Ethics statement

The study was conducted in accordance with the recommendations of the Declaration of Helsinki and the local Ethical Committee (“Comitato etico”, San Raffaele). Before the experiments, all participants signed the informed consent.

Participants

Eight healthy observers volunteered in the main experiment (6 females and 2 males, with normal or corrected-to-normal vision, aged between 22 and 34, naive as to the purpose of the experiment except two authors). They received a book in return for the collaboration.

Stimuli and task

Observers were seated in a darkened room about 60 cm in front of a computer screen (Sony Trinitron, 21 inches; frame-rate: 85 Hz; resolution: 800×600 pixels; background luminance: 19 cd·m⁻²), with the head resting on a forehead support. The stimuli (random dot kinematograms, Figure 1) were generated by a custom program written in MATLAB (6.5.0 R13, The MathWorks Inc., 2002, equipped with PsychToolbox 3), and consisted of 100 dots (diameter  = 0.2 deg), half black (luminance  = 0.2 cd·m⁻²) and half white (luminance  = 118 cd·m⁻²), moving linearly at a velocity of 5 deg·s⁻¹, with a lifetime of ∼82 ms (7 frames), presented within a central square aperture (side  = 11 deg). At each lifetime cycle, the dots were regenerated in a different position within the screen aperture; part of them was assigned a randomly-chosen motion direction (random motion), while the remaining part kept moving in the same direction (coherent motion, either rightward or leftward). The proportion of the latter dots determined the coherence of the stimulus, which could be 0%, 15%, 30%, or 60%.

Each trial started with the presentation of a central fixation dot (diameter: 0.3 deg; luminance: 118 cd·m⁻²), followed by the motion stimulus after a random interval comprised between 1 and 2 s (Figure 1). The offset of the fixation dot was the imperative cue to emit immediately the response (direction discrimination task), either verbally (“right”/”left”) or with the eyes (in this case the instruction was to make a saccade in the direction of motion, to the right or left border of the stimulus pattern). The response terminated the stimulus within 4 video-frames. We manipulated the duration of the fixation dot, which was switched off 0.2, 0.5, 1 or 4 s after the onset of the motion stimulus. Throughout the text, this manipulation – an experimental factor – will be called “urgency”, to denote the fact that, especially at 0.2 and 0.5 s, the imperative cue forced observers to emit a sudden, early response. Higher urgencies corresponded to shorter durations of the fixation dot. Observers could respond before the offset of the fixation dot if they had perceived clearly the global motion direction, which could happen especially at the lowest urgency (4 s).

The imperative cue was quite effective in determining a prompt response with a remarkably small variability of response times, except at the lowest urgency (Table 1). Observes were informed that in many cases the global motion direction might not be seen at all, in which case the discrimination response must be a blind guess. Before the beginning of the experiments, participants got acquainted with the task, which required about 100–200 trials.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Response times of the discrimination responses.

https://doi.org/10.1371/journal.pone.0017079.t001

In the control experiment (three additional participants, 2 females and 1 male, with normal or corrected-to-normal vision, aged between 22 and 27, naive as to the purpose of the experiment) the procedure was substantially the same as in the main experiment, except for the presence of a condition in which the perceptual judgment (actually a simpler yes/no task, see the Results section) was given upon the imperative cue. In another condition it was given after the discrimination response, i.e., at the end of the trial as in the main experiment. The two conditions were administered in different sessions (3 for each condition, 256 trials per session) in a counterbalanced order. Motion discrimination was performed only with verbal responses, and ocular fixation was monitored throughout the trial.

Recording procedures

The real-time stimulus presentation was monitored by means of an analog photocell system, and the signal recorded together with the eye position traces and the verbal traces. Eye movements (horizontal component, monocular) were recorded through infrared oculometry (Dr. Bouis Oculometer, nominal accuracy <0.3 deg), while the verbal discrimination responses were recorded with a directional microphone and an amplifier. The analog signals were visualized in real time on an oscilloscope, sampled through and A/D converter (16 bit, 1000 Hz), and stored for subsequent analyses. In the verbal response experiments, trials containing saccades or accidental utterances were discarded, and re-presented at the end of the session. In the saccade experiments, trials containing saccades before the offset of the fixation dot, or small saccades (<1 deg), or ocular artifacts (e.g., eye blinks) were discarded, and re-presented at the end of the session.

