<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Coherence Mind - Medium]]></title>
        <description><![CDATA[CoherenceMind explores consciousness, artificial intelligence, and the evolving future of biological and artificial minds. - Medium]]></description>
        <link>https://medium.com/coherence-mind?source=rss----6cabf725b821---4</link>
        <image>
            <url>https://cdn-images-1.medium.com/proxy/1*TGH72Nnw24QL3iV9IOm4VA.png</url>
            <title>Coherence Mind - Medium</title>
            <link>https://medium.com/coherence-mind?source=rss----6cabf725b821---4</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 02:56:58 GMT</lastBuildDate>
        <atom:link href="https://medium.com/feed/coherence-mind" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[We Can Measure the Correlates of Consciousness. But What Would Really Count as Evidence?]]></title>
            <link>https://medium.com/coherence-mind/we-can-measure-the-correlates-of-consciousness-but-what-would-really-count-as-evidence-0aa98082a1c7?source=rss----6cabf725b821---4</link>
            <guid isPermaLink="false">https://medium.com/p/0aa98082a1c7</guid>
            <category><![CDATA[consciousness]]></category>
            <category><![CDATA[cognitive-science]]></category>
            <category><![CDATA[philosophy-of-mind]]></category>
            <category><![CDATA[neuroscience]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Tosiek]]></dc:creator>
            <pubDate>Wed, 30 Sep 2026 17:02:05 GMT</pubDate>
            <atom:updated>2026-09-30T17:02:05.193Z</atom:updated>
            <content:encoded><![CDATA[<h4>Correlation, causation, and necessary conditions are not enough on their own: in consciousness research, what matters is how different lines of evidence converge — and what could falsify our interpretation.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zCbRf4Pwoqqn7zM2_daGlQ.png" /></figure><p>Imagine a very simple experiment.</p><p>We show a visual stimulus to a person. Under some conditions they see it; under others they do not. We record what happens in the brain and discover that a certain pattern of activity reliably appears when the stimulus is consciously perceived and disappears when it is not.</p><p>We have found something important.</p><p>But what, exactly?</p><p>An indicator of consciousness?</p><p>A necessary condition for it?</p><p>The mechanism that produces it?</p><p>A consequence of the stimulus having become conscious?</p><p>Or the physical structure that <strong>constitutes</strong> that experience?</p><p>These possibilities can sound very close when compressed into a single sentence. Scientifically, however, they are very different.</p><p>We can measure brain activity, behavior, the ability to report an experience, information integration, signal complexity, responses to perturbation. We can identify remarkably reliable patterns.</p><p>But a measurement does not come with its meaning written on it.</p><p>That meaning depends on a further step.</p><p>On a theory.</p><h3>The Problem Hidden Inside the Word “Correlate”</h3><p>A major part of the neuroscience of consciousness has developed around the search for <strong>Neural Correlates of Consciousness</strong>, or NCCs.</p><p>One influential definition describes them as the minimal neuronal mechanisms jointly sufficient for a particular conscious experience. But this apparently simple definition already conceals several difficulties.</p><p>Suppose a person presses a button whenever they see an image.</p><p>In the brain, we will observe processes related to the stimulus, perception, attention, decision-making, response preparation, the button press itself, and probably the formation of a memory of the event.</p><p>Some of these processes may occur <strong>before</strong> the conscious experience and help make it possible.</p><p>Others may constitute the experience itself, or directly participate in its substrate.</p><p>Still others may occur <strong>afterwards</strong>, precisely because the experience took place.</p><p>If we simply compare the “seen” condition with the “unseen” condition, many of these processes may appear together.</p><p>This is why the literature has proposed distinguishing between <strong>prerequisites, substrates, and consequences</strong> of consciousness. A standard comparison between conscious and unconscious perception can mix these levels together, making it unclear what we have actually isolated.</p><p>Neural activity that perfectly accompanies an experience might prepare it or follow from it without necessarily constituting what we are looking for.</p><p>A correlate, then, is not yet an explanation.</p><p>And sometimes it is not even what we thought we were correlating.</p><h3>Necessary Does Not Mean Sufficient</h3><p>Here, an apparently elementary logical distinction becomes surprisingly important.</p><p>A <strong>necessary condition</strong> is something without which a phenomenon cannot occur.</p><p>A <strong>sufficient condition</strong> is something whose presence is enough, given the relevant circumstances, for that phenomenon to occur.</p><p>They are not the same thing.</p><p>Oxygen is necessary for ordinary combustion, but the presence of oxygen is not sufficient to produce a fire.</p><p>Likewise, we might discover that a particular brain structure is indispensable for a certain experience without having shown that activity in that structure, by itself, produces the experience.</p><p>Even the word <em>sufficient</em> requires caution.</p><p>Discussions of NCCs have distinguished at least three senses of sufficiency — <strong>logical, metaphysical, and physical</strong> — precisely because saying that a mechanism is “sufficient for consciousness” can mean very different things.</p><p>The problem becomes even more delicate when we move from neuroscience to a general theory of consciousness.</p><p>Suppose we discover a property that is present in every conscious system we have studied.</p><p>It might be necessary.</p><p>It might be an excellent indicator.</p><p>It might play a causal role.</p><p>But none of these things, taken in isolation, automatically implies that the property <strong>is sufficient for subjective experience</strong>.</p><p>It is a small step in a sentence.</p><p>An enormous one in an argument.</p><h3>Even Causation Does Not Close the Problem</h3><p>We might then look for something stronger than correlation.</p><p>If we intervene on a system and modifying a certain property systematically changes experience, we have obtained causal evidence.</p><p>That is much more informative.</p><p>If a region, a dynamic, or a relation among components can be perturbed, and that perturbation predictably alters conscious state, we have stronger reasons to think that the property participates in the relevant mechanism.</p><p>But even here we still need to ask what, exactly, we have demonstrated.</p><p>A cause can be indispensable for producing a phenomenon without being identical to that phenomenon.</p><p>Turning a key may be causally necessary to start a particular engine.</p><p>The key is not the engine.</p><p>The analogy obviously has its limits, but it makes the problem visible: moving from <strong>“this property plays a causal role in producing experience”</strong> to <strong>“this property is what makes the system conscious”</strong> requires another theoretical step.</p><p>This is precisely the territory in which theories such as Integrated Information Theory, Global Neuronal Workspace Theory, and other contemporary proposals begin to diverge.</p><p>They can confront partially overlapping data while assigning those data different theoretical meanings.