Augmenting aesthetic chills using a wearable prosthesis improves their downstream effects on reward and social cognition

Previous studies on aesthetic chills (i.e., psychogenic shivers) demonstrate their positive effects on stress, pleasure, and social cognition. We tested whether we could artificially enhance this emotion and its downstream effects by intervening on its somatic markers using wearable technology. We built a device generating cold and vibrotactile sensations down the spine of subjects in temporal conjunction with a chill-eliciting audiovisual stimulus, enhancing the somatosensation of cold underlying aesthetic chills. Results suggest that participants wearing the device experienced significantly more chills, and chills of greater intensity. Further, these subjects reported sharing the feelings expressed in the stimulus to a greater degree, and felt more pleasure during the experience. These preliminary results demonstrate that emotion prosthetics and somatosensory interfaces offer new possibilities of modulating human emotions from the bottom-up (body to mind). Future challenges will include testing the device on a larger sample and diversifying the type of stimuli to account for negatively valenced chills and intercultural differences. Interoceptive technologies offer a new paradigm for affective neuroscience, allowing controlled intervention on conscious feelings and their downstream effects on higher-order cognition.


Methods
To test and improve the effects of the preliminary device (Figure 1), we conducted a preliminary study, whereby 52 subjects (International interdisciplinary students (American, European and Asian) from the Harvard Summer School at the Centre for Research and Interdisciplinarity) were distributed into three conditions. 23 subjects were exposed to treatment (chills stimulus alone), 19 to control (film about everyday life of student), and 10 subjects (3 females, M age=40, STD=8) were exposed to chill eliciting stimulus and received an additional chills stimulus from the first version of the actuator placed on their right arm (actuator).

Results
We build a comparison table (Table 1). In terms of pleasure, there is an extremely significant difference for treatment and actuator at p << .01, in comparison to control, the result is significant at p < .05. In terms of pleasure, highly significant difference between treatment and actuator ( p << .01). We test for the difference between treatment and actuator in terms of chills. The Fisher exact test statistic value is 0.3698. The result is not significant at p < .05. Table 1. Subjective pleasure and relaxation across groups.

Discussion
Actuator seems reliable in eliciting chills. However, arms may not be the optimal location. No significant difference in terms of chills (which we attribute to small sample size).
However, these chills are less pleasurable and relaxing than all conditions we tested. We attribute low pleasure to user experience and low relaxation to white coat effect and the preliminary nature of the protocol. All in all these results are encouraging but demand the user experience to be improved. New locations should be tested to build a gradient.

Methods
We used a within-subject design to investigate whether one can artificially modulate interoceptive inferences underlying aesthetic chills, their felt frequency and intensity. The hypothesis was that both psychological and physiological responses would be different between with and without the Frisson stimulation device. Experiments were conducted using self-report, a muscle-bend sensor allowing for silent report of chills experience, image capture for facial expression analysis, physiological sensors (heart rate and skin conductance) for measuring physiological changes concurrent with frisson, and both quantitative and qualitative surveys.

Protocol
The participant entered the laboratory between 12:00-4:00 pm, sat in front of a computer monitor, and was provided with a consent form. Participants were told that the study examined the relationship between temperature and attention, and were specified the definition of aesthetic chills as psychogenic waves of cold as opposed to external cold stimulations from the device, and asked to report only the former. The stimulation device was placed on their back and sensors were wrapped around the middle phalanges of their right hand's index and middle finger. Participants were told to clench their hand if they experienced aesthetic chills at any time during the film, such that the handworn sensor could collect a count of chills. Next, each participant was exposed to a calming film of a cold landscape (ice mountains and cascade) for 90s to control for stress baseline and prime the subject with the concept of cold. Next, the chill-eliciting video stimulus began, entailing a 213s long speech and introductory message about the film content. Artificial chills were delivered at timecodes 2:43, 3:52 and 4:03. The same video viewing procedure was done with or without the Frisson stimulation device, depending on condition, but participants wore hand-worn physiological sensors regardless of condition. We collected subjective data in the form of surveys after each condition which asked questions such as the frequency of chills experienced by the participant, the intensity of the chills experienced, degree to which subjects shared the speaker's viewpoint, the degree to which subjects shared the speaker's feelings, confidence of understanding the video and location of perception of emotion. The subjective questions were asked on a likert scale of 0-10 (see supplementary materials for further details about the procedure, stimulus and questionnaire). Once the experiment was finished, the experimenter disconnected the sensors and provided the subject with a questionnaire. Finally, the subjects were thanked for their participation and fully debriefed.
Each session lasted about 20 min.

Participants
A total of 21 students participated in the experiment (N females= 7, N males =14, M age=27, STD=6.5). Sample size acceptable range was determined based on a compiled exhaustive database of aesthetic chills research (see below). On their arrival at the laboratory waiting room, they were randomly assigned to one of the experimental conditions. Following a within subject design, each participant took part in both experimental conditions: once with and once without the device, each time exposed to the same audiovisual stimulus. The order was counterbalanced to account for learning effects.

Database of chills research
We constructed an exhaustive database of existing chills research. The RDF semantic web structured intelligent DB contains all existing academic literature on chills (N~40). All papers are listed in the Index tab. A user interface can be accessed at http://51.68.79.244:4001 . There one can select various properties and construct queries for the DB (e.g., listing all studies "on musc" conducted with "ECG"). If needed login/password: "guest". The compiled list of all aesthetic chills experiments are also available at the following url: https://bit.ly/2I8oFlV Table 2 : Articles selected from chills database to determine the sample size for this study.

Materials
The materials used in the experiment are in the following section.

Stimulus
The audiovisual stimulus was presented using a standard computer screen and headphones with a fixed volume. Following on a preliminary study using responses to a survey inquiring into the properties of chill-eliciting situations [32] and a software for searching YouTube videos in terms of their density of chills-related comments, we designed the stimulus combining two modalities (audio and visual) likely to trigger chills in the studied population.
Film audio tracks are more powerful than music in eliciting piloerection, a common marker

Sensors and software
Physiological data was collected using hand worn sensors from the Dormio device [14]. We collected heart rate using a Sparkfun Pulse Sensor Amped PPG sensor from the middle finger and electrodermal activity using dry electrodes from the wrist. We also attached a flexion sensor on the index finger and asked participants to clench their index finger when experiencing a chill sensation. The data was sampled at 100Hz. All the data was collected using Bluetooth to a PC. Facial expressions data was recorded using a camera mounted over the screen and stored in a SD card.We used Affdex SDK for detecting facial expressional from recorded video at 30fps. Affex SDK detected 12 relevant metrics from the video. The metrics were 'smile', 'anger', 'valence', 'browFurrow', 'noseWrinkle', 'joy', 'surprise', 'browRaise', 'upperLipRaise', 'mouthOpen', 'eyeClosure', 'cheekRaise'