Rating task

In order to capture as reliably as possible the first-person, subjective quality of visual experience [18], we used a measure of perceptual awareness structured as a type I task with an absolute 5-points scale. Firstly, we wanted the perceptual judgment to target specifically the visibility of the motion stimulus (type I task: judging an external event [19]) rather than the confidence of the observer's own discrimination response (type II task: judging an internal event): Being confident of the response is not equivalent of being aware of the stimulus [20]. Secondly, we used a multi-point scale instead of a yes/no task to capture the graded nature of perceptual awareness [6], [10], [15], especially because we worked with degraded stimuli. The number of points was chosen to make the scale simple and comfortable for the observers: It has been shown that four categories are enough for the PA scale to be informative, while a high number of categories may be confusing [15]. Thirdly, we took caution to anchor the scale to the minimum and maximum absolute values [21], namely, null visibility and full visibility, and ensured that participants understood that the scale should represent a linear quantity. We insisted that the observers assigned a score of zero only in case of complete invisibility of global motion direction. The instructions to the participants, which were given in both written and colloquial form, were the following: 0 =  You didn't see at all the motion direction; 1 =  Between 0 and 2: you had a raw feeling of the motion direction; 2 =  Half-way point of the scale: you probably saw the motion direction; 3 =  Between 2 and 4: you saw the motion direction, but not too well; 4 =  You saw clearly the motion direction.

Experimental design and data analyses

For each trial, the dependent variables were the actual viewing time (which in turn was determined by the urgency condition and the response time), the response direction, and the PA score. As for the independent variables, we used a randomized mixed design with 5 factors: motion direction (within-subjects factor, 2 levels: right/left), response modality (between-subjects factor, 2 levels: saccade/verbal response), stimulus coherence (within-subjects factor, 4 levels: 0%, 15%, 30%, 60%), urgency (within-subjects factor, 4 levels: 0.2 s, 0.5 s, 1 s, 4 s), repetition (N = 8), and block (N = 5). Each block consisted of a single experimental session with 256 trials (average duration: ∼1 hour, including pauses), for a total of 1280 trials per subject. Participants were invited to take pauses whenever they wished or felt tired. The five blocks were normally administered in different days.

By means of an interactive computer program we determined the onset of saccades (the moment at which the instantaneous horizontal eye velocity exceeded 30 deg·s⁻¹ for more than 15 ms), and verbal responses (the moment at which the envelope of the vocal emission exceeded the noise level). At this stage, 87 trials (∼0.8% of the total) were eliminated due to residual artifacts that went unnoticed during the data acquisition phase.

Within the framework of Signal Detection Theory (SDT, [22]), the observers' sensitivity in the rating task was assessed through both area and distance measures. We used the non-parametric index A', which estimates the area under the ROC (Receiver Operating Characteristics) curve (AUC), as well as the distance index d'. The area under the ROC curve is 0.5 when the performance is at chance, while it is 1.0 when the performance is errorless. The corresponding values for d' are 0.0 and ∞. The response bias was assessed by calculating the location criterion c =  −½[z(H)+z(F)], which is independent of d'. We treated the rating task as a multiple yes/no detection task: For each observer, and for each coherence and urgency condition, we calculated hits (on signal trials: coherent motion) and false alarms (on noise trials: coherence  = 0%) over four response combinations. Given the five PA scores used in our experiment, there were four possible pairs of hits and false alarms rates. That is, PA scores greater than 0 were first considered to be ‘yes’ responses, while the 0 score was considered to be a ‘no’ response; next, PA scores greater than 1 were considered to be ‘yes’ responses, while PA scores less than 2 were considered to be ‘no’ responses, and so on, until encompassing all pairs of hits and false alarms rates. In the control experiment, in which the rating task was replaced by a yes/no task, d' was computed simply as z(H)–z(F). In the motion discrimination task d' was computed also as z(H)–z(F), where hits are the correct responses to an arbitrary stimulus direction (in our case rightward), and false alarms are the correct responses to the other stimulus direction. To adjust for extreme values, the entries in the contingency tables have been increased by 0.25 [23].

The degree of correlation between the performance in the discrimination task and subjective visibility was assessed with the non-parametric Kendall τ, which tested the trial-by-trial relationship between the correctness of the discrimination response (a binary variable) and the PA score [24]. The discrimination rate was tested against chance with the one-tail binomial test (alpha level  = 0.01). We used repeated-measures ANOVAs to test the significance of the discrimination rate, AUC, and PA scores (all values were z-transformed, with the latter previously re-scaled in the 0.5–1 range) over coherence (excluding 0%) and urgency conditions. Separate repeated-measures ANOVAs were used to assess the significance of d'. Pair-wise comparisons at each urgency and coherence conditions were tested with the paired Student t.