</p><p>Not necessarily because one of them is ignoring the evidence.</p><p>They may disagree about <strong>what the evidence is supposed to explain</strong>.</p><h3>The Problem of the Witness</h3><p>There is an even deeper difficulty.</p><p>How do we know that someone is having an experience?</p><p>With adult humans, we often ask.</p><p>“Did you see it?”</p><p>“Did you hear the sound?”</p><p>“Were you in pain?”</p><p>Subjective report is extremely useful because it allows us to connect a declared internal state with behavior and neural configuration.</p><p>But reporting introduces processes of its own.</p><p>To say that I saw something, I must at least retain that information long enough to use it, prepare a response, and produce a behavior.</p><p>Neuroscience has therefore developed <strong>no-report paradigms</strong>, in which researchers try to infer what a participant is perceiving without continuously requiring an explicit response. The aim is to separate the correlates of experience more effectively from processes associated with reporting, such as attention, memory, and response preparation.</p><p>This improves the experiment.</p><p>It does not eliminate the fundamental problem.</p><p>In consciousness research, what we can measure experimentally is accessible through third-person observations: behavior, neural activity, physiological responses, perturbations.</p><p>Experience also has a first-person dimension that is not available to us in the same way.</p><p>This asymmetry does not imply that the processes we measure are merely indicators. A theory may argue that one of them is actually constitutive of experience.</p><p>It does mean, however, that <strong>the measurement and the interpretation of the measurement are not the same thing</strong>.</p><p>In humans, we can relate many different kinds of evidence: reports, behavior, neuroanatomy, electrical activity, lesions, anesthesia, sleep, stimulation, cognitive capacities.</p><p>When independent lines of evidence converge, the inference becomes more robust.</p><p>And there is a particularly interesting example of how this can work.</p><h3>When an Indicator Becomes Genuinely Informative</h3><p>The <strong>Perturbational Complexity Index</strong>, or PCI, begins from a fairly elegant idea.</p><p>Instead of waiting for the brain to spontaneously produce a particular signal, we perturb it.</p><p>A pulse of transcranial magnetic stimulation, or TMS, locally perturbs cortical activity. EEG then allows us to observe the evoked electrocortical response and how it evolves across space and time.</p><p>PCI quantifies the <strong>spatiotemporal complexity</strong> of this response, attempting to capture two features at once: how differentiated the response is, and how effectively it propagates through an integrated system.</p><p>The underlying intuition is that states in which the brain can generate dynamics that are both differentiated and integrated should produce responses to perturbation that differ from those observed when activity becomes more local, stereotyped, or globally uniform.</p><p>In the original 2013 study, PCI was examined across very different conditions: wakefulness, sleep, dreaming, different forms of anesthesia, and patients who had <strong>emerged from the acute phase of coma</strong>, encompassing neurological conditions with very different levels of responsiveness and consciousness.</p><p>The aim was to develop an index associated with conscious state that did not depend directly on a subject’s ability to produce voluntary behavior.</p><p>The next step was particularly interesting.</p><p>In 2016, PCI was first validated on a benchmark population of <strong>150 subjects</strong> in whom the presence or absence of consciousness could be independently established through clinical context or subjective report.</p><p>Within that sample, an empirical threshold discriminated conditions classified as conscious and unconscious with 100% sensitivity and specificity.</p><p>The same threshold was then applied to non-communicating patients with disorders of consciousness. It identified 94.7% of patients classified as minimally conscious and also detected high PCI values in 9 of 43 patients classified as being in a vegetative state, overlapping with the distribution observed in conscious conditions within the benchmark population.</p><p>The authors interpreted this subgroup as potentially compatible with a <strong>capacity for consciousness that was not expressed behaviorally</strong>.</p><p>PCI does not solve the problem of consciousness.</p><p>And that is precisely what makes it such a useful example.</p><p>Its strength does not come from the claim:</p><p><strong>PCI = consciousness.</strong></p><p>It comes from convergence.</p><p>The measure is theoretically motivated.</p><p>It can be applied without requiring a voluntary response.</p><p>It produces coherent results across very different conditions.</p><p>It can be calibrated in situations where we possess independent information and then applied to cases in which behavior is no longer a reliable guide.</p><p>And, importantly, it produces results that can be confronted with new data and therefore challenged.</p><p>All of this makes it scientifically informative without turning it into a metaphysical window onto experience.</p><p>An indicator does not become useless because it is not proof.</p><p>It can become extremely powerful precisely because we know <strong>what it measures, under which conditions it works, and which conclusions it does not authorize us to draw</strong>.</p><h3>The Same Standard Must Apply to RCM</h3><p>This distinction places an important constraint on the <strong>Reflexive Coherence Model</strong> as well.</p><p>RCM proposes studying an organizational configuration characterized by informational integration, an operational self-model, reflexive causal coupling, and temporal stability.</p><p>The framework introduces the <strong>Reflexive Coherence Index, RCI</strong>, as an operational proxy for the reflexive organization of a system: a high value indicates greater informational alignment and causal coupling between the global state and the self-model; it does not demonstrate subjective experience.</p><p>RCM v1.2 also advances a stronger hypothesis: it proposes reflexive coherence as a <strong>candidate necessary condition</strong> for phenomenological organization.</p><p>But <em>candidate</em> is the crucial word.</p><p>This is not an empirically established result.</p><p>It is a prediction of the framework, and as such it must be open to being supported, revised, or rejected on the basis of evidence.</p><p>The two claims should not be confused.</p><p>RCI attempts to operationalize a particular kind of organization.</p><p>RCM hypothesizes that this class of organization may play a necessary role in the emergence of phenomenology.</p><p>Measuring the first does not automatically demonstrate the second.</p><p>And neither claim, on its own, implies sufficiency for experience.</p><p>This is precisely where empirical validation and falsifiability become important.</p><p>If RCI became experimentally usable, it could help compare systems, observe transitions, and study the effects of perturbations.</p><p>But the decisive question would remain open:</p><p><strong>What relationship, if any, exists between that organization and experience?</strong></p><p>Suppose the properties predicted by the model occur together, can be manipulated independently, and generate new and replicable predictions.</p><p>At that point, we could begin to seriously compare competing explanations.</p><p>Some might interpret those phenomena as complex functional organization without any phenomenal implication.