Hide table of contents

Executive Summary

Rethink Priorities has been compiling and analyzing relevant scientific evidence regarding invertebrate consciousness. As explained in a previous post, we selected 53 features potentially indicative of the capacity for valenced experience as criteria for discerning whether invertebrates of 12 biological taxa are conscious or not (see database). However, if a specific feature can effectively operate non-consciously in humans, that function –by itself– may provide no evidence that an individual is conscious. Hence, we investigated which of those 53 features can occur unconsciously in humans. We observe that different noxious stimuli reactions and some simple forms of learning are the most likely to occur unconsciously in humans. These and other findings are presented and discussed in this report.


Let's say you accidentally touch a hot stove. Instantly, you will withdraw your hand from the stove and will feel pain. A simple circuit of neurons saved your hand from a burn: an excited afferent neuron that stimulates excitatory interneurons that in turn stimulate the efferent motor neurons supplying the biceps muscle in your arm. Thanks to them and without even thinking, your biceps contracts automatically, pulling your hand away from the hot stove.

Although you feel the pain almost simultaneously, your painful experience is more complex than that, and goes beyond the withdrawal reflex. When the afferent neuron stimulates still other interneurons, they carry the signal up the spinal cord to your brain (see Waterhouse & Campbell, 2008). Only when this impulse reaches the sensory area of your cortex are you aware of the pain, its location, and the type of stimulus. Only when the impulse reaches the brain, can you wonder how this happened and start thinking about moving the stove to a safer place.

Given that pulling your hand away from the stove is, per se, a pure reflex it occurs in humans without an attendant phenomenology. In fact, in comatose human patients, noxious stimulation of the lower extremity (produced by pinching the skin, for example) may produce a local withdrawal reflex with no purpose (Gelb, 2010). Hence, withdrawing the hand or any part of the body from a noxious stimulus, by itself, is unlikely to be a good indicator of conscious pain. For that reason, when assessing pain in nonhuman animals, it is considered that reflex responses to noxious stimuli provide information about nociceptive processing, but not about pain. In this regard, tests about more complex behaviors are needed (National Research Council, 2010).

The major implication of all the above is that, despite certain features are potentially relevant for consciousness, they –by themselves– may provide no evidence that an individual is conscious. In other words, if certain functions can effectively appear non-consciously in humans, that might be a good defeater for those features when assessing consciousness in other non-human individuals (for further discussion, see this prior post).

As explained in a previous post, we selected 53 features potentially indicative of the capacity for valenced experience as criteria for discerning whether invertebrates of 12 biological taxa are conscious or not. In addition, we investigated which of those features can occur unconsciously in humans. Those findings have been gathered and displayed in this database. Here, we present a narrative report that summarizes our main results. It is organized in five sections: first, we broadly describe the role of unconscious processes in human psychology. That allows us to assess, in a second section, the question as to whether it is possible for functions related to consciousness to operate unconsciously in humans. Third, we consider which potentially consciousness-indicating features appear to operate consciously in humans. Fourth, we explore which functions related to consciousness are especially likely to operate unconsciously in humans. Fifth and finally, we present some concluding remarks about the studied features.

Understanding the role of unconscious processes

The scientific study of consciousness has only received considerable attention during the last decades (Zeman, 2001; Pereira et al., 2010). These efforts have led to the proliferation of different theories of consciousness[1], but none of them has wide acceptance within the scientific community (Earl, 2014; Pereira et al., 2010). Hence, when addressing unconscious processes, we face two major challenges:

  • Lack of consensus and ambiguous definitions of what consciousness is, and as a result, unclear conceptions of unconscious processes and their role (Earl, 2014; Pereira et al., 2010). For example, some authors (e.g. Hassin, 2013) consider that mere automatic actions–even those learned through practice–can be unconscious, while other researchers assume that automaticity is not enough and unconscious processes are rather defined in terms of their unintentional nature (e.g. Bargh & Morsella, 2008). In a similar way, studies on behavioral priming have shifted from defining unconsciousness as “without awareness of the stimuli” to “without awareness of the influence of the stimuli” (Hesselmann & Moors, 2015). Additionally, in several cases, the underlying understanding of what is “unconscious” is not explicit and a definition must be inferred. In other cases the underlying theories for understanding conscious processes are not sufficiently specified. Hence, findings from different sources might not be comparable. When they are comparable, moreover, we may encounter conflicting evidence[2].

These inconsistencies may lead to different criteria about whether the same function or behavior qualifies as unconscious or not. If a concept is not properly specified, what counts or not as evidence and what the evidence points to will be subject to the researcher's interpretation.

  • Methodological constraints: most of the findings in the field of human consciousness are based on subjective reports of the inner states of individuals –like imagery, feelings and inner-speech. However, this approach faces two methodological limitations: first, these internal states cannot be externally observed. Second, our introspective capacity is limited (Hassin, 2013; see also McGovern & Baars, 2007). Furthermore, it must be considered that the use of other techniques based on brain activity may not necessarily yield unambiguous answers, since neural activity by itself is not an indicator of consciousness (Baars & Gage, 2010)[3]. Additionally, as Hesselmann and Moors (2015) highlight, this field of research is not exempt from failed replications; and when it comes to new studies, they haven't been replicated yet.

