Why It Must Feel Like Something
The Motivational and Computational Necessity of Phenomenal Experience
If a system has to weigh hunger against loyalty against fear and then actually move, what would the deciding state have to be — and does it have to be felt?
Pace, J. C. (2026). The Language of Stress: Why It Must Feel Like Something. FigShare. DOI: 10.6084/m9.figshare.31951794
Overview
An organism facing incommensurable demands has two problems, not one. It has to settle which matters most right now, and it has to be moved by the answer. A system that ranks perfectly and is unmoved by its own ranking is not a functioning organism; it is a very sophisticated observer.
This essay develops three lines of argument about what the state doing that work would have to be. Two concern architecture. The motivational argument extends a regress: where a system deliberates, a ranking is a representational state, something has to act on it, and that something has to be triggered by something else -- so the chain has to terminate in a state that requires no further trigger. A system in which tension resolves directly into action settles by reflex and faces no regress; what it cannot do is hold a demand unmet while it weighs ways of meeting it. The computational argument concerns what kind of system could do this at the speed and scale a living organism requires, across demands that share no common measure. Together these specify a state that is immediate, non-representational, commensurating across domains, and directly coupled to action.
The third argument is different in kind, and the essay is built around keeping that visible. It holds that a state which compels action by its own nature must be presented to the system as compelling, and that such self-presentation is what phenomenal experience is. That step is the framework's load-bearing inference. It is argued rather than demonstrated, a critic can accept the entire architecture and reject it, and the essay says so in the section where the argument is made. An earlier version presented all three lines as independent and jointly establishing the conclusion. They are not independent -- two establish a requirement and the third identifies what fills it -- and that claim is withdrawn.
Along the way the essay concedes what it owes. The primacy of affect over cognition is Panksepp's claim, argued from comparative neuroscience across a career. Mark Solms made an identity claim first, grounded it anatomically, and argued it from evidence this framework does not have. Predictive processing accounts for the immediacy and automaticity of motivated behavior without difficulty, so those phenomena are not offered as support. Where a competing account handles something as well, the essay says so rather than claiming the ground.
Version 2.1 sharpens the reply to active inference. Version 2.0 conceded that the motivational argument adds nothing against it. Version 2.1 locates the regress at one point: the precision that sets how decisively scored policies govern. It does that without changing how any policy is evaluated, and where habits enter as a prior over policies it can also change which policy governs. The essay names the empirical condition under which the account fails for human deliberation -- a dopaminergic gain separable from valuation -- and says that a brain implementing selection as competition with no separable gain would be a candidate implementation of the account rather than a rival. Section 04 now describes what the organism does as the resultant of many simultaneous pulls, not a contest won by the largest.
Version 2.2 follows the Canonical Axioms v2.4: the fourth condition for candidacy now includes the capacity to hold demands unmet while ways of meeting them are rehearsed, so a system confined to reflex is no longer a candidate -- a statement about the framework’s grounds for attributing experience, not a claim that such a system lacks it. Since deposit, the essay also says that the settlement of competing demands includes when each is acted on, describes the state that ends the regress as the measure of how much something matters right now, and points to the separate paper in which the regress is stated on its own (The Prioritization Regress, PhilArchive). No axiom has changed.
Version 2.3 states the dual-task prediction in signed form, following The Prioritization Regress and the empirical predictions register (Prediction 11): two strategies rehearsed at once should interfere more when their valence is opposite than when it is the same, at matched complexity, and each evaluation should shift toward the other’s valence. Accounts that evaluate options in a shared affective medium predict this too, so the test shows that deliberation runs on tension without by itself separating the framework from them. The design has not been run. No axiom has changed.
Read this if you want the argument for why experience might not be an accessory to the machinery -- and a clear statement of where that argument can be refused.