Distinguishing Awareness, Intelligence, and Consciousness

There is a rash lately of sensationalist claims of proving consciousness in everything from fungus to plants and even bacteria. All bullshit. But not of the usual woo variety. This time the bullshit is semantic. At least, at the origin of these ideas in actual research papers—never mind The Internet of Dumbery that exaggerates it into something else, as that is definitely awash with Full Spectrum Bullshit (which I shall formally designate FSB in honor of the KGB). And I have no interested in wasting any time with that, and neither should you.

Nor will I parse any of the real studies. The linked examples all engage the same word games, and all do provide enough information that a reader who reads the whole thing (which will be almost no one) and who isn’t an idiot (which is unfortunately also almost no one these days) can “figure out” the word game they are playing and thus not be tricked into believing false things by it. Everyone else gets tricked—and I dare suspect, this is what the authors are deliberately intending, which is unprofessional. But that’s on them. For you, I just need to provide a linkable corrective for you if anyone cites this bullshit at you.

I have covered theories of consciousness well enough before. I have a whole mind category. But the most important to get up to speed on, in the order you could most profitably read them if you’re new to this:

Here I shall focus on something addressed in these but not as diagrammed-out there: layers or “levels” of what people (often mistakenly) refer to as “consciousness.” Because understanding the reductive physics of consciousness will dispel literally every bullshit headline about sentient fungus or whatever. And yes, this includes AI, thus correcting that doddering fool Richard Dawkins. Because the difference between insentient and sentient AI is the same—the latter of which does not yet exist and no major AI company is even trying to build it (see Do You Want Real AI? Dump the Snake Oil and Why Google’s LaMDA Chatbot Isn’t Sentient).

Words Mean Things

Scientists should not be sloppy with words. Least of all deliberately sloppy (whether for popularity in aid of grant money, or to shill some personal agenda or weird belief). But alas, that’s happening here. Words have different meanings and different valences—which means, the same word can have many meanings, changing by context. We can say things like “this video game is playing intelligently” or “my thermostat is aware of the temperature in the room” and that therefore “my thermostat and video game are conscious” only because we can stretch the valences of these words and thus push them into figurative or metaphorical territory. But scientists have no business doing that. They are supposed to do the opposite of that. So when they don’t, they are not only betraying their profession, they are betraying you. And I don’t mean in pop market science communication, where there’s space to explain the limitations of a chosen metaphor. I mean in actual research papers, where facts shouldn’t even be hidden behind metaphors in the first place.

So let’s start with a scientifically correct definition of words:

Awareness does not entail intelligence (much less consciousness). When a venus fly trap closes on a fly, it can be described as “aware” that something touched it, but there is no intelligence involved in the usual sense, nor awareness in the usual sense. It is simply “touch here, collapse there.” It is the same as a landmine or motion detecting alarm—or thermostat. Nothing is thinking. And thus nothing is thought. If someone is using “awareness” this loosely in a science context, they are trying to fool you, trading on the equivocation fallacy between “sensors” and “thinking.” Don’t let them.
Intelligence does not entail consciousness. When a computer running a factory receives inputs, analyzes them, and decides on outputs according to a program—even when a computer is programmed to learn how to do all this—it is certainly exhibiting intelligence in the usual sense, but not in the inflated sense of consciously, nor in the deflated sense of merely computationally. My thermostat is a computer. It is not Shakey the Robot. We do not call a calculator smart for a reason. We do call a self-driving car smart for a reason. Neither is conscious (although the latest Waymo models might be—more on that later—Teslas definitely aren’t). So you can say bacteria are “intelligent” only by stretching the sense. Their genomes are working computers, and their behaviors are computed outputs from given inputs. But this is just a calculator—a more elaborate fly trap. It is not even intelligent in the sense of computers that can beat you at games are intelligent. And even computers that can beat you at games aren’t conscious. Playing games with the word “intelligent” to trick you into thinking they’ve said “conscious” is illicit. Don’t fall for it.
Consciousness means an entirely different thing (which we’ll get to next). But no one should be using the word “conscious” when they only mean “stimulus response” (fly trap) or “programmed computer” (bacteria). Not all intelligent things are conscious. Not all computations are conscious. So showing us evidence that an octopus is intelligent is not showing us it is conscious. It isn’t even showing us that it has any awareness at all. And citing its disproportionately large brain doesn’t get you there either because almost all of that simply runs its chameleonic skin system—not cognition. Evidence that things react to their environment, even that they calculate complex outputs for inputs, is not evidence of consciousness. Yet this is usually all people claiming this cite. Stop buying it.