Although debated, it is mostly agreed that conscious processes involve arousal and awareness of the perceived stimulus (Zeman, 2001). Thus, and for the strict purposes of this work, we understand conscious processes as phenomenological subjective experiences in which the individual is aware of an external object or an inner state (based on van Gulick, 2004). For their part, unconscious processes are those that take place outside of conscious awareness and yet influence a person's experience (Kihlstrom, 1997). Zeman (2001) adds that these ‘unperceived’ stimuli have effects on neural events and subsequent action (see e.g. Frith et al., 1999). In neurophysiology, one of the most widely quoted example of unconscious perception is blindsight[4].

These working definitions do not require adherence to specific theories of (un)consciousness, at least for the strict purposes of this project. In this sense, notice that we refer to conscious/unconscious processes (and not to “the unconscious” as a noun), since we do not necessarily imply that there is a conscious and an unconscious system or mental entity as such–as in the topographical model proposed by Sigmund Freud or Carl Jung.

At present, the existence of unconscious processes –which can be often contrasted with conscious mechanisms– is accepted in cognitive science and in neuroscience (McGovern & Baars, 2007). In general, it is agreed that there are dissociations between conscious awareness and stimulation and/or behavior. That is to say, there is considerable evidence that stimuli can be perceived and processed in the absence of conscious awareness, affecting an individual’s behavior (Cleeremans, 2001).

How’s that? For instance, we know that our mental paradigms—how we categorize things around us—like to lump similar themes and thoughts together. So if we show a subject the word “cat” (prime stimulus), even for a very short duration, and then one of two new words, “dog” or “car”, she will recognize “dog” quicker because brain activation spreads faster among related ideas. Under certain circumstances, a prime stimulus may unintentionally influence a subsequent behavior as well. In a gamble experiment, for example, it was observed that when people were exposed to betting related words they were significantly more likely to bet (Payne et al., 2016). Automatically and without intention or awareness, the exposure to one stimulus can influence the way we respond to another stimulus or aftermath situation (Ortells et al., 2006; Payne et al., 2016).

Various of these priming effects[5] have even been demonstrated when the first stimulus (in our example, the word “cat”) was unattended and presented outside the “spatial focus” of attention (e.g. Ortells et al., 2001) or displayed for a short time of exposure and then masked, in such a way that individuals claim to be unaware of the stimuli’s identity (e.g. Draine & Greenwald, 1998; Forster et al., 1990 in Kihlstrom, 1989; Marcel, 1983). In spite of this evidence, it must be noted that several specific experiments about priming effects (e.g. priming people to think about old age makes you walk slower; Bargh et al., 1996) have failed to reliably replicate[6]. Although there is still a vigorous debate over these findings, it seems that several behaviors and cognitive processes may have conscious and unconscious aspects, and they also may range from completely unconscious to completely conscious, with conscious and unconscious attributes or effects (McGovern & Baars, 2007).

In general, unconscious features (behaviors and cognitions) can take the form of automaticity (or ‘automatic behaviors’), implicit memory, implicit perception, implicit thought and language, and implicit learning (Kihlstrom, 1989; Zeman, 2001). A good example of the above is the way we learn our native language. We identify, learn and remember its complex and subtle regularities without even realising. For children, such implicit language learning seems to happen spontaneously in the first few years of life. Yet, when adults, learning a second language is generally far from effortless. Additionally, other authors include forms of unconscious evaluation and motivation (e.g. Bargh & Morsella, 2008) and even unconscious decision-making processes (e.g. Soon et al., 2008).

The above mentioned processes are mostly considered ‘unconscious’ due to two characteristics: their unintentional nature and automaticity (e.g. Bargh & Morsella, 2008; Lieberman, 2007; Morsella & Poehlman, 2013). However, not every automatic or unintentional action is unconscious per se. For example, riding a bike or driving a car may be automatic actions for someone familiar with these tasks, but these are not a proper unconscious process as they imply an intention (see McGovern & Baars, 2007). Hence, if someone goes to work everyday by car, she probably remembers both leaving home and arriving at work. But she probably may not recall the details about the drive in-between, when she turned the lights on or off, or how many times she stopped. She has driven for so many years, that after continuous practice, driving becomes automatic for her. Likewise, if we are learning how to play the guitar, during our first trials, we must think carefully about what each finger is doing. However, after practice, playing the guitar will eventually become automatic –at least, if we are playing a piece we know. In fact, automatic skills of this type, which may even require fast and highly organized muscle control, start out under cortical control, but after sufficient practice they tend to come under subcortical control (Haier et al., 1992; Chein & Schneider, 2005 in Baars & Gage, 2010).

Hence, when we find ourselves automatically playing the guitar, driving a car, riding a bike or walking, this does not mean that these actions are unconscious. They might be automatic behaviors, but we are not zombies at the wheel. We know why and where we are driving the car. We can change our route or destination. They are intentional behaviors and their automaticity results from overlearning ––a learned skill that becomes automatic after continuous training (see Shibata et al., 2017). Thus, after constant practice, we can perform these tasks without dedicating much attention to them[7]. Hence, not every automatic behavior is necessarily an unconscious action.

Besides overlearning for automatic behaviors, what is the explanation for unconscious processes? Why do they remain outside of consciousness? First, it must be considered that consciousness has a very limited processing capacity (Baddeley, 2007; Kahneman, 1973). That entails that an individual cannot consciously experience all the stimuli and information available at any one time. Therefore, that information may be relegated to unconscious processing. Taken together, unconscious processes can operate concurrently, have a greater processing capacity and are involved in all cognitive functions (McGovern & Baars, 2007; Baars & Gage, 2010)[8].

