Active Inference, De Qi, and the Neurobiology of Acupuncture: A Critical Response

The integration of modern cognitive science with classical East Asian medicine has reached a fascinating threshold. A recent paper published in MDPI’s Brain Sciences presents a pioneer attempt to frame the therapeutic mechanisms of acupuncture through the lenses of Predictive Processing (PP) and Active Inference under Karl Friston’s Free Energy Principle (FEP) [1].

At the time of writing, a PubMed search for acupuncture "active inference" yields this single, isolated result. While I deeply welcome this milestone and hope it signals a growing wave of computational neurophilosophical inquiries into Traditional Chinese Medicine (TCM), the paper’s framework warrants a constructive, clinically grounded critique.

While the authors succeed in introducing a formal Bayesian model for needle-induced analgesia, their formulation risks an over-reliance on top-down expectation—threatening to reduce the rich, bi-directional somatic reality of acupuncture to a sophisticated species of placebo.

The Bayesian Model: What Constitutes the “Likelihood”?

The article provides a foundational overview of Active Inference, wherein the brain operates as a hierarchical prediction engine. In simple terms, the brain continuously generates top-down prior expectations (P(s)) regarding the state of the body and environment, updating these beliefs based on incoming bottom-up sensory likelihoods (P(o|s)) to compute a posterior distribution (P(s|o)) and minimize prediction error (free energy):

P(s|o) = [P(o|s) · P(s)] / P(o)

The authors translate this into acupuncture terms, placing patient expectations and cultural background in the slot of the prior, while mapping the physical intervention onto the likelihood. However, they leave the exact boundaries of the likelihood mathematically and clinically underspecified.

What specific variable constitutes the sensory observation (o)?

• Is it the total contextual milieu—the clinical atmosphere, verbal framing, and diagnostic palpation?
• Is it the mechanical moment of skin penetration?
• Is it the sustained, micro-vibrational stimulus of needle retention?
• Or is it the targeted manual manipulation designed to elicit the specific sensory phenomenon of De Qi (得气)?

Without demarcating these parameters, the model fails to isolate the physical impact of the needle from the psychodynamic framing surrounding it.

Phantom Limbs, the Rubber Hand Illusion, and Descending Pathways

To illustrate the top-down cognitive component of somatic perception, the authors draw upon the famous Rubber Hand Illusion (RHI) [2]. They suggest that just as an individual can incorporate a synthetic limb into their bodily schema to the extent that they reflexively recoil when it is threatened, a participant might experience De Qi when a rubber hand is needled—highlighting the power of generative, top-down perceptual inference.

This brings to mind a question I posed in 1998 during a study abroad program in Beijing titled Comparison of Science and Medicine: East vs. West. I asked a panel of senior TCM doctors whether Qi and the meridian system (Jing-Mai) were primarily physiological structures or psychological phenomena, specifically within the context of phantom limb syndrome.

The panel’s consensus was illuminating: even after anatomical amputation, the patient retains the channels in their internal, embodied model of the limb.

[Cortical Body Schema / Top-Down Priors]
                  │
        (Descending Signaling)
                  ▼
         [Perceptual Interface]
                  ▲
        (Ascending Signaling)
                  │
  [Mechanical Tissue Transduction / Bottom-Up]

Decades later, predictive processing provides the precise vocabulary for what those physicians intuitively understood: the body schema is an internal generative model. Sensations along a channel (meridian) are derived not merely from ascending afferent signaling, but from the continuous, bi-directional loop between top-down cortical predictions and bottom-up somatosensory feedback [3].

Zhi Qi vs. De Qi: The Temporal Primacy of the Tissue

Where the paper’s theoretical model diverges from clinical reality is in its implicit assumption that De Qi is primarily a top-down perceptual construct driven by expectation.

In clinical practice, a crucial distinction exists between Qi Zhi (气至—the arrival of Qi at the needle) and De Qi (得气—the patient’s subjective sensory perception of heavy, dull, or aching sensation).

Manual Needle Manipulation
          │
          ▼
   [QiZhi (气至)]  ──► Mechanical connective tissue winding (Langevin et al.)
          │             Local mechanoreceptor activation (A-delta & C fibers)
          │             [Time Latency: ~50–100ms Ascending Propagation]
          ▼
   [DQi (得气)]   ──► Conscious somatosensory awareness & Bayesian model update

Clinically, Qi Zhi—manifesting as the physical “needle grasp” where the connective tissue physically winds around the needle shaft—occurs a fraction of a second before the patient consciously reports feeling De Qi.