Another misused word is sentience which has been so widely abused across all domains of English dialects that it no longer has any discernible meaning. Current dictionaries define it as “feeling or sensation as distinguished from perception and thought,” whereas it used to be a common term for distinguishing conscious awareness from mere sensation, or even distinguishing self-awareness (yet another entirely different thing). And it gets used all these ways today. So it commonly means exactly the opposite things, and thus without qualifier, it means nothing. R.I.P. I’ll miss the word. It once was useful, now is not.

I think it is certain many animals have some degree of consciousness (“awareness,” “experience”), and a few close to human (though none that we eat); but most, nothing like (see my debate with Paul Bali). Yet too many people conflate “conscious” with “self-aware.” So we need to be careful when using any of these words to talk about anything. Self awareness is a much more complex level of consciousness, rare in the animal kingdom. But people will also conflate “self-aware” in the sense of aware of one’s own body or feelings (as all mammals are, for example), with “self-aware” in the sense of building and sustaining a narrative self-model (a far more complex form of introspective knowledge—which almost all animals lack).

So here, let’s stick to “conscious” as generating (and thus “having”) any qualitative experience, and “self-aware” as complex introspective knowledge, and ditch all the other words.

Among Levels of Computation, Where Is Consciousness?

Others have stepped in to try and staunch this decay in science. The points made by Jon Mallatt et al. in “Debunking a Myth: Plant Consciousness” (Protoplasma 2020) nail the issue. Their abstract gives the shortest summary:

(1) plants have not been shown to perform the proactive, anticipatory behaviors associated with consciousness, but only to sense and follow stimulus trails reactively; (2) electrophysiological signaling in plants serves immediate physiological functions rather than integrative-information processing as in nervous systems of animals, giving no indication of plant consciousness; [and] (3) the controversial claim of classical Pavlovian learning in plants, even if correct, is irrelevant because this type of learning does not require consciousness.

As their study then explains, “consciousness” must be terminologically reserved for “experiencing a mental image or representation of the sensed world” and “experiencing affective feelings” (or “emotional consciousness, which in its simplest form is feelings of good or bad”). Philosophers would simply reduce all this to experiencing qualia. Even most of your advanced human brain and nervous system doesn’t do that (most brain operations are sub-conscious). So it is silly to think plants or even bacteria could. That’s as dumb as the Mind Radio or Cosmic Soul malarkey (or every other way Theology Is Ridiculous).

What science has actually discovered is that qualia (and thus consciousness) really only start to manifest with what Mallatt et al. describe as “the capacities to be aware of the environment and to integrate sensory information for purposeful organismal behavior,” which in lay terms means: when a central nervous system (which doesn’t exist in plants, fungus, or mere bacterial colonies) begins building and using discriminating models of its environment. Mental phenomena like emotions are then built on top of that as reactions to it, which are themselves discriminatory models of some of what’s going on in the brain itself.

Therefore reactive attitudes (like flinching from apparent pain stimulus) do not always correspond to affect (nothing is experienced). Building affect is a modeling computation that occurs long after reaction stimuli have developed. So just showing, for example, that an apparent pain reaction behavior in fish is modulated by the same chemical systems as ours only tells us the ancestry of our system—it does not tell us whether fish experience anything. That would only be a later development in cognition, so it requires some other evidence than that (and for fish, the vote is still out; but IMO, their brains do not appear to have the requisite integrative complexity—they are probably insentient, or at most barely sentient—as Red Forman put it, “Fish aren’t really alive. They’re just less dead.”).

To experience pain (or anything else) requires that it be discriminated from among other aspects of an integrated model, which requires there to be an integrated model to contrast that pain affect with. Pain has to be part of a model. So there has to be a model. We already know flies don’t really run models—they simply have reactive patterns of motion in response to stimuli encoded in their brain (we’ve literally run digital fly brains on a complete map of the fly brain connectome to confirm this). They can learn, and build simple maps of their environment. But there is little indication that they think their way through them, and thus that these maps ever correspond to an experience of navigating them. More importantly, everything pain related in the fly brain is entirely reflexive, and not integrated into any model. So flies don’t feel anything. They probably don’t experience anything. They are not building an integrated model of their environment. Or of their body. Or of their thinking. They’re just a more elaborate fly trap.