Second, although humans have the capacity to process information consciously, this often operates non-consciously, which would suggest that conscious and unconscious information processing are different strategies which may be more or less efficient for different tasks (Keller, 2014). Hassin (2013) argues that since conscious awareness, as humans experience it today, might be a relatively recent development (e.g. Dennett, 1991)[9], and given how evolution works, “cognitive functions are likely to occur outside of conscious awareness” (195). Hence, a new question arises: which kind of stimuli or tasks are more efficiently processed using unconscious mechanisms?

Answering this complex question exceeds the purposes of this document. Nevertheless, assessing whether numerous behavioral and neurobiological/architectural features associated to consciousness may occur unconsciously can shed some light on the subject. More importantly, if certain functions can appear non-consciously in humans, that might inform which features, by themselves, are not sufficient when assessing consciousness in other non-human individuals –which is the issue at hand.

Can some functions related to consciousness operate unconsciously in humans?

Brief answer: yes, they can (which is different from “yes, they do”). Some features can be performed unconsciously, but not always, probably not all of them and not totally. Here is an overview of our findings:

  • Some sophisticated information processing functions can be performed unconsciously: “Sophisticated” or high-level functions refer to cognitive processes that arise after perception and require what are usually considered complex computations and cognitive control (Diamond, 2013). Some examples are inhibition, decision-making, problem-solving, self-control, planning, executive functions and working memory. Why? Let’s consider the case of decision-making. Usually, taking a decision requires considering different options, weighing their benefits and costs, and coping with their implications: Having a burger or a salad? Quitting your job or not? You need to form an opinion and deliberate. Hence, making a decision, like other executive functions, requires some effort, or at least, paying conscious attention. But is it always like that?

In a study, thirsty and non-thirsty participants were exposed to a brand of ice tea or a neutral word. The words and the brand flashed before them for a few thousandths of a second, too quickly to be consciously perceived. However, some thousandths of a second were sufficient for reading the words and influencing the participants’ behavior: when choosing a drink, those thirsty individuals who were exposed to the ice tea brand preferred that brand (Karremans et al., 2006).

In recent years, new research has challenged the conventional view that decision-making and goal pursuit would necessarily occur consciously. According to Chartrand & Bargh (1996 in Chartrand et al., 2008) “goals can be activated by situational cues and can influence behavior outside of awareness until the desired outcome has been attained”. However, that will not happen under any circumstances. For example, note that in the ice tea experiment the effect of the non-consciously perceived brand only affected thirsty individuals. As Karremans et al. (2006) highlight, subliminal stimuli can affect behavior only when a person has the potential to perform a specific action and she is in the right context to perform it (see also Bargh et al., 1996). Notwithstanding the above, there seems to be no scientific consensus on the matter. Chartrand et al. (2008) clarify that further research is still needed in this field.

Language processing, for its part, is a complex cognitive process as well. But if you are told not to read the following sentence, you will probably fail: “Don't read this”. Did you read it? Presumably, yes. For single words or short sentences, reading is an automatic and non-intentional phenomenon that occurs without any voluntary effort–if you are a literate person, of course–and even outside of our conscious awareness, as the ice tea experiment and other studies suggest (see e.g. Forster et al., 1990 in Kihlstrom, 1989; Karremans et al., 2006; Marcel, 1983). This simple example points out that we do not have full control even of a complex cognitive process like reading, or of how its automatic processes are activated by our mind, even in cases when they conflict with our desired intentional behavior.

  • If a function can occur unconsciously, that does not mean that it is always or frequently unconscious: Our unconscious processing abilities may allow numerous behavioral and neurobiological/architectural features associated to consciousness to occur unconsciously, but that is compatible with said features actually operating in a conscious way some or most of the time.

The above may become more evident when studying particular functions in certain individuals. For example, consider navigation and tool use in somnambulists. During somnambulism episodes, individuals can navigate and display complex motor behaviors, such as getting and eating food. Popat & Winslade (2015) report a more unusual but well-documented case, in which a man drove his car, took a tire iron from it and beat his mother-in-law[10].

During episodes of parasomnia[11], physiologically, parts of the brain are in a sleeping state while others are in a waking state. However, subjects "have absolutely no mental interaction with the outside world", and their "sensory perceptions are virtually switched off" (Popat & Winslade, 2015), which would prove that these individuals are not aware. Sleepwalkers cannot intentionally initiate their actions and would not have control over them. If somnambulism is considered a non-conscious state, humans can navigate –at least through known areas– and use tools non-consciously. Nevertheless, it must be stressed that somnambulism is not a usual condition[12] and the cases here presented are rare[13].

Some scientists also caution against overstating the implications of certain functions occurring unconsciously. Roy Baumeister (2007 in Carey, 2007), a professor of psychology at Florida State University, explains that these findings about unconscious processes are “like showing you can hot-wire a car to start the ignition without keys. That’s important and potentially useful information, but it doesn’t prove that keys don’t exist or that keys are useless.”

  • If a specific behavior can occur unconsciously, that is not enough for making generalizations: If a certain function can be performed unconsciously that does not imply that other expressions of that function can operate unconsciously. Or if a function has unconscious components that does not entail that that same function is not indicative of consciousness.

How can that be? Our findings suggest that even when a person is fully awake and alert, unconscious processing occurs in many situations. In fact, unconscious processes seem to occur as a side effect of conscious inputs. Baars and Gage (2010) state that “there is now a large body of evidence that our perception of the world around us is partly unconscious, although the result of the perceptual process is conscious” (26). The authors add that almost all cognitive tasks (e.g. vision, audition, memory, emotions, executive control, social cognition) that take place during the waking state have both conscious (reportable) and unconscious (nonreportable) components. “There are no completely conscious cognitive tasks, as far as we know, and there may be no completely unconscious ones” (242-243).