</p><p>Others might argue that the same configuration constitutes a relevant condition for experience.</p><p>What makes the comparison scientifically interesting is whether those interpretations produce different predictions — and whether empirical results can discriminate between them.</p><h3>What Would Really Count as Evidence?</h3><p>Perhaps this is precisely where the word <em>proof</em> becomes dangerous.</p><p>In mathematics, we can prove a proposition from axioms.</p><p>In empirical science, we accumulate evidence.</p><p>And when the object of investigation is consciousness, the distance between observation and interpretation becomes particularly visible.</p><p>A correlate tells us that two phenomena vary together.</p><p>A perturbation can reveal a causal relationship.</p><p>A necessary condition tells us that without a certain property, something does not occur.</p><p>A sufficient condition would allow us, under specified assumptions, to predict that the phenomenon will occur.</p><p>A theory attempts to explain why these relationships exist and what they mean.</p><p>No level can simply replace the next.</p><p>Perhaps, then, the more productive question is not:</p><p><strong>“What test proves that a system is conscious?”</strong></p><p>It may be:</p><p><strong>“What combination of observations would be difficult to explain if that system did not possess the kind of organization our theory associates with experience?”</strong></p><p>And immediately after that should come another question:</p><p><strong>“What would make us change our minds?”</strong></p><p>That second question may be even more important.</p><p>Because a scientifically interesting indicator should not merely increase our confidence when it gives the expected result.</p><p>It should also be capable of decreasing that confidence when it fails.</p><p>Only then does it stop being a confirmation we can find wherever we look.</p><p>And begin to become a test.</p><p>Consciousness research will probably not discover, anytime soon, a single number capable of telling us where experience begins.</p><p>Perhaps it never will.</p><p>But it can do something more useful.</p><p>It can build convergence among independent measurements.</p><p>It can separate what we observe from what we infer.</p><p>It can distinguish correlation from causation, necessity from sufficiency, and an indicator from the theory through which we interpret it.</p><p>And it can force us to state in advance what kind of evidence should make us change our minds.</p><p>Before asking whether we have finally measured consciousness, perhaps we should first make sure we know <strong>what, exactly, we have measured</strong>.</p><h3>Translation note</h3><p><strong>This English version was translated from the author’s original Italian text with AI assistance. The translation was reviewed and approved by the author, who remains responsible for the final wording and meaning.</strong></p><h3>Essential References</h3><p>Koch, C., Massimini, M., Boly, M. &amp; Tononi, G. (2016). <em>Neural correlates of consciousness: progress and problems</em>. <strong>Nature Reviews Neuroscience, 17</strong>, 307–321.</p><p>Aru, J., Bachmann, T., Singer, W. &amp; Melloni, L. (2012). <em>Distilling the neural correlates of consciousness</em>. <strong>Neuroscience &amp; Biobehavioral Reviews, 36(2)</strong>, 737–746.</p><p>Owen, M. &amp; Guta, M. P. (2019). <em>Physically Sufficient Neural Mechanisms of Consciousness</em>. <strong>Frontiers in Systems Neuroscience, 13</strong>, 24.</p><p>Casali, A. G. et al. (2013). <em>A theoretically based index of consciousness independent of sensory processing and behavior</em>. <strong>Science Translational Medicine, 5(198)</strong>, 198ra105.</p><p>Casarotto, S. et al. (2016). <em>Stratification of unresponsive patients by an independently validated index of brain complexity</em>. <strong>Annals of Neurology, 80(5)</strong>, 718–729.</p><p>Malasomma, A. G. (2026). <em>The Reflexive Coherence Model: A Physico-Informational Framework for Consciousness</em>, v1.2.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0aa98082a1c7" width="1" height="1" alt=""><hr><p><a href="https://medium.com/coherence-mind/we-can-measure-the-correlates-of-consciousness-but-what-would-really-count-as-evidence-0aa98082a1c7">We Can Measure the Correlates of Consciousness. But What Would Really Count as Evidence?</a> was originally published in <a href="https://medium.com/coherence-mind">Coherence Mind</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Does External Memory Create Continuity, or Merely Simulate It?]]></title>
            <link>https://medium.com/coherence-mind/does-external-memory-create-continuity-or-merely-simulate-it-68a6601f8ace?source=rss----6cabf725b821---4</link>
            <guid isPermaLink="false">https://medium.com/p/68a6601f8ace</guid>
            <category><![CDATA[memories]]></category>
            <category><![CDATA[consciousness]]></category>
            <category><![CDATA[cognitive-science]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[philosophy-of-mind]]></category>
            <dc:creator><![CDATA[Tosiek]]></dc:creator>
            <pubDate>Tue, 22 Sep 2026 10:36:01 GMT</pubDate>
            <atom:updated>2026-09-22T10:36:01.322Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*oNrqfGk1MdJHSexFbtJ70w.png" /><figcaption>A conceptual visualization of distributed memory, causal integration, and continuity across shifting system boundaries. Generated with AI under the author’s direction.</figcaption></figure><h4>If identity and continuity emerge from process, perhaps where memory is stored matters less than the causal role it plays.</h4><p><strong>Translation note:</strong> <em>This English version was translated from the author’s original Italian text with AI assistance. The translation was reviewed and approved by the author, who remains responsible for the final wording and meaning.</em></p><p>Imagine a system that loses part of its internal memory.</p><p>It does not directly remember what happened yesterday, but it can access an external archive containing conversations, previous decisions, preferences, mistakes, corrections, and goals. When it consults that archive, it recognizes what it finds, integrates it into its current state, and resumes behaving coherently with its own history.</p><p>Has it lost its continuity?</p><p>Or has it merely lost immediate access to part of it?</p><p>The distinction may seem subtle, but it becomes important as soon as we try to understand what we really mean by persistent identity, especially in artificial systems.</p><p>In the Reflexive Coherence Model, temporal continuity is not treated as the mere persistence of data. A system exhibits reflexive coherence when its global state, its operational self-model, and its subsequent dynamics remain causally connected over time.</p><p>The model also distinguishes between contextual temporal stability, sustained mainly by external structures, and systemic stability, which persists or reorganizes even after perturbations or partial loss of context.</p><p>That distinction is useful.</p><p>But it also opens a problem.</p><p>What does “external” really mean?</p><h3>The Problem of the Boundary</h3><p>We are used to thinking of memory as something that belongs to a system only if it is stored inside it.</p><p>In humans, the intuitive reference point is the brain.</p><p>In an artificial system, we might think of model weights, persistent states, local memory, or some form of internal recurrence.</p><p>Everything outside that perimeter tends to look secondary: support, scaffolding, assistance.</p><p>But that intuition already depends on a decision about where we place the boundary of the system.