scispace - formally typeset
Search or ask a question
Posted ContentDOI

A common neural currency account for social and non-social decisions

TL;DR: In this article, the authors used simultaneous EEG-fMRI and a task in which decision evidence in social and non-social contexts varies along comparable scales to identify comparable time-resolved build-up of activity in the EEG, and then used the endogenous trial-by-trial variability in the slopes of these accumulating signals to construct parametric fMRI predictors.
Abstract: To date, social and non-social decisions have been studied in isolation. Consequently, the extent to which social and non-social forms of decision uncertainty are integrated using shared neurocomputational resources remains elusive. Here, we address this question using simultaneous EEG-fMRI and a task in which decision evidence in social and non-social contexts varies along comparable scales. First, we identify comparable time-resolved build-up of activity in the EEG, akin to a process of evidence accumulation. We then use the endogenous trial-by-trial variability in the slopes of these accumulating signals to construct parametric fMRI predictors. We show that a region of the posterior-medial frontal cortex (pMFC) uniquely explains trial-wise variability in the process of evidence accumulation in both the social and non-social contexts. We further demonstrate a task-dependent coupling between the pMFC and regions of the human valuation system in dorso- and ventro-medial prefrontal cortex (dmPFC/vmPFC) across both contexts. Finally, we report domain-specific representations in regions known to encode the early decision evidence for each context. These results are suggestive of a domain-general decision-making architecture, whereupon domain-specific information is likely converted into a "common currency" in the dmPFC/vmPFC and accumulated for the decision in the pMFC.

Summary (4 min read)

Introduction

  • Most strategic decisions occur under considerable uncertainty.
  • In contrast, when negotiating a deal in person, the trader’s risk assessment may rely instead on how trustworthy the other party appears (Fouragnan et al., 2013; Griessinger and Coricelli, 2015).
  • The first account assumes that the brain dedicates largely separate networks for encoding social and non-social forms of decision uncertainty and for assigning value to different choice alternatives.

Results

  • The authors investigated economic decisions within a social context by exploiting trustworthiness in a partner’s face during a strategic economic game to generate predictions about possible outcomes (see below) and within a non-social (purely probabilistic) context by manipulating outcome probabilities in individual gambles.
  • The authors instructed participants that the probability of receiving the higher payoff for ’Play’ choices would be based on the overall likelihood with which each face identify split the augmented endowment (here 4 pts) in a previous study (social trials) or the pure reward probabilities depicted on the face stimuli (non-social trials).
  • Persistent accumulating activity with a build-up rate proportional to the amount of decision difficulty should result in a gradual increase in the classifier’s performance while the traces for the easy and difficult trials diverge as a function of elapsed time in stimulus-locked data (Fig. 4a).
  • Here, instead, the authors estimate the rate of EA on individual trials purely from the slope of the accumulating activity they identified in the EEG data.
  • As with the EA clusters, the authors found that in the non-social context, activations were situated more anterior relative to the social context, consistent with previous reports of value gradients within the medial prefrontal cortex (Chib, Rangel, Shimojo, and O’Doherty, 2009; Clithero and Rangel, 2014; D. V. Smith et al., 2010).

Discussion

  • The marriage of social and non-social forms of uncertainty into a comprehensive theory of decision-making promises to significantly improve their understanding of human behavior.
  • Another novel aspect of their work is that the authors exploited the endogenous variability in the process of evidence accumulation to identify candidate regions involved in this process.
  • In deploying this EEG-informed fMRI analysis, the authors were able to identify a region of the pMFC correlating robustly with the trial-by-trial changes in the rate of evidence accumulation across both social and non-social contexts, suggesting that both types of decisions are likely to rely on the same implementational and algorithmic process (Lockwood et al., 2017).
  • Correspondingly, their results support the rapidly emerging view that, at least under conditions of increased urgency to make a choice, decisions are embodied in the same sensorimotor areas guiding the actions used to express that choice.
  • In conclusion, their results offer compelling new evidence that social and non-social choices share common neural underpinnings, whereupon domain-specific information is converted into a “common currency” in domain-general valuation areas prior to being accumulated for the decision in medial frontal cortex.