Consider the ability of language processing. As mentioned, grammar-processing skills are usually unconsciously exercised and perhaps meanings of words are often inferred unconsciously, i.e. using implicit learning (Baars & Gage, 2010). However, that does not entail that we can elaborate a sophisticated discourse unconsciously, nor that all forms of language processing can be performed unconsciously. In this case, it would be important to acquire evidence that unconscious language processing can also perform complex functions, such as reading a long philosophy text.

Similarly, taste avoidance behavior due to taste aversion provides another fruitful example. In this case, as a consequence of conditioned taste aversion (CTA), the taste becomes disliked (noxious stimulus) and it will be avoided in later presentations. CTA can operate unconsciously (see Gallo, 2008) and it seems to be resistant to aging. In humans, CTA is "typically robust and long lasting" (Clasen et al., 2017). In spite of the above, that does not imply that CTA is always or frequently unconscious, not even that other forms of long-term behavior alteration to avoid a noxious stimulus are unconscious.

Thus, to learn more about this issue and about other nonconscious processes in general, we need to figure out possible regularities of unconscious processes –that is to say, determine when we can expect certain functions to take place nonconsciously rather than consciously.

  • If a function can occur ‘unconsciously’ (‘automatically elicited’), that does not necessarily mean that the individual cannot report it: The nociceptive flexion reflex (or flexor withdrawal reflex or withdrawal reflex) is commonly presented as an unconscious reaction. As previously described, this reflex is activated automatically and does not require a conscious decision.

Although spinal centers may carry out commands from higher centers using sensorimotor reflexes, they also return feedback signals to the brain (Baars & Gage, 2010). Hence, the subject can be conscious of her reflex reaction –although she cannot control it– and can report it (see Feinberg & Mallatt, 2016).

Which features operate consciously in humans?

As previously stated, we understand conscious processes as phenomenological subjective experiences in which the individual is aware of an external object or an inner state. Following the literature reviewed (see McGovern & Baars, 2007), we can operationally concretize ‘conscious processes’ as those that meet the following conditions[14]:

(i) They can be claimed by the individual to be intentional,

(ii) They can be reported and acted upon…

(iii) …with verifiable accuracy.

In order to address our main question (**‘**which features operate consciously in humans?’), each indicator is rated in each of these three conditions of conscious processes. This categorization is not intended to be taxative, nor exclude unconscious aspects of conscious tasks. In fact, various types of well-known consciously mediated phenomena (such as sensorimotor skills, problem-solving, sensory-spatial integration) involve unconscious processes, and both conscious and unconscious components are likely to be crucial for normal, flexible and adaptable cognition (Baars & Gage, 2010).

Some features seem to be necessarily conscious:

For the following features, we found no evidence that they may operate unconsciously. Nevertheless, that does not mean that these functions are empirically proven to always occur consciously. We are only stating that, given the current state of the art:

  • The literature assumes that these functions take place consciously, and
  • No empirical evidence was found that these functions would occur unconsciously, although that doesn't discard unconscious components.

For the following cases, in general, less explicit references to the nature of consciousness were found. Therefore, for these features we have slightly less certainty. In some other cases, there is inconclusive evidence that the same function could take place unconsciously, or there is partial evidence that the feature may operate unconsciously under very specific circumstances (see ‘Observations’). Still, these features are often treated as intentional behaviors that require a certain flexibility, which leads us to assume that they are mostly conscious processes in human beings:

Note that for “mood state behaviors”, it is indicated that they can be described in detail by the individual ("Yes (reported)"), although the individual might not have control over these inner states.

With less certainty, ‘Defensive behavior/fighting back’ could be inferred as a conscious reaction and hence, be included in the previous table. Commonly, defensive behavior in humans is described as acting in self-defense, an action that can entail physical efforts and other non-physical mechanisms (see Downs et al., 2007). However, it can be claimed that under certain emotional states, defensive reactions could be elicited involuntarily. Regarding ‘Spatial memory’, there is conflicting evidence for this indicator: it is usually described as a conscious process, but other authors also propose an implicit (non-conscious) form of spatial memory. In general, many aspects of memory are unconscious, while episodic memory, for instance, implies the record of conscious events in the past (Baars & Gage, 2010). However, if spatial memory refers to developing cognitive maps (a form of declarative consciousness), it may not occur non-consciously, although it may require to some degree unconscious processes that are often implicit and hard to articulate (e.g. contextual phenomena, such as the assumptions we make about visual space, the direction of the incoming light in a visual scene. See Baars, 1988).

In general, for various inferred cases additional research is needed. Arguments may render a feature as necessarily (or plausibly) conscious, but these hypotheses should be supported by empirical data.

Some features are frequently described as conscious processes:

For this category, we consider some functions that are usually described as conscious processes in the literature. However, we also found consistent evidence that the same functions may occasionally be displayed non-consciously. These exceptions are pointed out in the ‘Observations’ column:

For its part, ‘Contextual learning’ could also be included in the previous table. Although it seems to be a conscious feature, the contextual cueing effect[15] is commonly cited as an unconscious form of contextual learning in humans and it may require to some degree unconscious processes (e.g. contextual phenomena, see Baars, 1988). However, recent research discusses whether this type of learning would actually occur nonconsciously.