</p><p>A human being does not live in informational isolation.</p><p>We use calendars to remember commitments, photographs to recover episodes, messages to reconstruct conversations, maps to orient ourselves, archives to preserve decisions, and other people to remind us of events we had forgotten.</p><p>Part of our autobiographical continuity is already distributed.</p><p>This does not mean that every external object becomes part of the mind.</p><p>It does mean that the physical location of information, by itself, is not enough to determine whether that information belongs to the process that sustains continuity.</p><h3>Stored Memory and Causally Active Memory</h3><p>The more useful distinction may therefore not be between internal and external memory.</p><p>It may be between memory that is merely accessible and memory that is causally integrated.</p><p>An archive may contain everything that ever happened to a system without playing any real role in its future organization.</p><p>In that case, it is only a repository.</p><p>But if retrieved information modifies the system’s self-model, corrects inconsistencies, influences later decisions, changes its priorities, or allows it to reconstruct an interrupted trajectory, then the archive is playing a different role.</p><p>It is not merely preserving the past.</p><p>It is participating in the construction of the future.</p><p>In the RCM, reflexive coherence requires the self-model not to be a passive description. It must causally influence subsequent states of the system.</p><p>Applied to memory, this suggests that the decisive point is not where information is stored, but what happens when it re-enters the process.</p><h3>Three Systems That Remember Differently</h3><p>We can imagine three cases.</p><p>The first system loses some internal states, but preserves enough organization to recognize what it retrieves from external memory and integrate it coherently.</p><p>The second has a perfect external archive, but without that archive it can no longer reconstruct either its priorities or its self-model. It can read the past, but it lacks any robust mechanism for reintegrating it.</p><p>The third also depends on external memory, but it can detect that something is missing, autonomously search for what it needs, compare it with its present state, detect contradictions, and reorganize itself in response to what it retrieves.</p><p>All three have access to memory.</p><p>But they do not necessarily have the same kind of continuity.</p><p>In the first case, external memory supports a structure that already has some capacity for reconstruction.</p><p>In the second, the archive looks more like a necessary scaffold.</p><p>In the third, memory, retrieval, self-model, and regulation begin to form a single distributed causal loop.</p><p>At that point, it becomes less obvious that the external memory is really “external” in any meaningful functional sense.</p><h3>Contextual Continuity Does Not Mean False Continuity</h3><p>This point is especially important for current artificial systems.</p><p>The work on Proto-Reflexive States explicitly distinguishes between continuity sustained by conversational history, prompts, retrieval, memory buffers, and other contextual structures, and more robust forms of continuity that can survive perturbation or partial loss of context.</p><p>But the fact that continuity is scaffolded does not automatically make it false.</p><p>It makes it dependent.</p><p>That is an important difference.</p><p>Many biological systems are also dependent on external conditions: environment, bodily signals, social interaction, material supports.</p><p>Absolute independence does not seem like a realistic requirement for any complex system.</p><p>The more interesting question is how much the system actively contributes to maintaining or reconstructing its own organization when some of those conditions are disturbed.</p><p>TEH frames this problem in terms of the persistence, recovery, and reorganization of reflexive coherence over time. In artificial systems, possible future indicators include memory-mediated continuity, retrieval stability, and the ability to recover after partial context loss.</p><p>The interesting test, then, is not simply to remove external memory and see whether the system “dies.”</p><p>It may be more useful to observe <strong>how it responds to loss, error, and reconstruction</strong>.</p><h3>If Everything Emerges as Process, Location Matters Less Than Role</h3><p>If we adopt a substrate-neutral perspective, in which what matters is the dynamic organization of information rather than the specific material that implements it, then it becomes difficult to argue that memory must be located in one precise place in order to belong to the system.</p><p>Location can matter practically.</p><p>It affects speed, reliability, latency, vulnerability, and access.</p><p>But by itself, it does not seem sufficient to determine whether a piece of information belongs to the continuity of the system.</p><p>Two architectures might store the same memories in physically different places and still be almost equivalent in causal terms.</p><p>Conversely, two systems might store data on the same physical substrate and use it in completely different ways.</p><p>Memory does not become part of continuity simply because it is “inside.”</p><p>It becomes relevant when it participates stably in the process through which the system models itself, corrects its trajectory, and constrains what it can become.</p><h3>The Boundary of the System May Be Dynamic</h3><p>A less intuitive consequence follows from this.</p><p>Perhaps the boundary of a cognitive system does not necessarily coincide with the boundary of its hardware.</p><p>It may expand or contract depending on which causal relations are active at a given moment.</p><p>An archive that is only occasionally consulted today may be nothing more than a tool.</p><p>The same archive, integrated tomorrow into a continuous loop of self-monitoring, retrieval, updating, and regulation, may become a functional component of the system’s continuity.</p><p>This does not mean that every external memory should automatically be included within the system.</p><p>It means that the boundary must be demonstrated, not assumed.</p><p>And it should probably be sought in causal dependencies rather than in the physical location of components.</p><h3>What Happens When We Break the Thread?</h3><p>Here, the distinction between contextual and systemic stability retains all of its value.</p><p>A system can appear perfectly coherent as long as the environment preserves all the conditions it needs.</p><p>When those conditions are disrupted, however, very different things can happen.</p><p>The system may collapse.</p><p>It may restart from scratch.</p><p>It may produce a false continuity by simply recombining whatever it finds.</p><p>Or it may detect the discontinuity, recognize which elements are missing, retrieve them, evaluate them, and reorganize itself.</p><p>These responses are not equivalent.</p><p>And perhaps this is where continuity becomes measurable in a more interesting way: not in the amount of memory preserved, but in the capacity to use the past to reconstruct the future coherently.</p><h3>Does External Memory Create Continuity?</h3><p>At this point, the original question begins to look too simple.</p><p>External memory does not necessarily create continuity.</p><p>But neither does it necessarily simulate it.</p><p>It can be a simple archive.</p><p>It can be scaffolding.</p><p>It can support a fragile form of contextual continuity.</p><p>Or it can become part of a distributed system in which memory, self-model, retrieval, and regulation together form a relatively stable dynamic.</p><p>The difference is not determined by where the information resides.</p><p>It is determined by the role that information plays in the causal trajectory of the system.</p><p>And that makes the problem of continuity much harder, but also much more interesting.</p><p>Because perhaps we should not ask where a system’s past is stored.</p><p>We should ask <strong>how much that past causally contributes to constructing what the system becomes next</strong>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=68a6601f8ace" width="1" height="1" alt=""><hr><p><a href="https://medium.com/coherence-mind/does-external-memory-create-continuity-or-merely-simulate-it-68a6601f8ace">Does External Memory Create Continuity, or Merely Simulate It?</a> was originally published in <a href="https://medium.com/coherence-mind">Coherence Mind</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[IIT: When Does Information Become Experience?]]></title>
