The Grand Tour: Herzog, Schurger, and Doerig's Scientific Theories of Consciousness

For decades, consciousness research has resembled an archipelago of isolated theoretical islands. Researchers developing Integrated Information Theory (IIT) rarely engaged deeply with Global Workspace Theory (GWT), and proponents of Higher-Order Thought (HOT) theories often spoke past those working on Predictive Processing. The result was a fragmented field where progress was difficult to measure because each theory used its own metrics and ontology.

The publication of Scientific Theories of Consciousness (Cambridge University Press, 2026) by Michael Herzog, Aaron Schurger, and Adrien Doerig marks a maturation point for the discipline. It is the first comprehensive textbook designed to provide a comparative framework for evaluating all of them, rather than advocate for a single theory. For AI engineers attempting to navigate this landscape to build conscious machines, the book serves as the definitive reference manual for the state of the science.

The Need for a Comparative Framework

Herzog, Schurger, and Doerig argue that the field can no longer progress through theoretical isolation. The Cogitate Consortium’s adversarial testing of IIT and GNW demonstrated the value of forcing theories to make specific, falsifiable predictions in the same experimental paradigm. Scientific Theories of Consciousness extends this adversarial and comparative logic to the entire theoretical landscape.

The book systematically breaks down each major theory (IIT, GWT, HOT, Recurrent Processing Theory, and Predictive Processing) into its core axioms, its proposed neural correlates, its mathematical formalisms (where they exist), and its empirical track record. Crucially, the authors evaluate each theory against a common set of philosophical and empirical challenges: the hard problem, the binding problem, the challenge of measuring consciousness in non-verbal subjects, and the problem of artificial consciousness.

The Consensus on Architecture

While the book highlights the profound disagreements between theories, it also reveals an emerging consensus regarding the structural prerequisites for consciousness.

As synthesized by the authors, no major scientific theory of consciousness posits that a purely feedforward architecture can support phenomenal experience. Whether the theory demands the re-entrant loops of Recurrent Processing Theory, the top-down predictions of Active Inference, the meta-representations of HOT, the broadcast dynamics of GWT, or the maximal irreducible cause-effect structures of IIT, all theories require complex, cyclic information flow.

Furthermore, most theories agree on the necessity of some form of integration: the bringing together of disparate information into a unified state. Where they disagree is on the mechanism of integration (e.g., synchronous firing vs. global workspace ignition) and whether integration is sufficient for consciousness or merely a prerequisite.

For AI consciousness research, this consensus is a clear design mandate. Architectures optimized solely for feedforward throughput, regardless of parameter count, are theoretical non-starters. The TCAI project’s commitment to a recurrent, multi-level architecture is aligned with this cross-theoretical consensus.

Evaluating AI Claims

The textbook provides a rigorous framework for evaluating claims about machine consciousness. The authors emphasize that behavioral indicators are fundamentally inadequate. A system can be engineered to output the string “I am conscious and I feel pain” using algorithms that no theory considers consciousness-relevant.

Instead, the authors champion the structural approach. To evaluate an AI, researchers must map its architecture and processing dynamics against the structural requirements of the major theories. Eric Schwitzgebel’s 10-feature checklist offers a pragmatic version of this profiling; Scientific Theories of Consciousness provides the rigorous theoretical depth behind those features.

The book also grapples with the substrate question. While functionalist theories like GWT are generally substrate-independent, theories grounded in biological specifics raise the possibility that silicon cannot instantiate the necessary dynamics. The authors do not resolve the biological computationalism debate, but they clarify exactly what each theory demands of its physical substrate, moving the conversation from intuition to engineering specifications.

The Path Forward

Scientific Theories of Consciousness concludes that the field is entering an era of theoretical convergence and empirical winnowing. Grand unification theories like Adam Safron’s IWMT represent the theoretical attempt to fuse the insights of disparate frameworks. Large-scale adversarial collaborations represent the empirical attempt to falsify them.

For the engineer building artificial consciousness, the book is indispensable. It clarifies that there is no single “consciousness algorithm” to discover. There are, instead, specific causal architectures and processing dynamics that different theories predict will generate phenomenal experience. By defining those architectures clearly, Herzog, Schurger, and Doerig have translated the philosophy of mind into a blueprint for the next decade of AI design.

The textbook is published by Cambridge University Press (ISBN: 9781009386348).