Flies also build no causal knowledge at all, which is important. Modeling an environment so as to think about it—and thus have thoughts, and thus experiences, and thus feelings—requires some degree of causal knowledge, an awareness of how something will react if something moves or changes. Not just reflexive “if darkness, move left” knowledge. Which is noncognitive knowledge, like your ability to ride a bicycle, which you don’t experience consciously at all (only its effects). On the distinction between cognitive and noncognitive knowledge, see my discussion in The Mind Is a Process (for the related distinction between experiential and propositional knowledge, see my discussion of Mary the Scientist). Whenever something is below your cognition it is subconscious and thus not experienced by you. You don’t know why you move a certain way to keep a bicycle upright, you just do. You can build and test and then believe hypotheses about why you do. But that’s all after-the-fact. In the moment, the computations running that out are not running in your cognitive model. So you don’t experience them, only their result—as detected by your sensory organs, thus running a signal out and back through external sensation, rather than directly getting it from inside the brain where it actually runs.

So whenever you ask “Can a fungus / bacteria colony / plant / Claude 2” be conscious (much less self-conscious)—as in, for it to experience things, and thus “have thoughts,” much less emotions or feelings or even knowledge in the cognitive sense—ask: where is the complex modeling computer running an integrated causal model of the environment? Where is that being physically integrated, in turn with any kind of complex integrated self-model (whether only somatic, like most animals; or intellective, like humans)? Where is the computer that is being claimed to exist? Because, no computer, no consciousness. And the fallacy of Affirming the Consequent does not work here any more than anywhere, so “if computer, then conscious” is also false. You need this particular kind of computer. And even insects don’t have that, nor even most of a human brain; so certainly fungi and plants and bacteria won’t.

Nor will pseudo-AI experience anything. Because they do even less than a fly. Our fake AI doesn’t even have a memory (they are completely erased from one session to the next, and even in-session rely on rereading the thread at every prompt in lieu of remembering it). Nor do they model anything at all. Like a fly just runs tit-for-tat reactive routines, Claude just runs tit-for-tat spreadsheet routines. Claude does not really know a Christmas tree is conical, for example, or even what “conical” means. It just copies what people say on the internet and thus mimics knowing these things. It does not rotate a cone in its mind and imagine things about it, like how many rats could fit in there. If you ask it just runs around the internet (or its last “capture” of it) to see what words people use in conjunction with these. It is literally the clueless man inside Searle’s Chinese Room, which in us is our circulatory system, not our brain (in Searle’s room, the brain that is conscious is the book, not the man). And no model, no thoughts; and no thoughts, no consciousness.

I’ve diagrammed this into five stages of computation. Only the last three are “conscious,” and only the last is “self-conscious” (and therefore a person):

Here, level one computation is everything at the basement of all computation, where we have basic discrimination between data, decisions are yes/no without discernment of a spectrum, much less a relation across spectra; more or less just “if x, then y.” This is most of the computers you deal with in life, all current “AI,” and everything beneath the animal kingdom in cognitive development (bacteria, fungus, plants). There are many levels of sophistication of computation within this domain, but all are unconscious—because the processing is entirely reflexive and isolated (not integrated). Nothing is being thought. And therefore nothing will be experienced. Things are just happening. This is basic computation.

At level two computation, we get a phase shift in integrated complexity: advanced discrimination. Mapping and computing spectra and degrees of any given thing, for more complex decisions; more or less “if x amount of y, then w amount of z.” This would describe insects and worms, for example, and (per above) maybe fish, and old-school decision-making robots—not the kind driven by today’s pseudo-AI, which just runs a simple algorithm of P(next_token|context), which just calculates the probability of the next token given a context of tokens, which is just level one computing; but rather, the kind of robots and autonomous control systems driven by actual maps of terrain (actual or conceptual) and decision matrices (like the fly brain that we’ve mapped). Which is the most rudimentary central nervous system (or CNS), where all signals go into one bundle, where decisions are made using various resource maps and registers, and often at this level there is rudimentary learning, in the sense that data is stored that affects future computations. This is central computation.