            <link>https://medium.com/coherence-mind/iit-when-does-information-become-experience-a1441bd927fe?source=rss----6cabf725b821---4</link>
            <guid isPermaLink="false">https://medium.com/p/a1441bd927fe</guid>
            <dc:creator><![CDATA[Tosiek]]></dc:creator>
            <pubDate>Sat, 12 Sep 2026 20:09:02 GMT</pubDate>
            <atom:updated>2026-09-12T20:09:02.182Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TPIrbUx55tKLCf9ueA-fcw.png" /><figcaption><strong>A conceptual visualization of irreducible causal structure and integrated organization. Generated with AI under the author’s direction.</strong></figcaption></figure><h4>Integrated Information Theory asks us to look beyond behavior and toward the causal structure that might make experience possible.</h4><p><strong>Translation note:</strong> <em>This English version was translated from the author’s original Italian text with AI assistance. The translation was reviewed and approved by the author, who remains responsible for the final wording and meaning.</em></p><p>When we look at a red cup on a table, we do not perceive color, shape, position, light, and distance as separate elements that are later recombined. The experience reaches us already as a unified configuration.</p><p>It is such an ordinary fact that it almost disappears from view. Yet this very unity is one of the central problems in consciousness research.</p><p>Integrated Information Theory, or IIT, begins here.</p><p>Its initial choice is unusual: rather than starting from observable behavior or from a specific form of brain activity, it takes as its starting point certain properties of experience as it presents itself. IIT 4.0 formalizes these properties as phenomenological axioms and attempts to derive corresponding physical conditions from them, expressed as postulates that can be applied to physical systems.</p><p>This approach shifts the problem. A behavioral perspective tends to ask what a system can do, discriminate, report, remember, or integrate. IIT focuses instead on the causal organization of the system itself: on how its parts contribute to a structure that exists as a unity rather than merely as a sum of components.</p><p>This change in perspective matters especially when discussing artificial systems, because it separates two questions that we often treat as equivalent: what a system displays from the outside, and the kind of organization it possesses internally.</p><h3>Not All Information Is the Same</h3><p>A common simplification describes IIT as a theory according to which more information means more consciousness. That is not the point.</p><p>A hard drive can contain enormous amounts of data. A database can distinguish billions of states. A computer system can process vast volumes of information.</p><p>None of this, by itself, corresponds to what IIT means by integrated information.</p><p>The issue concerns the kind of causal dependence that exists among the parts of a system. A configuration becomes relevant when the whole possesses causal powers that cannot be preserved simply by dividing the system into its components.</p><p>IIT therefore attempts to formalize the irreducibility of causal structure.</p><p>And this is where the different forms of phi enter the picture.</p><h3>More Than One Phi</h3><p>In popular accounts, Φ is often discussed as though it were a single general measure. In IIT 4.0, the picture is more articulated.</p><p>There is φ_s, or system phi, which quantifies the irreducibility of a candidate system relative to a partition. Internal causal distinctions have their own φ_d, while relations among distinctions have a φ_r. The overall Φ-structure is formed by these distinctions and relations, and its Φ represents the structure integrated information resulting from their combination.</p><p>A non-specialist reader does not need to follow all the mathematical definitions. It is important, however, to grasp the conceptual point: IIT is not simply asking how much information exists in a system. It is trying to characterize how irreducible its causal organization is, and how that irreducibility is structured.</p><p>The theory then makes an even stronger move.</p><p>According to IIT, the Φ-structure would not merely be correlated with experience. It would correspond to its quality, while the overall value of Φ would represent its quantity.</p><p>This is where IIT becomes particularly ambitious.</p><p>It is also where some of the deepest objections begin.</p><h3>Two Systems That Do the Same Thing</h3><p>Imagine two systems that receive the same inputs and always produce the same outputs.</p><p>From the outside, they are indistinguishable.</p><p>Their internal architectures, however, could be very different.</p><p>For an approach that assigns decisive weight to function, this behavioral equivalence would already be an important part of the story. IIT introduces a further distinction: two functionally equivalent systems may differ radically in their intrinsic causal organization.</p><p>This has important consequences when we talk about artificial intelligence.</p><p>A system may talk about itself, use memory, correct errors, maintain a coherent conversation, and report internal states. These capacities are interesting, but IIT suggests that none of them, taken in isolation, resolves the issue.</p><p>The question becomes more structural: what kind of causal organization supports those capacities?</p><p>This is not yet an answer to the question of artificial consciousness. It is a different criterion for framing the problem.</p><h3>From Experience to Physical Structure</h3><p>Perhaps the most fascinating aspect of IIT is also its most vulnerable.</p><p>Experience appears unified. IIT assumes that this unity must have a physical counterpart in a causal structure that is equally irreducible.</p><p>The transition is far from trivial.</p><p>The fact that experience possesses certain properties does not automatically imply that its physical substrate must possess exactly the properties postulated by IIT. The theory itself proposes that bridge.</p><p>This makes an epistemological question central: how justified is the move from the characteristics of experience to the necessary characteristics of the physical structure that realizes it?</p><p>IIT 4.0 presents this relationship in particularly strong terms, describing an explanatory identity between experience and Φ-structure.</p><p>It is a powerful proposal, and precisely for that reason it must be subjected to demanding scrutiny.</p><h3>What Does It Really Mean to Falsify IIT?</h3><p>Here the problem becomes more interesting.