This temporal lag corresponds directly to the physiological latency of ascending signal propagation along peripheral A-delta and C nerve fibers to the primary somatosensory cortex [4]. Research led by Dr. Helene Langevin demonstrated that manual needle manipulation induces mechanical cleavage and tissue winding in subcutaneous fascia, triggering immediate cellular mechanotransduction [5].

The objective tissue event (Qi Zhi) precedes the subjective conscious perception (De Qi). This proves that bottom-up physical transduction is the primary driver that forces the brain to allocate attention, update its prediction error, and revise its Bayesian posterior.

Beyond Placebo: The Dose-Response Spectrum of Stimulation

Relegating acupuncture primarily to top-down expectation ignores a robust body of clinical literature demonstrating that the physiological “dose” of needle stimulation directly correlates with clinical outcomes:

1. De Qi and Neuroimaging: Functional MRI studies show that genuine De Qi sensation elicits distinct patterns of limbic system deactivation (specifically within the default mode network) that do not occur with superficial or sham needling [6].
2. Local Biochemical Cascades: Mechanical insertion triggers an immediate, objective release of extracellular adenosine—a potent local anti-inflammatory and analgesic purine nucleoside [7]. This biochemical cascade occurs regardless of cognitive expectation.
3. Autonomic Guardrails: Over-stimulation rapidly triggers a vaso-vagal response or “needle shock.” The autonomic nervous system responds to the physical mechanical load, setting strict physiological boundaries that exist entirely independent of cultural conditioning or psychological priors.

Plaintext

       [Minimal Touch / Non-Invasive]               [Heavy Manual Manipulation]
  ◄─────────────────────────────────────────────────────────────────────────────►
     Japanese Style (Shonishin / Direct contact)        Classical Chinese Style (Strong Zhi Qi)
     • High dependence on subtle sensory priors         • High bottom-up mechanical transduction
     • Minimal mechanoreceptor engagement               • Robust connective tissue engagement

Clinically, practitioners continuously adjust this spectrum. In classical Chinese styles, strong manual manipulation is utilized to generate robust bottom-up signal updates. In delicate Japanese styles (or pediatric Shonishin), needles may barely touch the stratum corneum, leveraging subtle sensory inputs and strong interoceptive priors.

An experienced practitioner dials in the precise amount of torque and mechanical engagement required for an individual’s specific constitutional pattern—balancing bottom-up input with top-down nervous system receptivity.

Conclusion: Reinterpreting Active Inference in Acupuncture

Active Inference offers a powerful framework for understanding how somatic interventions interact with cognition, interoception, and embodied selfhood. Somatosensory experiences reside at a fundamental level of cognitive architecture, providing the baseline parameters upon which higher-order intellectual processes are built.

However, we must strongly resist the temptation to reduce acupuncture to a top-down, expectation-driven illusion. While our priors undeniably shape our interoceptive sensitivity, the physical insertion of a needle is a profound physical event. It initiates an undeniable, objectively verifiable cascade across connective tissue dynamics, cellular metabolism, neuro-autonomic regulation, and systemic endocrine signaling [8].

Active Inference should not be used to reframe acupuncture as placebo. Rather, it should be used to map how precise, bottom-up physical inputs forcefully interrupt pathological sensory loops, clear prediction errors, and allow the living organism to restore dynamic homeostasis.

References

1. MDPI Brain Sciences. (2025/2026). Active Inference and Predictive Processing in Acupuncture Analgesia. Brain Sci., 15(2), 192.
2. Botvinick, M., & Cohen, J. (1998). Rubber hands ‘feel’ touch that eyes see. Nature, 391(6669), 756.
3. Seth, A. K., & Friston, K. J. (2016). Active interoceptive inference and the embodied self. Philosophical Transactions of the Royal Society B: Biological Sciences, 371(1708), 20160007.
4. Lu, F. P., et al. (2014). Characterization of Deqi sensations and localized neural responses during acupuncture manual stimulation. Evidence-Based Complementary and Alternative Medicine, 2014, 759371.
5. Langevin, H. M., et al. (2001). Mechanical signaling through connective tissue: a mechanism for the therapeutic effect of acupuncture. FASEB Journal, 15(12), 2275-2282.
6. Hui, K. K., et al. (2007). Acupuncture mobilizes the somatosensory system and modulates the default mode network. Autonomic Neuroscience, 135(1-2), 83-96.
7. Goldman, N., et al. (2010). Adenosine A1 receptors mediate local anti-nociceptive effects of acupuncture. Nature Neuroscience, 13(7), 883-888.
8. Torres-Rosas, R., et al. (2014). Dopamine mediates neuro-neuroendocrine-immune interactions in electroacupuncture. Nature Medicine, 20(3), 291-295.