Some functions are more likely to occur unconsciously

In this section, we consider features whose presence can be verified, but (i) the individual may not claim them to be intentional and (ii) the individual may not be aware of them (does not report them). In operational terms and for the purposes of this work, functions that meet criteria (i) and (ii) have been qualified as "unconscious". Additionally, although a function can be reported, it may not be operated upon or avoided.

As shown in the previous table, those functions may occur unconsciously, but that does not mean that they always or most of the time do. Additionally, subliminal perception and learning that might be involved in some of these unconscious functions may take place for highly significant and stereotyped stimuli (see Baars & Gage, 2010).

On the other hand, we found other functions that are more frequently described in the literature as (i) not being initiated intentionally and that (ii) the individual may not be aware of them (does not report them). This is elaborated in the next section.

Are there functions that are especially likely to operate unconsciously in humans?

In the following table, we consider features whose presence can be verified, but (i) the individual may not claim them to be intentional and (ii) the individual may not be aware of them (does not report them). All the cases here included meet criterion (i). Criterion (ii) may or may not be met, with indeterminate precision. Additionally, although a function can be reported, it may not be operated upon or avoided.

We hypothesize that under optimal reporting conditions, the previous functions may be reported by the individual –see "lean yes" in the fifth column of the table.


  • In humans, consciousness has a very limited processing capacity. Additionally, conscious and unconscious information-processing are different strategies which may be more or less efficient for different functions.
  • Some functions related to consciousness can operate unconsciously in humans. Indeed, some sophisticated information-processing functions can be performed unconsciously and conscious and unconscious processes usually go together. Nevertheless, if a function can occur unconsciously, that does not mean that it is always or frequently unconscious. Furthermore, if a specific behavior can occur unconsciously, that is not enough for making generalizations about a given feature. Finally, if a response can occur ‘unconsciously’ (automatically elicited), that does not necessarily mean that the individual cannot report it.
  • According to our findings, the following features are the most likely to occur unconsciously in humans: almost all the noxious stimuli reactions (except for ‘defensive behavior/fighting back’) and some forms of learning (ie. classical conditioning, sensitization, habituation and taste aversion behavior). As it is observed, in general, all these features refer to basic kinds of learning or to automatized rigid reactions that enable an extremely rapid response to potentially dangerous stimuli. As emergency strategies, these reactions have probably been selected for extreme speed and, therefore, bypass or supersede the agency system. Thus, it could be argued that these are the least solid functions for inferring conscious states in other non-human individuals.
  • Notwithstanding the above, it should be noted that if a function may be executed unconsciously in humans, it does not mean it is executed unconsciously in other non-human individuals. What is functional and adaptive to humans may not be functional and adaptive for other non-human individuals; and in turn, what is crucial for some non-humans’ survival may not be comparably relevant to human beings (e.g. anti-predatory responses, see Barrett, 2005).
  • Our findings are provisional and should be considered with caution, given the current state of the art, the lack of consensus on foundational definitions and the ongoing and lively methodological discussions on how to appropriately study conscious and unconscious processing in human beings.
  • Hence, there are many unresolved questions. Baars and Gage (2010) state that we still do not know what humans can do completely unconsciously. Further, as ‘consciousness’ is a construct inferred from behavioral observations, its study should not only rely on subjective reports but also be based on empirical observation methods –brain recording techniques. In this sense, a number of reliable methods for studying both conscious and unconscious brain events are being developed (Baars & Gage, 2010), and a growing integration of behavioral and neural brain evidence might contribute to improving the reliability of research designs and encouraging replication studies. In this regard, some key challenges for future research are (i) operationalising shared foundational definitions, (ii) understanding empirical differences between conscious and unconscious processes, (iii) identifying patterns –if there are any– for determining when we can expect certain functions to take place nonconsciously rather than consciously, and (iv) identifying the specific adaptive purposes of these phenomena, in an effort to understand to what extent our consciousness-indicating features are relevant for assessing consciousness in non-human individuals.


This essay is a project of Rethink Priorities.

It was written by Daniela R. Waldhorn with contributions from Jason Schukraft, Peter Hurford, and Marcus A. Davis. Jason Schukraft and David Moss provided helpful comments on this essay.

If you like our work, please consider subscribing to our newsletter. You can see all our work to date here.

  1. Muehlhauser (2018) lists several physicalist functionalist theories of consciousness. See Appendix Z.1. ↩︎

  2. Muehlhauser (2018) provides a set of examples in order to illustrate the lack of consensus about several aspects of consciousness. See Appendix Z.5. ↩︎

  3. If a human individual becomes conscious of a particular stimulus, then specific neurons in her brain will become highly active. However, the opposite is not necessarily true: “if you have some activity in a given brain area, this does not mean you will necessarily perceive the stimulus. Without enough activity in the right brain areas, awareness may simply fail: the result is unconscious perception” (Baars & Gage, 2010: 190). ↩︎

  4. Blindsight patients with lesions to the occipital lobe of the brain are phenomenologically blind, but can nonetheless perform normally on a number of visual tasks (Zelazo et al., 2007: 2). ↩︎

  5. Priming refers to the incidental influence of environmental context on cognition and behavior. According to Molden (2014), priming effects are described as resulting from the activation of mental representations that facilitate associated subsequent evaluations, judgments, or actions. This effect occurs either without awareness of the prime’s influence or without intention to utilize the prime stimulus (although not necessarily without awareness of the stimulus presentation). ↩︎