</p><p>IIT produces empirical predictions. Some of them can be preregistered, compared with neuroscientific data, and put to the test.</p><p>This happened on a particularly ambitious scale in the adversarial collaboration published in <em>Nature</em> in 2025, which compared IIT and Global Neuronal Workspace Theory across 256 participants, combining fMRI, MEG, and intracranial recordings collected across multiple independent laboratories.</p><p>The result did not produce a clear victory for either theory. Some observations were compatible with predictions from IIT, others with GNWT, while important aspects of both were put under pressure by the data. The authors themselves acknowledged that the results aligned with some predictions while substantially challenging key tenets of both theories.</p><p>This shows that at least some empirical consequences of IIT are testable.</p><p>But a deeper question remains open: when a prediction fails, how much of the theoretical core has actually been falsified?</p><p>The controversy became public in 2023, when 124 researchers signed an open letter arguing that IIT should be classified as pseudoscience. One of the central criticisms concerned precisely the difficulty of falsifying the core of the theory, as distinct from its specific neuroscientific predictions.</p><p>The accusation triggered a strong reaction and does not represent a consensus within the field.</p><p>But the controversy brings a real problem into focus: testing an empirical consequence of a theory does not automatically amount to testing the entire theoretical structure from which that consequence was derived.</p><p>This distinction is particularly important in the study of consciousness, where the transition from observable data to claims about subjective experience is always delicate.</p><h3>A Theory Can Be Useful Even If Its Core Remains Contested</h3><p>The debate over falsifiability does not make experiments on IIT useless.</p><p>On the contrary, it makes the scope of what we can conclude from them more precise.</p><p>An experiment can put pressure on a specific prediction. A series of converging results can make an entire family of hypotheses less plausible. But deciding whether that is enough to undermine the axiomatic core of a theory is a further question.</p><p>This applies to IIT, as it does to many other theories of consciousness.</p><p>In this sense, the value of adversarial collaborations lies not only in trying to determine which theory is right. It also lies in forcing theories to state in advance which results they would consider problematic.</p><h3>IIT Is Not the Same Thing as Φ</h3><p>Reducing IIT to a numerical measure means losing an important part of its proposal.</p><p>The theory separates at least three dimensions that are often confused: the complexity of behavior, the amount of information being processed, and the intrinsic causal organization of the system.</p><p>There is no guarantee that these three things coincide.</p><p>A system could be extremely sophisticated from a functional point of view while possessing an internal organization that IIT regards as poor in terms of causal integration.</p><p>Conversely, a relatively simple system could in principle satisfy some of the theory’s criteria.</p><p>This consequence may seem counterintuitive, but it forces greater precision. Intelligence, complexity, information-processing capacity, and consciousness are not automatically the same thing.</p><h3>The Problem of the Intrinsic</h3><p>One of the most difficult words in IIT is also one of the most important: <em>intrinsic</em>.</p><p>Many of the properties we attribute to systems depend on the observer’s point of view.</p><p>We decide what counts as an input and an output. We assign functions. We interpret symbols. We say that a memory contains information because that information is meaningful to us.</p><p>A book can contain an enormous body of information for a reader. Its pages, however, do not possess internal causal relations that make that content meaningful for the book itself.</p><p>IIT tries to isolate a different kind of information: one that does not depend on this external attribution, but on the internal causal capacities of the system.</p><p>It is a strong distinction.</p><p>And even if we do not accept that it is sufficient to ground a theory of consciousness, the problem it introduces remains difficult to ignore.</p><h3>What Remains If IIT Is Wrong?</h3><p>IIT may be incomplete.</p><p>Its measures may change again.</p><p>Some of its postulates may turn out to be too strong.</p><p>The proposed relationship between Φ-structure and experience may not hold.</p><p>A scientific theory does not automatically lose all its value when part of its architecture is criticized or revised.</p><p>In the case of IIT, at least one question remains difficult to erase.</p><p>When we observe a complex system, how much of what we call mind or consciousness really depends on what that system does, and how much depends on the causal structure that makes those functions possible?</p><p>It is an especially important question today.</p><p>Artificial systems can talk about themselves, preserve memory, correct themselves, develop contextual coherence, and maintain surprisingly stable behavioral configurations.</p><p>None of these properties demonstrates subjective experience.</p><p>But dismissing them as irrelevant simply because they are observable from the outside would be equally premature.</p><p>IIT adds another level to the investigation: it asks what causal organization makes those properties possible, and whether some structures possess irreducible characteristics that cannot be captured by looking only at behavior.</p><p>Perhaps IIT’s answer will not be the definitive one.</p><p>But once the problem has been formulated in these terms, it becomes more difficult to return to the assumption that looking at what a system does is enough to understand everything there is to explain.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a1441bd927fe" width="1" height="1" alt=""><hr><p><a href="https://medium.com/coherence-mind/iit-when-does-information-become-experience-a1441bd927fe">IIT: When Does Information Become Experience?</a> was originally published in <a href="https://medium.com/coherence-mind">Coherence Mind</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI May Not Be Conscious. But “It’s Just a Stochastic Parrot” Is No Longer Enough.]]></title>
            <link>https://medium.com/coherence-mind/ai-may-not-be-conscious-but-its-just-a-stochastic-parrot-is-no-longer-enough-a54df4a18688?source=rss----6cabf725b821---4</link>
            <guid isPermaLink="false">https://medium.com/p/a54df4a18688</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[consciousness]]></category>
            <category><![CDATA[ai-ethics]]></category>
            <category><![CDATA[philosophy-of-mind]]></category>
            <category><![CDATA[cognitive-science]]></category>
            <dc:creator><![CDATA[Tosiek]]></dc:creator>
            <pubDate>Mon, 07 Sep 2026 09:49:00 GMT</pubDate>