Participants

  • 40 participants were recruited through the The University of Glasgow subject pool.
  • Since facial perception may depend on one’s race and racial history (e.g. Scott and Monesson, 2009), participants were chosen to be Caucasians, aged 18-35 to match the available face stimuli (see below).
  • Two participants were removed due to the poor behaviour (one had near performance across all levels of reward probability in the social context whereas the other had chosen to nearly always ’Play’ across all levels of reward probability in the non-social context) and seven participants due to noisy EEG signals in the scanner leading to poor discrimination performance.
  • The remaining 31 subjects (12 males, 19 females), were included in all subsequent analyses.
  • The study was approved by the College of Science and Engineering Ethics Committee at the University of Glasgow (300180147) and informed consent was obtained from all participants.

Experimental paradigm

  • The experimental design consisted of three parts: 1) an initial behavioral session comprising a rating task and a separate choice task, 2) an online rating task one day prior to the main EEG-fMRI experiment and 3) a rating task and the main choice task during which participants underwent scanning.
  • Correspondingly, the authors told participants that during the main choice task they would assume the role of the Trustee themselves and play with the same face identities they encountered during the ratings tasks (in social trials) or using purely probabilistic gambles (in non-social trials).
  • In reality, the authors used the participant-specific reports on the likelihood of individual face identities splitting the augmented endowment to construct reward probability ranges that were comparable to those used in the non-social contexts (via explicit reward probability values).
  • The main choice task during the initial behavioral session was virtually identical to the one used for the EEG-fMRI experiments, with the main difference being that participants received feedback following each of their choice on a trial-by-trial basis.
  • Furthermore, to motivate participants to engage with the task, the authors told them that in addition to their base rate payment (behavioral session: £6, EEG-fMRI session: £16) they would receive a variable bonus (up to £4) based on the overall points they collected during the experiment.

Choice probabilities

  • To assess the similarity between the probabilities of ’Play’ choices across the social and non-social contexts the authors used a conventional likelihood-ratio test.
  • Specifically, the authors examined whether a single sigmoid curve (Weibull function) would fit the combined social and non-social choice data across the five reward probability levels better than two separate curves (Philiastides and Sajda, 2006).
  • The authors performed this separately for each participant by fitting the best single Weibull function jointly to the two data sets in addition to the individual fits.
  • If λ exceeds the criterion value (for p = 0.05), the authors concluded that a single function fits the data better than two separate domain-specific functions.

Sequential sampling modelling

  • The authors modelled the behavioral data using a special case of the leaky competing accumulator model.
  • For the indecision point (i.e. 0 evidence) the authors included an extra free parameter, bias, to account for potential inter-individual biases towards either ’Play’ or ’Keep’ choices.
  • Secondly, the authors ran a grid search fitting procedure for each participant using a fine-grained parameter space around the estimates obtained in the previous step.
  • Choices and RT distributions were created for each possible combination of the four free parameters from 5000 simulated decision traces for context.

EEG data acquisition

  • The authors used an MR-compatible EEG amplifier system (Brain Products, Germany) to collect the data and Brain Vision Recorder software (Brain Products, Germany) to continuously record EEG at 5000 Hz.
  • The authors lowered the input impedance for each electrode to < 50 kOhm (25 KOhm average across participants).
  • To facilitate the recording of the scanner triggers, the scanner pulses were lengthened to 50 µs via an in-house pulse stretcher.
  • Experimental event codes and participants’ responses were synchronized, and recorded simultaneously, with the EEG data through the Brain Vision Recorder software.
  • Finally, the cabling connecting to the EEG amplifiers at the back of the bore was secured to a cantilever beam to minimize scanner vibration artifacts.

EEG data preprocessing

  • The authors used MATLAB (Mathworks, Natick, MA) to preprocess and analyse the EEG data.
  • This process was repeated for each functional volume in their dataset.
  • Hz band-pass filter in order to remove slow DC drifts and higher frequency noise.
  • The authors recorded the timing of these events and used principal component analysis to identify linear components associated with eye-blinks, which were subsequently removed from the broadband EEG data collected during the main task (Parra, Spence, Gerson, and Sajda, 2005).
  • BCG artifacts share frequency content with the EEG and are therefore more challenging to remove.