  6. In an open letter to priming researchers, psychologist D. Kahneman (2012) called to check the robustness of the findings of priming experiments. He argued that “to deal effectively with the doubts you should acknowledge their existence and confront them straight on”, claiming that the researchers in this field should “examine the replicability of priming results”. ↩︎

  7. In contrast, an unconscious equivalent behavior would be the case of a somnambulist person driving her car, who could not intentionally initiate her actions, would not have control over them and would not respond in a flexible manner to the external world (see Popat & Winslade, 2015). This example is discussed below. ↩︎

  8. In other words, limited functions are closely associated with conscious experience, while very large capacity functions are generally unconscious. For more details, Baars and Gage (2010) provide two tables (2.1 and 2.2) on ‘limited capacity tasks’ and ‘very large-capacity functions’ (43). ↩︎

  9. It must be noted, however, that other authors argue that consciousness dates back hundreds of millions of years. See for example M. Graziano and its Attention Schema Theory (Graziano & Webb, 2017) or Gopnik (2019). ↩︎

  10. Popat & Winslade (2015) refer to Kenneth Parks's case that took place in 1987 in Ontario, Canada. The same night, right after the crime, Parks turned himself in to police, confessing "I think I have just killed two people" (CityNews, 2006). His defense argued that he was asleep during the entire incident and that he was not aware of what he was doing. After careful investigation, the specialists could find no other explanation. The Supreme Court of Canada upheld the acquittal in the 1992 decision ‘R v Parks’ (Broughton et al., 1994). ↩︎

  11. Parasomnias are sleep disorders that result in abnormal physical actions while sleeping (Wilson & Fundukian, 2008). ↩︎

  12. A meta-analysis provided by Stallman & Kohler (2016) showed the estimated lifetime prevalence of sleepwalking was 6.9%. The current prevalence rate of sleepwalking –within the last 12 months– was significantly higher in children (5.0%) than adults (1.5%). ↩︎

  13. More exceptionally, under special circumstances, individuals suffering from somniloquy can describe vividly their dreams while sleeping, after being subjected to hypnosis and ordered to do so (Arkin, 1981). However, no further evidence about this was found and the results of this experiment do not seem to be conclusive. ↩︎

  14. This articulation of three criteria should be understand as operational indicators in an attempt to clarify exactly what we are trying to measure. Otherwise, we may end up with evidence that measures something other than what we intended, especially if the literature employs ambiguous or different definitions of unconscious processes. ↩︎

  15. A learning phenomenon where repeated exposure to a specific arrangement of target and distractor items leads to progressively more efficient search. ↩︎

Sorted by Click to highlight new comments since:

Great article! I like the conceptual clarification that you do about what it means to say that a process is unconscious and how people use this term inconsistently in the literature. I've never seen that put so well and it's important.

I was wondering what you think of cases where a good idea 'spontaneously' occurs to someone while there thinking about something unrelated or while their mind is wandering. I only know anecdotes about this phenomenon, but I think it's a widespread phenomenon that most people would experience something like this themselves.

Some people have some of their best ideas in this way and it seems to satisfy both criteria for being an unconscious process. I am not sure if it's directly related to any of the potential consciousness indicating features, but it seems like an example of very complex cognition being unconscious. Albeit it's a bit murky how it occurs.

Hey Max! Thanks for your feedback and for your vital contribution to this project. Sorry I couldn't get back to you before –I had taken a few days off.

The example you provide fits well in what I classified as "sophisticated information processing functions that can be performed unconsciously". Of course we can come up with creative ideas after a period of conscious thought, but it doesn't necessarily happen that way. As you describe, unconscious processes play an important role in achieving creative insights, during what is called "the incubation period". Neuroimaging studies suggest that the association cortices are the primary areas that are active during this state and that the brain is spontaneously reorganizing itself. Recent research also supports the idea that it is not merely the absence of conscious thought that drives creativity incubation effects, but that during an incubation period unconscious processes contribute to creative thinking.

It's still not clear which are the functional advantages of conscious over non-conscious thinking. In general, which kind of stimuli or tasks are more efficiently processed using unconscious mechanisms is an issue that remains to be elucidated. We also need a more refined distinction between neural correlates of unconsciousness (the absence of any conscious contents) vs. neural correlates of disconnectedness (the absence of perception of the environment) in different altered levels of consciousness.

How these findings can be applied to research on consciousness in invertebrates? I'm unsure. Perhaps we can assess if equivalent structures of their CNS are activated when performing tasks that challenge them to "make associations" –and this may shed some light on how likely they are to show flexible ("creative") responses. Currently, in my opinion, it is clearer how these findings can contribute to improving our thinking: for instance, we are likely to benefit more from an incubation period when we get stuck, or when we are dealing with a problem where the conventional approach is wrong. For specific tasks, a break of 3 min can be enough to promote unconscious thought. Understanding and facilitating creativity can have a direct application in the EA community since creativity plays a vital role in research and designing innovative solutions.

This all seems to assume that there is only one "observer" in the human mind, so that if you don't feel or perceive a process, then that process is not felt or perceived by anyone. Have you ruled out the possibility of sentient subroutines within human minds?

Hi JoshYou. Thanks for your very pertinent comment.

We are aware of the possibility of hidden qualia. It is a valuable hypothesis. Nevertheless, we found no empirical evidence to support it, at least in the literature on invertebrate sentience. If you will, you can view our project as a compilation and analysis of the existing evidence about the sentience of individual invertebrate organisms, as opposed to subroutines within those systems. Under this reading, what we call ‘unconscious processes’ would be understood as processes which are inaccessible to the organisms’s first-person perspective.