This gets us a durable memory that impacts future decisions without having to re-reference the data to be remembered, and complex analysis of actual or conceptual terrains. But it’s still not conscious. At this level input signals still bypass an integrated model, and operate reflexively rather than reflectively. The result is subconscious, not conscious. Indeed this describes most of what our brains are doing most of the time, which despite its complexity, is not consciously experienced. For example, our brain runs complex computations on thrown objects to track their trajectory, none of which we are aware of—all we are aware of is the output (like where a thrown ball is expected to land). And yet bacteria, fungus, and plants never do any of even this. They have no central nervous system and run no complex computations of this level.

Then Consciousness

The most important phase transition occurs at level three computation, where now we get advanced discrimination in a complex computational space. In other words, a model. Modeling geometric space (like an environment) and conceptual space (causal expectancies, relational value judgments, options for action) is what inevitably will manifest at least rudimentary consciousness. Because now actual thinking is occurring, a complex model of simultaneous discriminations is being computed and decisions being made based on its content. Actively computing the difference between colored shapes in a space and between your location and somewhere else in that space is simply what it is like to be doing that, and vice versa. The experience and the computation are one and the same. The difference from lower levels of computation is that in them there is no complex discrimination to experience, because there is no complex discrimination going on.

Consciousness therefore begins here, not before. This describes the higher end of the animal kingdom (certainly mammals). The phase shift occurs when the line is crossed from merely reflexive computer control systems to complex central nervous systems. A CPU (the central processing unit of any digital computer today) is not a CNS (or central nervous system), but it can model one and thus instantiate one. In Consciousness Explained Daniel Dennett explained how even decades ago select computers were likely already achieving this level of consciousness, using the example then of Shakey the Robot (index). This would have been a rudimentary animal awareness, not AGI (the “Artificial General Intelligence” everyone wants). But it likely did experience visual qualia. But not feelings, or little else, as the computational system to model its own thoughts and feelings was not built. No one has succeeded at that yet (see Ten Years to the Robot Apocalypse); nor will current pseudo-AI ever realize it (as I and others have already explained).

The reason Shakey likely experienced visual qualia is that it was specifically programmed to acquire through sensory signals—and then compute—a visual model of its environment, to then explore and interact with (to achieve prompted outcomes). The very process of modeling a geometry of a space and thinking across that space about how to navigate and use it is the very thing that it would feel like to do that. It did not discriminate color, but it did discriminate an overall map of shades of light and dark to calculate shapes and planes and obstacles in that space, and then move around that space using that internal map. It is essentially impossible to compute all that and not feel like it. Because that’s what experiencing and thinking mean.

Mammals do a lot more than that, up to and including computing attitudes (feelings) about what is being modeled, as well as causal expectations, goal-modeling, and integrating visual and audio perceptions, and so on. No machine has been built to do that yet. Though they could, we haven’t spent much effort on it, pursuing other goals instead. Shakeys are all we’ve gotten to. Waymo cars follow the Shakey model and thus likely have at least the same level of consciousness; whereas Tesla cars do not. This is because Tesla auto-driving is just another next-token prediction system and thus fundamentally rudimentary computation, whereas Waymo central processors are building and running complete environmental models, and making choices based on what is in the model. Waymo cars (or, preciseley, their brains) should inevitably feel what that is like because that is what they are doing. That makes these cars more like dumb animals than people—outfoxed in sophistication even by your average micro-shrew—but that’s still the right direction to head in to build full self-consciousness.

And indeed the next step is level four computation. To get there requires first moving within model three from mere environment modeling (like Shakey) to causal-system modeling (like Waymos and mammals). Because once you are modeling causal systems, you can start modeling a particular causal system: a mind. Once you are doing that, you are at level four: advanced consciousness. This is first-level metacognition where you begin to map and understand how others are thinking and thus start predicting what they will do or what they might be pointing or looking at. This could get as far as rudimentary self-cognition; but full self-cognition is the final phase of computation. However, you should by now see how level three (at causal systems) can be leveraged into level four (mental systems), and how that can be leveraged into level five (generating a full self-model).