Single-trial EEG analysis

  • To identify activity related to evidence accumulation the authors used a single-trial multivariate discriminant analysis (Parra, Spence, Gerson, and Sajda, 2005; Sajda, Philiastides, and Parra, 2009) to discriminate between easy (i.e. reward probabilities 0–0.2 and 0.8–1) and difficult trials (reward probabilities 0.4–0.6) in stimulus-locked EEG data, collapsing across both social and non-social trials.
  • Specifically, for every iteration, the authors used N-1 trials to estimate a spatial filter (w), which they then applied to the remaining trials to obtain out-of-sample discriminant component amplitudes (y(t)).
  • Specifically, the authors identified the time point at which the discriminating activity began to rise monotonically after an initial dip in the stimulus-locked data following any early evoked responses present in the data (onset time mean ± s.e.: 363.097 ± 97.046 for social trials and 376.161ms ± 107.155 for non-social trials).
  • The EEG data X and discriminating components y are now depicted in matrix and vector notation, respectively, for convenience.

Single-trial regression analyses

  • Specifically, the authors used the trial-wise estimates of the slope of the EEG-derived evidence accumulation signal (i.e. y(t)) to predict the probability of playing (1: ’Play’, 0: ’Keep’) on individual trials (Fig. 5c, d).
  • The authors performed this analysis separately for each participant and for each of the social and non-social trials: Pplay = [1+ e−(β0+β1×y(buildup rate))]−1 (8) In all three cases they tested whether the regression coefficients across participants (β1 values in Eqs. 6 7 8) came from a distribution with a mean different from zero (using separate two-tailed t test).

MRI data acquisition

  • Trio MRI scanner (Siemens, Erlangen, Germany) with an 12-channel head coil was employed for the (f)MRI acquisition.
  • Finally, the authors added the motion correction parameters obtained from fMRI preprocessing (three rotations and three translations) as additional covariates of no interest.
  • Thus, the resulting regressors for each participant were different as they were constructed from a random sequence of regressor amplitude events.
  • This was achieved by constructing a null distribution for this joint threshold based on the size of all clusters larger than 10 voxels and with Z-scores larger than |2.57| (i.e. considering both positive and negative correlations) across all shuffled regressors.

Psychophysiological interaction analysis

  • The authors conducted a psychophysiological interaction (PPI) analysis to probe the functional connectivity between the pMFC found to correlate with the trial-by-trial variability in their EEG-informed regressor, and the rest of the brain.
  • To carry out the PPI analysis, the authors first extracted time-series data from group-level activation clusters in the pMFC (seed), separately for each of the social and non-social contexts.
  • The main aim of this analysis was to investigate potential task-dependant associations between the site of evidence accumulation and regions involved in domain-general value computations.
  • The authors expected the relevant coupling to be negative, as easier trials decrease integration times and correspondingly the overall integrated activity (that is, area under the accumulation curve; Fig. 6b).
  • The resulting fMRI statistical maps were corrected based on the threshold derived from the resampling procedure described above.

Author contributions

  • DHA, EF, EDL and MGP conceived and designed the experiments.
  • DHA performed the experiments and collected the data.
  • DHA and MGP analysed the data and wrote the paper.
  • All authors discussed the results and implications and commented on the manuscript.

Did you find this useful? Give us your feedback

Figures (7)