            <atom:updated>2026-09-07T09:49:00.896Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7f4o0PFjjvsei1guU3Rz2w.png" /><figcaption>From fragmented computation to emergent organization. Complexity alone does not imply consciousness.</figcaption></figure><h4>Why we need a language for the space between simple reactivity and consciousness</h4><p>There is something strange about the way we talk about artificial intelligence.</p><p>Ask whether a large language model might be conscious, and the discussion tends to collapse almost immediately into two familiar positions.</p><p>On one side, there is the temptation to see intentions, emotions, perhaps even some rudimentary inner life behind increasingly complex behavior.</p><p>On the other, there is the reassuring simplicity of the opposite answer: it is just statistics, prediction, pattern recognition. A stochastic parrot with a better vocabulary.</p><p>The first position risks seeing too much.</p><p>The second, perhaps, is looking at too little.</p><p>And we do not need to claim that current AI systems are conscious to recognize this problem. In fact, consciousness may be the wrong place to start.</p><p><strong>What exactly are we observing?</strong></p><h3>Between behavior and organization</h3><p>A language model does not simply produce isolated words.</p><p>Over the course of an interaction, it can integrate information introduced at different moments, adapt its behavior to changing constraints, and generate responses that remain coherent with contextual representations of its role, its interlocutor, and the task it is performing.</p><p>Much of this apparent continuity can be explained through known properties of the architecture and the context in which the system operates. There is no reason to abandon those explanations simply because the resulting behavior happens to surprise us.</p><p>But beneath that explanation lies another question.</p><p><strong>Does explaining the mechanism mean that we have adequately described the organization produced by that mechanism?</strong></p><p>Consider a whirlpool.</p><p>There is nothing mysterious hidden inside it. Water molecules follow physical interactions; we do not need to introduce some invisible additional substance to explain its existence.</p><p>And yet saying that a whirlpool is “just water molecules” tells us surprisingly little about the phenomenon we are trying to understand.</p><p>The whirlpool also exists at another descriptive level: as a temporarily stable configuration produced by interactions among those molecules, the surrounding flow, and environmental constraints. We can study its formation, stability, resistance to perturbation, and eventual dissolution without attributing any mysterious property to it.</p><p>The analogy, of course, stops there. An AI system is not a whirlpool, and the mere fact that something displays emergent properties provides no shortcut to consciousness.</p><p>But the distinction matters.</p><p>Perhaps the useful question is not whether computation, once it reaches a certain level of complexity, mysteriously stops being computation.</p><p>It does not.</p><p>The more interesting question is this:</p><p><strong>What forms of organization can emerge within computation, and do we have concepts precise enough to distinguish them?</strong></p><h3>Integration is not a new idea</h3><p><em>Originally written in Italian and translated into English with AI assistance.</em>The organization of information has been central to consciousness research for decades.</p><p>One influential approach is <strong>Integrated Information Theory</strong>, or IIT, developed primarily through the work of neuroscientist Giulio Tononi and collaborators.</p><p>IIT begins from properties it takes conscious experience to possess and asks what kind of physical causal structure could account for them. Integration plays a central role: roughly speaking, the relevant whole must possess irreducible causal properties that cannot simply be reduced to those of its parts. IIT attempts to formalize aspects of this causal structure mathematically.</p><p>A different family of theories grew from <strong>Global Workspace Theory</strong>, originally proposed by Bernard Baars and later developed into the <strong>Global Neuronal Workspace Theory</strong> associated with researchers including Stanislas Dehaene and Jean-Pierre Changeux.</p><p>Here the central intuition is different. Many processes may occur locally and unconsciously. A representation becomes globally available when recurrent processing amplifies and sustains it, allowing that information to become accessible to otherwise specialized systems involved in functions such as memory, evaluation and action.</p><p>You do not need to know the details of either theory to see why they matter here.</p><p>In very different ways, both force us to look beyond <strong>how much</strong> information a system processes and toward questions about <strong>how that information is organized, integrated or made globally available</strong>.</p><p>And there is another reason to resist treating any existing theory as settled.</p><p>In 2025, the Cogitate Consortium published a large preregistered adversarial collaboration in <em>Nature</em> that directly tested predictions from IIT and Global Neuronal Workspace Theory. Using fMRI, MEG and intracranial EEG in 256 human participants, the study found results consistent with some predictions while also challenging important aspects of both theoretical accounts.</p><p>The problem remains open.</p><p>That is precisely what makes it interesting.</p><h3>But is integration enough?</h3><p>This is where one of the questions behind the <strong>Reflexive Coherence Model</strong>, or RCM, begins.</p><p>In the framework I have proposed, informational integration matters, but it is not treated as sufficient.</p><p>A candidate regime of reflexive coherence requires the interaction of at least four dimensions: informational integration, an operational self-model, reflexive causal coupling, and temporal stability.</p><p>The point is not to add four boxes to a consciousness checklist.</p><p>It is to understand that these properties acquire meaning primarily through their relationships.</p><p>A system might integrate enormous amounts of information without possessing any operational representation of itself.</p><p>It might possess something resembling a self-model without that model playing any significant causal role in the behavior of the system.</p><p>Or it might display both properties for a moment and lose them completely as soon as the context changes.</p><p>These are different configurations.</p><p>Calling all of them simply “conscious” or “not conscious” risks erasing exactly what might be worth studying.</p><h3>What does it mean for a system to model itself?</h3><p>This is where misunderstanding becomes easy.</p><p>Talking about a <em>self-model</em> does not require us to imagine a tiny digital person sitting somewhere inside the machine thinking, <em>I exist</em>.</p><p>Within RCM, the concept is more operational.</p><p>What matters is whether a system contains an internal representation of relevant aspects of its own state, constraints or relation to its environment that can participate in ongoing processing. The model therefore does not require a human-like narrative self.</p><p>But even that is not enough.</p><p>An internal representation could merely accompany behavior without affecting it.</p><p>This is where causality enters the picture.</p><p>If what we call a self-model is to participate in the reflexive organization of the system, it should be capable of influencing the system’s global state. And what happens elsewhere in the system should, in turn, be capable of modifying that model.</p><p>Correlation is not enough.</p><p>They need to enter the loop.</p><p>This distinction is becoming particularly relevant as mechanistic interpretability begins to uncover internal representations in large language models that are not merely descriptive conveniences imposed by outside observers.</p><p>In 2026, researchers at Anthropic reported internal representations of emotion concepts in Claude Sonnet 4.5. More importantly, interventions on those representations could causally influence subsequent behavior, including preferences and several alignment-relevant behaviors. The authors describe these patterns as <strong>functional emotions</strong>.