Process Dependence

Overview

Many theories associate consciousness with processes unfolding over time, such as global broadcasting, recurrent processing or the execution of an algorithm. But this creates a potential tension with fundamental physics, whose equations are generally Markovian: the future evolution of a system depends on its complete present physical state and subsequent inputs, rather than independently on its history.

Consider two computers in precisely the same present physical state. One arrived there by executing an algorithm proposed to generate consciousness; the other was assembled directly into that state without executing the algorithm. A process-dependent theory attributes consciousness differently to these systems because their histories differ. Yet under standard physical laws, their subsequent behaviour must be identical, given identical future inputs.

This creates a dilemma. If the difference in consciousness is causally efficacious, affecting subsequent physical events, then the system's evolution must depend on something beyond its complete present physical state. This requires a different kind of physical law, with irreducible dependence on history. Such laws are not logically impossible, but depart from the state-sufficient dynamics of established fundamental physics and require justification. Why is there one type of law for consciousness and another for everything else in physics?

Alternatively, if the historical difference produces no further physical effects, the proposed difference in consciousness cannot be directly tested through subsequent behaviour. Why should we prefer the process-dependent account over one in which consciousness depends only on the present physical state?

The challenge is not to the importance of processes in producing conscious states, but to the claim that undergoing a particular past process remains an independent, causally relevant condition for consciousness once the present physical state is fixed.

Responses

But Markovian systems can perform complex computations and retain memories of past events.
AGREED, BUT: The argument does not claim that computation requires non-Markovian physics. Present physical states can encode information about the past and perform sophisticated computations using that information. The issue is whether consciousness can depend on actual past events beyond the information and causal consequences retained in the present state.
But contemporary physics does contain non-Markovian equations - perhaps the functions of consciousness are related to those. For instance, equations governing viscoelastic materials make references to past events over various historic intervals, and collision integrals introduce memory kernels in the Boltzmann equation.
BUT: All non-Markovian equations in contemporary physics are non-fundamental. They represent simplifications of fundamental equations, such as Maxwell's equations for electromagnetism, Einstein's equations for relativity, and the broader framework of classical field theory in classical physics, or the Klein-Gordon and Dirac equations of Quantum Field Theory. All the fundamental theories are Markovian, and science treats these fundamental theories as a deeper truth closer to the ontological reality.
Non-Markovian equations emerge only as coarse-grained simplifications. Such references to past events are only an 'epistemic' form of memory, where we, as modelers, need to keep track of history because we've averaged out microstates. There is no true ontological memory defined in the equations of contemporary physics, where the universe itself remembers past states. If consciousness is defined via epistemic models, i.e. levels of description relative to some modeler, the theory needs to be made explicitly relational and address concerns about the arbitrariness of choosing a model/modeler or different mutual-modeling possibilities. Perhaps all possible models simultaneously exist, but this risks identifying consciousness in very many systems.
NB. Some interpretations of wave function collapse could constitute a counter-example, but it is unclear how contemporary functionalist theories could exploit this.
The present physical state may contain everything about the past that matters for consciousness.
YES, BUT: This places a strict constraint on the types of temporal history that functionalists can rely on and types of causal mechanism they can invoke. Systems' present-states do provide information about past trajectories but typically a wide range of past trajectories could lead to the same present-state. If the distinction is irrelevant, the original process is not independently necessary for consciousness. If the distinction is relevant, the present-state cannot know which of those past trajectories created it, at least not without solutions like a block universe or a metaphysical observer.
Alternatively, functionalists would need to show that all possible equivalent past trajectories would preserve the motivating intuitive causal mechanism for their theory (e.g. global broadcast, meta-representation etc.). Or they might argue that what actually happened does not matter, provided the system interprets its present-state as implying the relevant history. But this requires rejecting the intuitive causal mechanisms with their powerful counterfactuals. It would be enough to falsely believe a mechanism occurred, rather than it actually having occurred.
NB. If we assume deterministic and time-reversible fundamental laws and allow access to the complete physical microstate, then the present state would, in principle, fix a unique past history. But this relies on substantive interpretive commitments — e.g., rejecting fundamental wavefunction collapse — and in any case no actual subsystem (including any conscious system) has access to such a state, since it is itself only a tiny part of it. So uniqueness of the past is a feature of a God's-eye theoretical description rather than of any physically or cognitively available perspective.
Consciousness might be an extended process rather than an instantaneous state, obeying physical laws that differ from currently established physics.
POSSIBLY, BUT: Such laws should be motivated and held to the same high standard as existing accounts in physics, rather than simply posited as possibilities that cannot be discounted through logic alone. Indeed, we know contemporary physics is incomplete, e.g. there is no established reconciliation of quantum physics with gravity. Perhaps the true underlying theory, if we ever get there, will be shown to be non-Markovian in nature, providing a realist underpinning to functionalist theories of consciousness - but theorists pursuing this route should acknowledge the grandeur of their assumptions.
Fundamental Markovian equations can be rewritten in a non-Markovian format with no loss of predictive power. Perhaps the non-Markovian formats are actually more fundamental? In general, given any non-Markovian dynamics x(t) depending on history, we can formally define new variables (x, ẋ, ẍ, ... or integral expressions over the past) until we get a first-order system. The shift to Markovianity starts to look like a bookkeeping choice.
PERHAPS, BUT: The preference for Markovian formats is not arbitrary. The added variables in successful physical theories are not generic memory registers but correspond to independently measurable quantities, transform naturally under the theory's symmetry group, and support a local initial-value formulation: these strongly constrain the space of viable options.
One example is the equal footing of position and momentum in the uncertainty theorem in quantum mechanics: neither can be eliminated without loss of structure, and they are related by a duality (Fourier transform) rather than a simple derivation. State-space enlargement can reflect genuine physical structure rather than representational excess.
The Markovian formulation is often the unique fixed point of deep structural constraints (locality, symmetry, variational principles, well-posedness), whereas non-Markovian formulations form a large equivalence class with no canonical internal ordering (you can typically generate other non-Markovian variants by removing different subsets of state-space variables in different orders). That asymmetry gives a non-aesthetic reason to prefer the Markovian formulation even when non-Markovian ones are formally equivalent. Non-Markovianity also brings other worries: temporal non-locality (physicists tend to require particularly compelling motivations to introduce non-localities, e.g. given difficulties reconciling with relativity) and infinite regress needed to determine future trajectories (where is this information stored?).
It might be possible to restructure contemporary physics in non-Markovian formats (and that might ultimately prove a deeper truth, as Nancy Cartwright argued when rejecting simple universal physical laws in The Dappled World, 1999). But this is not a simple motivation - and goes counter to the historical trend where originally non-Markovian formulations have been replaced over time. Functionalists could pursue this path but should be explicit in what type of non-Markovianity meets their needs and why it is better motivated than current Markovian equations.

Overview

Responses

Further reading