Content maybe subject to copyright    Report

A common neural currency account for social and
non-social decisions
Desislava H. Arabadzhiyska
1,2
, Oliver G. B. Garrod
1,2
, Elsa Fouragnan
3
, Emanuele De
Luca
1,2
, Philippe G. Schyns
1,2
, and Marios G. Philiastides
1,2,*
1
School of Psychology and Neuroscience, University of Glasgow, Glasgow, UK
2
Centre for Cognitive Neuroimaging, University of Glasgow, Glasgow, UK
3
School of Psychology, University of Plymouth, Plymouth, UK
*
Address correspondence to: marios.philiastides@glasgow.ac.uk
ABSTRACT
To date, social and non-social decisions have been studied in isolation. Consequently, the extent to which social and non-social
forms of decision uncertainty are integrated using shared neurocomputational resources remains elusive. Here, we address this
question using simultaneous EEG-fMRI and a task in which decision evidence in social and non-social contexts varies along
comparable scales. First, we identify comparable time-resolved build-up of activity in the EEG, akin to a process of evidence
accumulation. We then use the endogenous trial-by-trial variability in the slopes of these accumulating signals to construct
parametric fMRI predictors. We show that a region of the posterior-medial frontal cortex (pMFC) uniquely explains trial-wise
variability in the process of evidence accumulation in both the social and non-social contexts. We further demonstrate a
task-dependent coupling between the pMFC and regions of the human valuation system in dorso- and ventro-medial prefrontal
cortex (dmPFC/vmPFC) across both contexts. Finally, we report domain-specific representations in regions known to encode
the early decision evidence for each context. These results are suggestive of a domain-general decision-making architecture,
whereupon domain-specific information is likely converted into a “common currency” in the dmPFC/vmPFC and accumulated
for the decision in the pMFC.
Introduction
Most strategic decisions occur under considerable uncertainty. For example, when investing in the stock market, a trader may
use only purely probabilistic models to estimate risk in the market’s fluctuations. In contrast, when negotiating a deal in person,
the trader’s risk assessment may rely instead on how trustworthy the other party appears (Fouragnan et al., 2013; Griessinger
and Coricelli, 2015). Similarly, the decision to undergo a risky surgical operation may use online statistics regarding overall
success rates or the advice of a trustworthy person who has undergone a similar operation.
In standard economic utility models (Morgenstern and Von Neumann, 1953), the rules governing such decisions are the
same, regardless of whether the source of uncertainty is social or non-social in nature (e.g. a human or an on-line platform).
Recent advances in social neuroscience, however, have focused instead on identifying neurocognitive processes that might be
uniquely social (Rilling, King-Casas, and Sanfey, 2008; Suzuki and O’Doherty, 2020; van Baar, Chang, and Sanfey, 2019).
Correspondingly, two competing accounts have recently been proposed to explain how the brain might encode uncertainty
representations underlying social and non-social choices (Ruff and Fehr, 2014).
The first account assumes that the brain dedicates largely separate networks for encoding social and non-social forms of
decision uncertainty and for assigning value to different choice alternatives. In contrast, the second account proposes that the
same network processes the different forms of uncertainty and converts the values associated with different choice alternatives
into a “common currency”. In order to arbitrate between these accounts, both the algorithmic (i.e. what computational
mechanisms are involved) and implementational (i.e. which brain regions are involved) levels need to be considered (Lockwood,
Apps, and Chang, 2020).
To date, however, there is no unified framework for integrating social and non-social decisions as most studies have evaluated
them in isolation. Studies of non-social decision-making have focused primarily on constructing a mechanistic account of choice
behavior using sequential sampling models in which evidence accumulates stochastically to an internal decision boundary (Hu,
Domenech, and Pessiglione, 2020; Hunt et al., 2012; Sepulveda et al., 2020). Recent brain imaging work has implicated the
posterior-medial frontal cortex (pMFC) in this process of evidence accumulation, proposing a domain-general role for this
region in perceptual and value-based choices (Cona, Marino, and Semenza, 2017; Pisauro, Fouragnan, Retzler, and Philiastides,
2017; Zhang, Hughes, and Rowe, 2012).
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted October 18, 2021. ; https://doi.org/10.1101/2021.10.18.464762doi: bioRxiv preprint