Assessing perceptual awareness

This study addressed the issue of conscious perception. Consciousness is in the first place regarded, both in popular and scientific psychology, as a subjective, private phenomenon. As such, it appears to be ultimately inaccessible to a third-person approach. This notion of consciousness corresponds roughly to what the philosopher Ned Block termed phenomenal consciousness [34]. The distinction between phenomenal consciousness and access consciousness (i.e., a conscious state that is not entirely private) justifies the distinction between a purely phenomenological approach to consciousness, and the approach typical of scientific psychology and neurosciences. However, this passage is far from obvious [35], and enduring disputes about the nature of consciousness persist, as also witnessed by the polysemy of the term consciousness, and the variety of related terms and qualifiers. Throughout the paper we use the terms “awareness”, “consciousness”, “phenomenal”, “sensation”, “subjective visibility”, as synonyms, and are intended to denote the first-person visual experience of “seeing”, whereas the term “perception” is used in a rather liberal way. The difficulty to assess conscious perception in controlled experiments, in particular, gave rise to endless methodological discussions as to the best approach to consciousness, and what counts as index of conscious experience – e.g., direct vs. indirect, objective vs. subjective measures [18], [25], [36].

In order to capture as directly as possible the first-person, subjective quality of visual experience, while at the same time keeping an objective stance, we took as a primary index of visual consciousness the explicit visibility judgments given by the observers [18]. We opted for a simple PA scale that accommodates different degrees of subjective visibility without constraining visual perception in the Procrustean bed of the conscious/unconscious dichotomy (see Methods), an issue that is particularly cogent under uncertainty conditions. Although a perceptual judgment may be regarded to involve meta-cognitive processes distinct from the “genuine” phenomenal fact, we preferred here to avoid such a clear-cut separation, and considered simple perceptual judgments as the basic unit of measurable, subjective perceptual experience. That is, we tried as much as possible to go to the heart of conscious visual perception by tagging the natural notion of “seeing” – regardless of its fuzzy or illusory nature – from the point of view of the observer.

A posteriori, the observers' perceptual judgments turned out to be both reliable and veridical. They were reliable, because the pattern of results was very systematic, both within and between observers. They were veridical, because when the stimulus was pure noise (coherence  = 0%), the reported visibility of global motion direction was almost null (mean PA scores ranging from 0.08 to 0.16 over urgency levels, not shown), and because especially at the highest stimulus coherence and long viewing time perfect visibility was easily attained (PA score  = 4). That is, observers used the entire PA scale without difficulty.

Note that despite our observers were carefully instructed to assign a zero score only when they didn't see at all the global stimulus direction, it is possible in principle that a zero visibility score could in fact tag sometimes cases in which visibility was not truly “null”, but simply not worth a score of one. However, a visual experience that for a normal observer is not even worth a minimum visibility judgment should be labeled “blindness”, and therefore we took the trials with PA = 0 as trials in which observers were blind to motion direction in a substantial way (blind trials).

By using SDT we aimed at providing a common currency for quantifying the performance in the two tasks, and at assessing the contribution of top-down and bottom up components in the perceptual judgment (criterion and sensitivity), without however pretending that perceptual awareness is entirely captured by a single objective quantity [36]. Yet, we believe that combining a somewhat qualitative approach (e.g., Figure 4) with a quantitative analysis based on the transformation of the visibility ratings into an equivalent detection performance (e.g., Figure 3) is a simple way to capture satisfactorily the subjective quality of perceptual experience while at the same time remaining within a robust methodological framework (see also [37]).

Perceptual decisions can precede awareness

When considering the temporal dynamics of perception, the superiority of motion discrimination capability over subjective motion perception has an interesting consequence: awareness and discrimination become dissociated in time, i.e., perceptual awareness lags discrimination, in the sense that it takes longer to reach the same performance level and to stabilize. This does not imply strictly serial processes, as the processes underlying discrimination and awareness can coexist in time (race model). Note that we are not suggesting to take the saturation of perceptual awareness as the temporal marker of the perceptual delay; Rather, its gradual buildup suggests that the notion of a precise point in time where conscious perception is realized may be too strict, at least with our degraded motion stimuli.