We are also aware that, on some accounts, one really does not need empirical evidence to determine whether a process (or subroutine or algorithm) is conscious. All of them are. On such an account, the relevant distinction is between processes that matter morally and those who don’t. Someone who endorsed this view should interpret our position as agnostic about (but compatible with) the thesis that there are hidden qualia.

Are studies on the capabilities of people with impaired conciousness (vegitative or minimally conciousness states, maybe dementia or delerium) considered by studies looking at the limits of human conciousness? I assume doing something like learning and memory research with such patients isn't high priority for their carers, but I assume that, for instance, tasks a person in a vegitative state can do are unlikely to require conciousness.

Hi, gavintaylor. Thanks for your comment. Research in patients with vegetative or minimally consciousness states –different from dementia or delirium, which should be better described as acute disturbances in consciousness– would probably shed some light on this matter. However, this area of research might be challenging by itself.

Disorders of consciousness are heterogeneous, and judging the level of actual awareness has proved a complicated process. Traditional tests and observations have been criticized since they require some level of subjective interpretation –such as deciding whether a patient's movements are purposeful or not. In fact, recent research has revealed that about 40% of vegetative state diagnoses is incorrect.

We know, for instance, that some vegetative patients and other individuals in a minimally conscious state are capable of simple learning (i.e., classical conditioning). In a study, it was observed that the amount of learning correlated with the degree of cortical damage and was a good indicator of future recovery. But none of these effects were found in control subjects under the effect of anesthesia.

Furthermore, integrative brain processing, a proposed prerequisite of awareness, has been observed in minimally conscious state patients as well. Previous neuroimaging work has shown that some vegetative patients, when asked to imagine performing physical tasks such as playing tennis, still had activity in premotor areas. In other patients, verbal cues sparked language sectors.

Hence, these results have two interpretations. First, individuals with disorders of consciousness may have partially preserved conscious processing, which cannot be exhibited clearly via voluntary movement or verbal responses. Or, a second interpretation is that conditioning, for example, can indeed be acquired in the absence of consciousness.

Given that (i) individuals in a pharmacologically controlled unconscious state were incapable of displaying signs of learning, and (ii) learning was a good predictor of recovery, researchers consider that the first interpretation is more likely. However, this comparison must be made cautiously and complementary evidence about these processes–including neuroimaging studies–should also be taken into account.

As you suggest, more research in this field may pave the way for more definitive and accurate assessments of consciousness in humans, and probably, in non-human individuals as well.

This article might have some more useful results, and there's a nice summary in Table 1.

There's also an ongoing debate about the necessity of attention for consciousness.

At the level of theories, Global Workspace Theory, or more narrowly the Global Neuronal Workspace Model, one of the leading theories of consciousness, requires it for consciousness, and I think Dennett's views are similar (see his discussions here and more recently here). Basically, it requires attention and for the attended states to be accessible in some "global workspace", which might include things like working memory and executive functions. Attended Intermediate-level Representation Theory requires attention but not for the state to be passed into any global workspace. Recurrent Processing Theory only requires recurrent connections, i.e. directed loops. It also seems like if you try to abstract away the details of GWT (global to what?), the necessary conditions for consciousness might reduce to Recurrent Processing Theory. See the discussions here and here.

As for evidence, attention (or specific kinds of attention, e.g. "top-down") may not be necessary for consciousness, based on iconic memory, gist, animal and gender detection in dual tasks, and partial reportability, according to this article (see a summary in Table 5.1). It's also argued unnecessary in this (and insufficient for consciousness, too), this and this, and some recent experiments here and here. EDIT: I should add that I think recurrent processes reduce to feedforward ones when you draw them out as causal diagrams over time.

Attention is argued necessary here. Some experimental results here (consistent with being necessary, since we still "attend" to things we don't focus on), on iconic memory here and here, ensemble perception (gist?) here, and more discussion here (leans towards attention being necessary).

Finally, "no-report paradigms" could be useful. They can be used to address the problem that subjective reports may involve capacities and structures that are only necessary for reporting, not the conscious experiences themselves, and so cause us to falsely label these capacities and structures as necessary for consciousness. No-report paradigms do this by measuring "reflexive behaviors correlated with conscious states to provide a window on the phenomenal that is independent of access" (SEP article section, which also discusses some limitations: you can't know for sure the reflexive behaviours still indicate consciousness in the absence of report). There's some recent discussion here, and here's the abstract, which I hope further illustrates its usefulness and subtleties:

Cognitive approaches to consciousness dictate that consciousness involves frontal and parietal circuits that are devoted to thinking, reasoning, evaluating, reporting, deciding, and memory. By contrast, sensory approaches allow for consciousness in creatures that have little or no ability to think and reason.
To decide between cognitive and sensory accounts, we must distinguish between the neural basis of consciousness and the neural basis of reports.
The no-report paradigm purports to solve this problem by using the reports of some subjects to calibrate indicators of consciousness, allowing experimental subjects to make no report.
The problem with the no-report paradigm is that you cannot keep subjects from thinking, and their thought processes may be stimulated by and reflect the contents of perception.
The solution is a ‘no-post-perceptual cognition’ paradigm, one version of which is illustrated here.