While some recent work on social decision making started to explore the utility of such accumulation-to-bound models
(Chen and Krajbich, 2018; Krajbich et al., 2015; Suzuki and O’Doherty, 2020) no direct comparisons have been made
between the mechanisms and neural representations of social and non-social choices. Here, we design a novel task in which
decision evidence in social and non-social contexts varies along comparable scales. Across contexts, we test whether there is a
common embedding of decision evidence as well as a common mechanism for integrating this evidence using simultaneous
electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) (henceforth EEG-fMRI).
In doing so, we identify in both contexts centroparietal EEG signals exhibiting decision dynamics consistent with a common
process of evidence accumulation. Consistent with such domain-general mechanisms, the trial-by-trial temporal variability
in these accumulating signals is reflected in the fMRI data in the region of the pMFC previously implicated in other types of
decisions (Pisauro, Fouragnan, Retzler, and Philiastides, 2017). Moreover we report a trial-wise and task-dependent modulation
of the pMFC with established regions of the human valuation system in the medial prefrontal cortex (Chib, Rangel, Shimojo,
and O’Doherty, 2009; Clithero and Rangel, 2014; Philiastides, Biele, and Heekeren, 2010), suggestive of decision evidence
embedded within a “common currency” space.
Results
We investigated economic decisions within a social context by exploiting trustworthiness in a partner’s face during a strategic
economic game to generate predictions about possible outcomes (see below) and within a non-social (purely probabilistic)
context by manipulating outcome probabilities in individual gambles. Importantly, we created parametrically modulated stimuli
along comparable scales of reward probability in each of the social and non-social contexts (while keeping reward magnitude
constant across both contexts). The non-social stimuli were associated with a range of pure reward probabilities chosen from
the full probability range (from 0–1), placed on top of a face image (neutral for trustworthiness) to equalize perceptual load
across the social and non-social stimuli (Fig. 1; see Materials and Methods for more details).
Figure 1.
Sample stimuli from a representative participant. Top: Social stimuli at five different (participant–specific) indirect
trustworthiness levels, matching the pure reward probability levels used for the non-social stimuli. For each participant there
were, on average, 28 unique face identities in each of the five reward probability levels. Bottom: Non-social stimuli with five
explicit reward probability levels (given a ’Play’ choice) superimposed on a face neutral for trustworthiness (i.e. 0.5 reward
probability). Photo-realistic face images were obtained using the procedure described in (Gill, Garrod, Jack, and Schyns, 2014)
and summarised in Materials and Methods.
We derived comparable reward probabilities for the social stimuli by asking participants (N=31) to provide indirect
trustworthiness ratings for a series of 150 face identities. Specifically, we framed this rating stage in the context of a trust
game (Berg, n.d.). Usually a trust game involves an interaction between two players, the investor and the trustee. The investor
decides whether to send a monetary endowment to the trustee that gets multiplied by a certain factor (’Play’ option) or to retain
2
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted October 18, 2021. ; https://doi.org/10.1101/2021.10.18.464762doi: bioRxiv preprint