That perceptual awareness is more sluggish than motion discrimination may appear somewhat unsettling, as we tend often to assume that conscious perception precedes decision. However, phenomena such as blindsight [24], [28] and unconscious perception [18], [25] suggest that automatic decisions are indeed possible under certain conditions. In more ordinary contexts, many sensory-driven actions, as well as the stimuli that originated them, pass mostly unnoticed, as when driving or in sports. Similarly, we incessantly decide where to make the next gaze shift, despite poor or null awareness of peripheral - and sometimes also central - visual information [38], [39]. Awareness may just follow.

In general, a slow dynamics of perceptual awareness may reflect the need to go beyond the contingencies of a continuously changing world. For example, if the temporal dynamics of perceptual awareness were strictly tuned to the timing of exploratory eye movements, we would probably see the world as a rapid sequence of snapshots, one for each fixation period. Indeed, we recently found that a spontaneous increase of the saccadic latency from ∼200 to ∼500 ms resulted in a progressive increase of their susceptibility to an illusory mislocalization effect [8], which suggested that perceptual awareness of the target position lagged short-latency saccades (blind saccades). Also, very fast responses to natural objects (∼100 ms, [40]), and even more so the amazingly fast color discrimination capability recently discovered in the monkey (∼30 ms, [41]), call for an automatic process that may not wait for conscious perception [31]. Thus, transient dissociations, long or short, between decision and perception may be the rule, rather than the exception, during everyday life, in the sense that our sluggish perceptual awareness can lag visually-guided motor responses [8], [9], [30], action selection [42], perceptual decisions [41], and even intuitive choices [43]. Another reason why awareness should be delayed is that, in order to experience the world as unitary despite the various asynchronies deriving from multiple sensory (and also non-sensory) signals, a buffer system is needed that accommodates all possible asynchronies; The readout of a buffered system is by necessity delayed. A general picture thus arises that depicts awareness as a post-hoc construct emerging on the top of a developing or even mature decision, both when the stimulus is internal, as in unconscious initiation of a voluntary motor act [7], or external, as in the present study.

In disentangling decision from conscious perception, our study warns against an indiscriminate use of monkeys' saccadic eye movement as a proxy for conscious visual perception (e.g., for what monkeys “see”), even when accuracy is rewarded [1], [2], [3]. More generally, our findings indicate that, especially when time is an issue, objective forced-choice responses may not provide a full account of visual perception, as the perceptual decision can be taken when the formation of perceptual awareness is still underway. Note that in our data there was a non-negligible proportion of intermediate visibility trials in which observers anticipated the imperative cue, which is a hint that even well-trained observers may tend to respond somewhat automatically before perceptual awareness is stabilized. In the less corticalized monkey, automatic visuo-motor decisions could be even more pronounced. Thus, because selecting correct responses is a too lenient criterion to ensure that perceptual awareness is already stabilized, it would be important to measure directly the degree of monkey's perceptual awareness, an issue that however is theoretically far from obvious [35], [44]. Building upon past work [45], [46], [47], it might perhaps be possible to train monkeys to pair the motion discrimination response with a simplified rating task for perceptual awareness. In principle, comparing the buildup of perceptual awareness and perceptual decisions in monkeys would permit to assess how much their visual awareness is flattened to the swift dynamics of a perceptual decision.

Seeing global motion: neural correlates of conscious vision

How can a perceptual decision be taken when perceptual awareness is still developing? At least for global motion direction, the answer may lay in the particular mechanism that is thought to regulate the formation of the decision signal from the underlying visual signal. Several findings, both in humans and monkeys, indicate area MT as a crucial node for motion processing (see [48] for review). In recent years a growing body of data have disclosed also the role of area LIP (Lateral Intraparietal) in perceptual decisions involving motion stimuli (see [3] for review). When a monkey is instructed to make an eye movement to report the direction of a random-dot kinematogram, neurons in LIP pick up sensory evidence, presumably from MT, and integrate it for some hundreds of ms until a decision bound is reached, and an oculomotor command issued. Importantly, the decision is reached even when the stimulus is still available or the response procrastinated, because LIP neurons exploit only the initial part of the discharge of MT neurons [49], [50]. Thus, monkey LIP seems to work as a device that implements a relatively quick rise-to-threshold mechanism for various types of visuo-motor responses when a perceptual decision is required [51]. In this way, the decision is ready even though MT neurons are still processing the motion input. Note that structures other than LIP could be involved in motion discrimination when the response is verbal, perhaps as part of a circuit for more abstract decision-making [4], although the fact that we found a pattern of results virtually identical for saccades and verbal responses suggests that similar mechanisms may be at play.

