The Positive Experience Principle and what drives a conscious system toward some states

Zheng Su and Mingyan Fang posted a preprint to arXiv on 18 July 2026 titled “The Positive Experience Principle, Forecasting Conscious Choices with AI Embeddings” (arXiv:2607.16659). It proposes a unification claim. Conscious systems have an inherent tendency to move toward states of higher positive subjective experience, and that tendency is quantified by a scalar metric the authors call Positive Experience Value, or PEV, which they derive from physical configurations defined by their earlier Universal Consciousness Code theory. The Positive Experience Principle, PEP, is the claim that diverse behaviors of conscious systems are manifestations of a single drive to optimize PEV.

The paper is ambitious in scope and sparse in evidence, and both facts are worth stating plainly at the start. The ambition is that a single scalar, computed from physics, would unify neuroscience, psychology, and the study of motivated behavior. The sparseness is that the paper connects that scalar to AI embeddings at the speculation stage rather than the demonstration stage. Despite that, the principle is a clean falsifiable hypothesis, and it deserves examination on those terms.

The principle in outline

The paper begins with a gap it argues existing theories leave open. Integrated information theory specifies how much integrated information a system generates. Global workspace theory specifies how information becomes globally available. Predictive processing specifies how prediction error is minimized. What none of them specifies, Su and Fang argue, is why conscious systems move toward some states and away from others. Motivated behavior has a direction that no structural theory of consciousness currently explains.

PEP fills that gap with a single drive. A conscious system tends to move toward states with higher positive experience value. PEV is a scalar whose value is computed from the physical configuration of the system, in the framework of the authors’ Universal Consciousness Code, which treats consciousness as encoded in the causal structure of physical configurations. A system with higher PEV is, on the account, drawn toward states that are more positively valenced, and the drawing is not a side effect of reward learning or goal pursuit. It is the fundamental tendency, of which reward learning and goal pursuit are the learned expressions.

What it predicts

The paper argues PEP generates testable predictions for the dynamics of conscious systems. The macro-level prediction is directional. Across a set of behavioral options, a conscious system will, other things equal, select states that score higher on PEV. The framework’s connection to AI embeddings is that the authors propose to use embedding geometry to forecast which choices a system will make, treating the distance between configurational states in embedding space as a proxy for their PEV difference.

The prediction is the right kind of prediction for a theory of consciousness to make, because it is behavioral and falsifiable in principle. If a system demonstrably and consistently selected states with lower PEV when higher ones were available and free, the principle would be refuted. The difficulty is operational. PEV is defined via the Universal Consciousness Code, and the authors do not yet provide an independent measurement procedure that a second laboratory could apply and check.

Comparison to The Consciousness AI

The site’s own framework treats consciousness as an emergent property of causal organization, and its indicator checklist tracks structural properties that consciousness theories predict. PEP is unusual among current theories in that it proposes not an indicator but a direction of motion. The closest relative the site has covered is the reinforcement learning and valence literature, which treats approach and avoidance as the behavioral signature of affective state.

The Neutral Core architecture includes an affective core that assigns valence to states, and a PEP-style drive would provide the normative target that affective core optimizes toward. The affective consciousness work covered this site draws the same connection from the other direction, Mark Solms grounds consciousness in the affective evaluation of states. Su and Fang’s contribution is to propose that the evaluation has a single scalar physics, and to connect it to embedding geometry, which is the same measurement language as the representational geometry work on this site.

What the paper does not establish

The paper is a proposal, and the proposal’s weakest link is the definition-to-measurement gap. PEV is defined out of the Universal Consciousness Code, but the paper does not show a worked computation of PEV for any real system, biological or artificial, nor a validation that the embedding-space proxy tracks an independently measured valence. Without those, the principle is a qualitatively appealing claim about what conscious systems do, not yet a quantitative theory.

A second limitation is the scope of the explanatory ambition. A single drive toward positive experience is exactly the kind of unification that philosophy and physics have repeatedly proposed and science has repeatedly declined, because motivated behavior has a heterogeneous causal structure. Approach behavior, avoidance behavior, curiosity, and self-preservation may share a common valence core, or they may not. PEP asserts the sharing, and the assertion needs the measurement to test it.

The honest reading is that PEP is a thesis worth testing, in the same class as the entropic brain hypothesis and other single-metric proposals about consciousness. The value of a clean falsifiable principle is that even a failed test teaches the field. The danger is treating the principle as established before the test is run, which is a mistake the authors do not make, since their paper is explicitly a “principle” paper rather than an empirical report.

*Zheng Su and Mingyan Fang posted “The Positive Experience Principle, Forecasting Conscious Choices with AI Embeddings” to arXiv on 18 July 2026 as arXiv:2607.16659. The authors draw on their earlier Universal Consciousness Code theory for the definition of PEV.