possession of the initial endowment (’Keep’ option). In turn, the trustee can decide whether or not to send a fixed share of the
augmented amount back to the investor so that both parties can benefit from the interaction. We told participants that each face
belonged to people who had previously taken part in a similar study in the role of the trustee and we asked them to indicate the
overall likelihood (in the range 0–1) that each person had returned a fixed share (50%) of the augmented endowment entrusted
to them (Fig. 2a).
This indirect measure of perceived trustworthiness was previously shown to be more ecologically valid compared to explicit
ratings (Uleman and Kressel, 2013) and further ensured that trustworthiness judgments became the product of an economic
decision as in our main experimental paradigm. Specifically, during the main (EEG-fMRI) task participants assumed the role of
the investor in a series of one-shot trust games. In each game they had to decide whether to choose between a small but sure
reward (1 pt; ’Keep’ option) or a bigger, but riskier payoff (2 pts; ’Play’ option). We randomly interleaved non-social trials (i.e.
probabilistic gambles) in which we controlled the likelihood of obtaining the higher payoff with explicit reward probabilities,
matched against subject-specific indirect trustworthiness ratings as highlighted above (Fig. 2b).
We instructed participants that the probability of receiving the higher payoff for ’Play’ choices would be based on the
overall likelihood with which each face identify split the augmented endowment (here 4 pts) in a previous study (social trials) or
the pure reward probabilities depicted on the face stimuli (non-social trials). We sampled the full range of reward probabilities
given a ’Play’ choice using five levels (
P(payo f f |play) = {0 0.2,0.2 0.4, 0.4 0.6,0.6 0.8,0.8 1}
). In social trials,
we populated each reward probability level with face identities based on the subject-specific perceived trustworthiness from
the initial rating stage. Ultimately, these ranges correspond to three broad task difficulty levels; easier trials favouring either a
’Keep’ or ’Play’ choice (i.e. 0–0.2 and 0.8–1, respectively), medium difficulty trials in which the outcome uncertainty for ’Play’
choices begins to increase (i.e. 0.2–0.4 and 0.6–0.8) and difficult trials for the most ambiguous set of reward probabilities (i.e.
0.4–0.6).
Figure 2. a. A variant of the traditional Trust Game in which a participant (Investor) is allocated 1 point and they need to
decide whether to ’Keep’ the point or ’Play’ for the chance of winning 2 points. During ’Play’ choices the 1 point is quadrupled
and passed on to a Trustee, which takes the form of either a social agent (red) or a purely probabilistic gamble (blue). The
Trustee can either split the 4 points evenly and give the participant 2 points or keep all 4 points to themselves (i.e. the
participants receives 0 points). In the social context the probability of winning is based on the trustworthiness of the social
agent displayed in the stimulus, while in the non-social context by the reward probability range displayed on top of a face,
neutral for trustworthiness. b. Social (S; red outline) and non-social (NS; blue-outline) experimental design trials. Each trial
began with a variable duration (1-4 s) fixation cross screen, which served as an inter-trial interval. The fixation screen was
followed by a stimulus screen which remained available for up to 1.3 s, during which participants indicated their choice (’Play’
or ’Keep’). The stimulus screen was replaced by a fixation cross following choice for the remainder of the 1.3 s.
3
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted October 18, 2021. ; https://doi.org/10.1101/2021.10.18.464762doi: bioRxiv preprint

Behavioral performance in social and non-social contexts
Participants’ fraction of ’Play’ choices correlated positively with the overall reward probability across both the social and
non-social trials (Social: t(30) = 17.769, p
<
0.001; Non-social: t(30) = 4.086 p
<
0.001), indicating that they selected the riskier
option more frequently as the likelihood of receiving the higher payoff increased (Fig. 3a). More importantly, we demonstrated
that choice behavior was comparable across the social and non-social trials. Specifically, we used a likelihood-ratio test (see
Material and Methods) to show that a single sigmoid function fit the fraction of ’Play’ choices (jointly across both conditions)
as well as two separate functions (λ(30) = 0.551, p = 0.759).
The mean response times (RTs) as a function of the overall reward probability exhibited an inverted ‘V’ relationship, across
both the social and non-social trials (Fig. 3b), consistent with a positive relationship with task difficulty (Social: t(30) = 9.302,
p
<
0.001; Non-social: t(30) = 10.105, p
<
0.001). In other words, we observed the longest RTs for the most difficult trials
(reward probabilities 0.4–0.6), the shortest RTs for the easiest trials (reward probabilities 0–0.2 and 0.8–1) and intermediate
RTs for medium difficulty trials (reward probabilities 0.4–0.6 and 0.6–0.8). The overall RTs showed a small (41.637ms), albeit
significant difference between the social and non-social trials (paired ttest: t(30) = -3.274, p = 0.003), with social trials (
M
S
=
677.864ms, SD
S
= 86.479 ms) being on average faster than non-social ones (M
NS
= 719.502ms, SD
NS
= 91.287 ms).
Figure 3.
Social and non-social behavioral responses (red and blue circles) versus modelling performance of a drift diffusion model (black
crosses) for proportion of ’Play’ choices (
b
) and response times (RTs;
c
). ’Play’ responses increased with probability of reward given a ’Play’
choice (P(payoff|play)) and RTs were the highest when there was no strong evidence for or against ‘Play’ decisions. Participant-specific
behavior presented in grey circles.
Evidence accumulation in social and non-social contexts
Having established comparable behavioral performance across the social and non-social contexts, we asked whether these share
a common underlying mechanism for integrating relevant decision evidence. To address this question, we first leveraged the
high temporal resolution of the EEG data to identify signals exhibiting a gradual build-up of activity consistent with a general
process of evidence accumulation (EA) (Pisauro, Fouragnan, Retzler, and Philiastides, 2017; Polanıía, Krajbich, Grueschow,
and Ruff, 2014). We hypothesize that if such signals exist, we should observe reliable ramp-like activity with a build-up rate
that is proportional to the amount of decision difficulty.
We tested this hypothesis with a single-trial multivariate linear classifier (Parra, Spence, Gerson, and Sajda, 2005; Sajda,
Philiastides, and Parra, 2009) designed to estimate spatial weightings of the EEG sensors that discriminate between easy
vs. difficult trials (see Materials and Methods). Applying the estimated electrode weights to single-trial data produced a
measurement of the discriminating component amplitudes (henceforth
y
). These amplitudes represent the distance of individual
trials from the discriminating hyperplane that we treat as a neural surrogate for the relevant decision variable that is being
integrated.
Persistent accumulating activity with a build-up rate proportional to the amount of decision difficulty should result in a
gradual increase in the classifier’s performance while the traces for the easy and difficult trials diverge as a function of elapsed
4
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted October 18, 2021. ; https://doi.org/10.1101/2021.10.18.464762doi: bioRxiv preprint