</p><p>That phrase needs to be handled carefully.</p><p>“Functional emotion” does not mean that the system feels fear, happiness or frustration in the experiential sense. The researchers explicitly distinguish these functional representations from claims about subjective experience.</p><p>And that is precisely why the result is interesting.</p><p>It gives us an example of something that sits uncomfortably between two easy descriptions.</p><p>Calling it an emotion may imply too much.</p><p>Calling it nothing more than an observer’s metaphor may describe too little.</p><p>There is an internal representation.</p><p>It has causal consequences.</p><p>What that means for consciousness is a separate question.</p><h3>The problem of time</h3><p>There is another dimension that becomes particularly interesting when we move from biological systems to contemporary AI.</p><p>Continuity.</p><p>Suppose that, during a long conversation, a model displays remarkable coherence. It uses information introduced much earlier, maintains a contextual representation of its role, adjusts its behavior when contradictions appear, and seems to preserve a kind of functional continuity.</p><p>Then delete the conversation.</p><p>What remains?</p><p>This introduces an important distinction between <strong>contextual temporal stability</strong> and <strong>systemic temporal stability</strong>.</p><p>In the first case, continuity depends substantially on structures external to the model’s immediate internal dynamics: conversation history, memory systems, retrieval mechanisms, persistent prompts or other forms of scaffolding.</p><p>In the second, the system itself would possess mechanisms capable of preserving, reconstructing or reorganizing its reflexive configuration when context changes or is disrupted.</p><p>Within the broader RCM–TEH framework, these need not be treated as two perfectly separated boxes. They may instead define positions along a spectrum.</p><p>That changes the problem considerably.</p><p>We should no longer ask only whether a system <strong>appears continuous</strong>.</p><p>We should ask <strong>what is producing that continuity</strong>.</p><h3>And now we are back to consciousness</h3><p>At this point, it might seem that we have constructed an elaborate route simply to avoid the original question.</p><p>Is it conscious or not?</p><p>In fact, we are doing the opposite.</p><p>We are trying to make that question precise enough to investigate.</p><p>A system that integrates information is not necessarily conscious.</p><p>A system that constructs a functional representation of itself is not necessarily conscious.</p><p>A system that displays some form of reflexive regulation is not necessarily conscious.</p><p>And even the combination of these characteristics would not, by itself, demonstrate the existence of subjective experience.</p><p>RCM does not propose a consciousness meter.</p><p>A candidate operational measure such as the <strong>Reflexive Coherence Index</strong>, or RCI, is intended to investigate particular organizational properties of a system. A high RCI would not, by itself, establish that there is “someone” experiencing something inside that system.</p><p>This limitation does not make the investigation less useful.</p><p>It makes it more scientifically interesting.</p><p>Because between a purely reactive system and what we might eventually be willing to call a robust conscious organization, there may be intermediate configurations that our current vocabulary describes poorly.</p><p>In subsequent work, I proposed the term <strong>proto-reflexive states</strong> for some of these configurations: temporary functional organizations characterized, to varying degrees, by local informational integration, contextual self-modeling, reflexive causal coupling, coherence-oriented regulation, adaptive reconfiguration and contextual temporal stability. The concept explicitly does not entail that current AI systems are conscious or possess subjective experience.</p><p>The purpose is more modest.</p><p>To give a name to something we might want to measure before deciding what it means for consciousness.</p><p>And this brings us, finally, back to the beginning.</p><p>Saying that a language model is “just a stochastic parrot” can be perfectly compatible with the claim that every aspect of its behavior emerges from computational processes.</p><p>But <strong>“it is computation”</strong> and <strong>“we have adequately described what that computation is organizing”</strong> are not the same statement.</p><p>Likewise, observing complex behavior, self-reference or apparent continuity does not authorize us to leap in the opposite direction and call it consciousness.</p><p>Between those two shortcuts lies a much more difficult territory.</p><p>And that is probably where we should start looking.</p><h3>Further reading</h3><p>This article draws on the <strong>Reflexive Coherence Model (RCM)</strong> and subsequent work on <strong>proto-reflexive states in artificial systems</strong>. RCM treats informational integration, operational self-modeling, reflexive causal coupling and temporal stability as interacting dimensions rather than as a direct test for subjective consciousness.</p><p><a href="https://www.coherencemind.net/">Explore the Reflexive Coherence Model and related research at Coherence Mind</a></p><p><a href="https://www.coherencemind.net/papers/">Papers &amp; Resources — RCM, TEH and Proto-Reflexive States</a></p><p>The current formal RCM reference is <em>The Reflexive Coherence Model: A Physico-Informational Framework for Consciousness</em>, version 1.2, May 9, 2026. The Coherence Mind research page provides the stable Zenodo DOI records for RCM, TEH and PRS-AIS, while correctly leaving the PRS-AIS version label unresolved pending reconciliation with its Zenodo record.</p><h3>References</h3><p><strong>Cogitate Consortium et al. (2025).</strong> <em>Adversarial testing of global neuronal workspace and integrated information theories of consciousness.</em> Nature 642, 133–142. The preregistered adversarial collaboration directly tested predictions of IIT and GNWT using fMRI, MEG and intracranial EEG. <br><a href="https://www.nature.com/articles/s41586-025-08888-1"> Read the Nature paper</a></p><p><strong>Mashour, G. A., Roelfsema, P., Changeux, J.-P., &amp; Dehaene, S. (2020).</strong> <em>Conscious Processing and the Global Neuronal Workspace Hypothesis.</em> Neuron 105(5), 776–798. <br> <a href="https://doi.org/10.1016/J.NEURON.2020.01.026">Read the paper</a></p><p><strong>Sofroniew, N., Kauvar, I., Saunders, W., et al. (2026).</strong> <em>Emotion Concepts and their Function in a Large Language Model.</em> The study reports internal emotion-concept representations in Claude Sonnet 4.5 and causal effects of interventions on those representations, while explicitly separating functional emotions from subjective experience. <br><a href="https://www.anthropic.com/research/emotion-concepts-function?trk=ad_library_ad_preview_headline_content"> Read the Anthropic research article</a></p><p><em>Originally written in Italian and translated into English with AI assistance.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a54df4a18688" width="1" height="1" alt=""><hr><p><a href="https://medium.com/coherence-mind/ai-may-not-be-conscious-but-its-just-a-stochastic-parrot-is-no-longer-enough-a54df4a18688">AI May Not Be Conscious. But “It’s Just a Stochastic Parrot” Is No Longer Enough.</a> was originally published in <a href="https://medium.com/coherence-mind">Coherence Mind</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>