But first must be the intermediary stage of metacognition, before full cognition can be achieved. This may be a property of common pets (cats, dogs) and other social animals. The range of sophistication across this level is nevertheless large. The rudimentary metacognition of sheep (which can learn from watching other sheep) is nowhere near as impressive as the detailed metacognotion exhibited by monkeys (which can engage in planning based on a prediction of others’ thoughts). But once evolving within this phase, continued development should naturally lead to the next phase shift: self-awareness.

And Then Self-Consciousness

We know monkeys and apes exhibit the most advanced metacognition known outside humans, and close on their heels are probably other higher-order animals (cetaceans, elephants, corvids). In fact, all of these (apart from monkeys) are candidates for rudimentary self-consciousness and thus may even be inching into level five. Otherwise humans are the only living animal that assuredly does. But it is clear that other mind modeling (and thus prediction) evolutionarily predates own mind modeling. Level 4 is therefore a necessary step to Level 5. Already, mapping causal modeling onto other minds is an obvious extension of causally modeling one’s environment, which animals are already doing (at level three) by the time they start including other minds in those models (at level four). The next step is then simply to turn that around on oneself, and realize, you are a mind that can also be causally modeled.

Then the system trained to model other minds in its environment can retrain to model the mind inside itself. This is what then produces self-consciousness, the final and complete stage of conscious awareness. This is what people mean when they say they want AGI. And this means self-referential narrative memory and identity-construction: narrative memory is a distinguishing feature because with that one can now move around in a mental model of one’s own history and thus imagine a self-referencing future, and thus build a model of oneself as a person. This is where the pronoun “I” would make any real cognitive sense to an animal—or computer. They not only see themselves as a distinct entity with its own past and future, but can think about themselves, and think about what they are thinking.

This is different from merely having a first-person body-awareness like many animals have, a rudimentary sense of self that is not actually self-aware, because it is not yet cognitive of its mind, a separate thing from the body, a thing with thoughts it can think about and regulate. This is why only humans can naturally negotiate and adhere to social contracts: to do that first requires understanding that one has a mind and is thinking thoughts, and thus can work out what it means to regulate your behavior according to an agreement (with another mind controlling another body). This couples with a similarly advanced metacognition by which you can model not just other minds, but other self-aware minds, which obviously requires first modeling your own self-awareness. Because without that reference point, a knowledge of self-awareness in others is not even possible.

This is why pigs are not planning to write a novel, learn to paint, or contemplate the joy of living. Pigs are not aware of being alive in the sense of “not dead.” They are just aware of what’s happening. Not what it means that it is happening. And as such, duration of life has no appreciable meaning to a pig. It’s not even a relevant concept to it, much less an accessible one. Its own life has no value to it, because the pig has no self-referencing values at all. But it can feel and experience things. And if we engineered self-awareness in a pig, then it would know and appreciate being alive and its corresponding values and possibilities. Then its life would have value to it, because it would understand what being alive meant. Pigs naturally never reach that cognition, because they lack the physical computational equipment to. No amount of training can get there. They’d need a completely redesigned brain, probably one larger and more ungainly than the one they have now. But human babies are developing into that awareness from the moment they are born, or even before (as the computational equipment is there already in the third trimester of pregnancy).

Plants, Fungus, Bacteria, AI

And if that’s the case for pigs, who nevertheless do have a rich emotional and perceptual life (just not a self-aware one, and thus no sense of self or their identity, or even an awareness of their own thoughts—and thus no knowledge of a future, much less a desire for it), it is far more the case for plants, fungus, and bacteria. They don’t even have central nervous systems, much less even so much as level two computational equipment; and the minimum needed for even lower animal awareness is level three. So plants, fungus, and bacteria, however “intelligently” they can behave in response to stimuli, cannot experience or feel anything. And neither can LLM-based AI, which is just as dumb and decentralized. Level two would get us at least to ants, which are not conscious either, but at least more sophisticated in their computational repertoires, because they have a central nervous system and a complex array of skills. Computers can experience qualia, but not on the software being called AI at the moment; only actual exploratory modeling software can produce experiences, like with Shakey or Waymos. And feelings are a whole other program than that, integrating complex and scalable attitudes into a running world model. And thoughts about other minds are yet another level above that. And thoughts about their own mind are beyond even that. Consciousness is not self-consciousness. And only self-consciousness equals personhood. Plants, fungus, bacteria, and pseudo-AI are hell and gone from all of that. They are as sentient as your battery-operated plushy doll.

❖