Perceptual awareness would instead require a longer processing of visual signals. At a first sight, the long discharge of MT neurons, that outlasts the perceptual decision, could be thought of as the neural basis for the formation of visual awareness for motion. However, in the anaesthetized monkey, MT neurons respond promptly to the onset of random dot motion, and the information rate saturates very quickly (within ∼100 ms, [52]. Obviously, such automatic, almost time-invariant response cannot give rise to perceptual awareness, and is rather involved in smooth pursuit initiation. Moreover, in the awake monkey the sensitivity to global motion of individual MT neurons, while increasing over time (for about 2 seconds), is comparable to the sensitivity of motion discrimination of both monkeys and humans, as assessed through coherence threshold [53]. Yet, we showed that, at least in humans, the sensitivity in the discrimination task is higher than the sensitivity in the rating task. Thus, if the same holds for monkeys, the sensitivity of MT neurons would be too high to be compatible with perceptual awareness. Therefore, a distinct, sub-optimal readout mechanism should be hypothesized, perhaps through the pooling a sub-population of MT neurons containing a large amount of correlated noise that collectively have a lower signal-to-noise ratio than those feeding LIP for a perceptual decision [54]. Thus, in this scenario neither MT nor LIP are sufficient for the formation of perceptual awareness, as at least an additional readout mechanism would be required, exhibiting integration properties compatible with the slow buildup of perceptual awareness. Clearly, an alternative scenario is that in the monkey motion discrimination and perceptual awareness are not dissociated, in which case monkeys may have only an immediate, faint subjective sense of motion, in principle entirely accountable by the activity in MT/LIP.

In humans, the contribution of area MT to motion awareness remains a matter of speculation [55], although findings such as that a bilateral damage to this region causes motion blindness [56], or that implied and imagined motion activate MT [57], [58], or that MT exhibits spatiotopic properties [59], suggest an important role in high-level motion processing. To shed light on the role of human MT in the buildup of awareness of global motion, it would be important to know whether its temporal integration properties are compatible with the long buildup of perceptual awareness or whether they just comply with the relatively quick dynamics of perceptual decisions. Because temporal summation in human MT does not appear to be very long-lasting when tested with stimuli similar to those employed in our study (M.C. Morrone, personal communication), it is doubtful that this cortical structure can fully support the slow buildup of perceptual awareness for motion. Thus, it would seem that MT can be considered a high-level area as far as global motion processing is concerned, but a low-level area as far as the associated subjective quality of motion vision is concerned.

The above considerations do not exclude that neural circuits in MT and/or LIP may take part in an early phase of perceptual elaboration as a precursor of conscious perception. MT could contribute to form a faint feeling of motion, as in motion imagery [57], [58], [60], perhaps reinforced by multiple connections with earlier striate and extrastriate areas (especially human V3A, whose response to motion is remarkable [61]), or by the recruitment of other cortical circuits [13], [30], [62]. Decision-related activity in LIP may contribute to form the subjective confidence in the discrimination performance [46], which depends on an internal evaluation of both one's own decisional capability and the degree of awareness of the stimulus [19]. The involvement of a sensori-motor area such as LIP in visual perception would be generally in line also with current views of embodied cognition and motor theories of perception [63].

In sum, among the mosaic of visual areas activated by global motion in humans [64], only those supporting temporal integration properties compatible with the slow buildup of perceptual awareness would candidate as a specific neural correlate of consciousness, sufficient to confer us the full phenomenic depth of motion perception. Clearly, a less simplistic view is that awareness is a large-scale, distributed property [13], in which case no single cortical structure may exhibit a macroscopic activation that co-varies on its own with the conscious percept.

Conclusions

Our study has documented a remarkable capability of identifying the direction of degraded motion, with a saccade or a verbal response, at a time after stimulus onset when motion is still subjectively invisible or poorly visible. This sort of “transient blindsight” suggests that many ordinary perceptual decisions – not necessarily motor – can be effectively taken when our sluggish conscious representation of the world is still a void or a faint impression. As a consequence, forced perceptual decisions may not tell the entire story about visual perception, especially when time is an issue.