Second-order conditioning, with which a "conditioned stimulus can be associated with some other conditioned stimulus or action, and so on, building up long chains of associative links between stimuli and actions", and unlimited associative learning, which requires second-order conditioning as well as conditioning on compound and novel stimuli, have been proposed as markers of consciousness. Apparently there haven't been any studies on whether second-order conditioning is possible on subliminal stimuli, though. Some discussion in this review by Jonathan Birch of the book The Evolution of the Sensitive Soul: Learning and the Origins of Consciousness by Simona Ginsburg and Eva Jablonka:

They use this marker (unlimited associative learning) to defend a generous view about the distribution of consciousness in the natural world, on which a capacity for conscious experience is common to all vertebrates, many arthropods and some cephalopod molluscs.


For example, I don’t know of any demonstration of the possibility of second-order conditioning of any kind on subliminal stimuli. This raises the intriguing possibility that second-order conditioning alone might already be a positive marker of consciousness, whether or not the stimuli are compound or novel. Second-order conditioning (but not on novel, compound stimuli) has been found in honey bees (Hussaini et al. 2007) and even snails (Papini 2010, p. 366). Yet I also don’t know of any experiments that have actively looked for second-order conditioning on subliminal stimuli in humans and failed to find it. 

In short, the link between UAL and consciousness is promising yet largely uncharted territory. The hypothesis that UAL requires consciousness must be considered very tentative, since it has not been directly tested.

New paper from the Foundations of Animal Sentience project just published that seems relevant:

The search for invertebrate consciousness by Jonathan Birch.

The author argues for a middle ground between theory-heavy and theory-neutral approaches after criticizing each. The theory-light approach's only theoretical commitment is the facilitation hypothesis:

Phenomenally conscious perception of a stimulus facilitates, relative to unconscious perception, a cluster of cognitive abilities in relation to that stimulus.

The author goes on:

We already have some good, scientifically plausible candidates for abilities other than verbal report that are facilitated by consciousness. Here I will briefly review three such candidates.

The candidates given are trace conditioning, rapid reversal learning, and cross-modal learning, described further in the article, and in the last section of the paper, the author looks at studies on these abilities in bees.

These three examples are enough to illustrate the “theory‐light” strategy. Without committing to any particular theory of consciousness, we can investigate, in humans, the question of which cognitive abilities are facilitated by conscious perception. If we have only one such ability, and find that one ability in the target non‐human species, a critic will say: that could be done without consciousness, even if it happens to involve consciousness in humans. So we need a cluster of correlated abilities, not just one, in order to build up a case that is harder for the critic to resist. The larger and more diverse the cluster, the stronger the case will be.

Once we have constructed, on the basis of evidence from humans, a tentative, defeasible hypothesis about the cluster of consciousness‐linked abilities, the next step in the theory‐light approach is to look for the cluster in the target species of nonhuman animal. Some elements of the cluster will, inevitably, be absent. We will not find verbal report. What we might find is a substantial fraction of the cluster. For example, we might find that bees can do trace conditioning of the right kind, reversal learning of the right kind, and cross‐modal learning of the right kind. I say “of the right kind” in each case as an acknowledgement that more work is still needed here to pin down the precise type of each ability that is linked to consciousness in humans.

This will still not be enough to convince a reasonable critic, who will say: I'm afraid I can seriously envisage all of those abilities occurring without conscious experience, even though they are all facilitated by conscious experience in humans. You've shown the abilities are present, but you haven't shown their facilitation by consciousness. This is a fair criticism, but we can overcome it. The next step in the theory‐light approach should be to investigate protocols with the potential to cause unconscious perception in the animal: backward masking, the attentional blink, flash‐suppression, distracting tasks, and so on. For brevity, I will refer to this whole family of procedures as “masking”. We need to find out whether the identified cluster of putatively consciousness‐linked abilities is selectively switched on and off under masking in the same way it is in humans.

For example, in humans, presenting a tone subliminally appears to switch off trace conditioning while leaving delay conditioning in place. We can ask: is the same true of our target species of animal? Do we see a similar pattern of sensitivity to masking? When the stimulus is masked, does this selectively switch on and off the entire cluster of consciousness‐linked abilities?

This seems similar to no-report paradigms. The indicator of consciousness seems to be improved performance  on a task under conditions which would involve conscious processing in humans over conditions which would involve only unconscious processing in humans. I think no-report paradigms instead use reflexive behaviours correlated with conscious perception, but this correlation could just be due to cause common to both the reflex and conscious perception, not because one causes the other. On the other hand, if conscious perception facilitates some ability, it causes the expression of that ability.

Super interesting, I really like seeing this work being done.

I wonder if there is a meaningful difference between how you define consciousness:

‘conscious processes’ as those that meet the following conditions:
(i) They can be claimed by the individual to be intentional,
(ii) They can be reported and acted upon…
(iii) …with verifiable accuracy.

and conscious states that are associated with positive or negative experienced value. One example that came to my mind are dreams: Sometimes I remember having had very negative or positive experiences, but mostly I don‘t remember anything. I strongly suspect I still have those dreams (right?), but those states seem to involve no intentionality, they cannot be acted upon and have no connection to verifiability.

Another candidate process that just came to mind (very uncertain) that might be indicative for experiencing evaluative states is planning. You are mentally laying out paths into the future and need a flexible evaluation function that gives you feedback that guide your planning.

P.S.: Have you thought about posting it on the LessWrong forum? I think they are also a very informed crowd with respect to the topic of consciousness and might give you valuable feedback.

Curated and popular this week
Relevant opportunities