the system the phototropic/photophobic difference would correspond to a partition of that total space by the hidden layer.
If the hidden layer state merely replicates the dynamically changing input or responds randomly (as would be the case in a network prior to training) then this presumably won’t be the case.
So if we identify knowledge states with fluidly changing states recording the passing scene, we get the reductive result that all we can know is the passing scene. We could also get to a similar position if we simply reject the claim that some objects – like homeostats – have internal states with causal roles (input-output conditions). I suppose OOO fans have to commit to some such but this doesn’t follow from anything known about cybernetic systems unless this knowledge excludes the possible of hypothesis-generating mechanisms and merely considers the raw input from sensory transducers – which is not the case.