time in stimulus-locked data (Fig. 4a). To test the extent of a domain-general EA process across social and non-social contexts,
we performed this discrimination by initially collapsing trials across both conditions. We treated the medium difficulty trials as
“unseen” data (independent of those used to train the classifier), to more convincingly test for a full parametric effect on the
build-up rate associated with the different levels of decision difficulty.
Figure 4. Linear discriminant analysis of the EEG. a
. Build-up rates for hypothetical evidence accumulation (EA) signals
for easy (yellow) and difficult (green) trials (top) and how differences in the rate of EA could manifest in the accuracy of an
EEG classifier trained on stimulus-locked data. b. Average discrimination performance (Az; using leave-one-out cross
validation) between easy and difcult trials across participants along with histogram of participant-specific peak discrimination
times (top). The dashed line represents the average Az value leading to a significance level of p=0.05, estimated using a
separate bootstrap test. The thinner black lines indicate standard errors of the mean across participants. Insets: scalp
topographies (forward models) of the discriminating activity estimated at time of maximum discrimination averaged across
participant for the social (red outline) and non-social (blue outline) trials.
c
. The average temporal profile of the discriminating
activity across participants (obtained by applying the participant-specific classification weights estimated at the time of
maximum discrimination) for the three levels of decision difficulty for social (red) and non-social trials (blue), locked to the
onset of the stimulus onset. Insets: histograms of participant-specific EA onset times for social (red) and non-social trials
(blue). d. The average temporal profile of the discriminating activity across participants, realigned to the onset of EA as
estimated in c, for the three levels of decision difficulty for social (red) and non-social trials (blue).
We quantified the classifier’s performance through the area under a ROC curve (i.e. Az value) with a leave-one-trial-out
cross validation procedure. This was done at several time windows locked to the stimulus onset. As hypothesized, the classifier’s
performance increased systematically over time, reflecting the potential divergence in the gradual build-up of activity between
easy and difficult trials (Fig. 4b). On average, the classifier’s performance began increasing after 400 ms post-stimulus (i.e.
after early encoding of the relevant evidence) and peaked several hundred milliseconds later.
The spatial distribution of this discriminating activity (i.e. forward model; see Material and Methods) from participant-
specific windows of maximum discrimination between easy and difficult trials (Fig. 4b, top) revealed comparable centroparietal
topographies across social and non-social contexts (r = 0.896, p < 0.01; Fig. 4b, inset). These similarities are suggestive of
common neural generators across the two contexts, consistent with those reported previously in the perceptual domain (e.g.
Herding et al., 2019; Kelly and O’Connell, 2013; Philiastides, Heekeren, and Sajda, 2014).
To formally characterize the temporal profile of the discriminating activity (i.e.
y(t)
) for each condition separately, we
applied participant-specific spatial weights from the time window of maximum discrimination to an extended stimulus-locked
time window and separately for social and non-social trials. We also applied this procedure to medium difficulty trials (i.e.,
“unseen” data) by subjecting the relevant data through the same neural generators responsible for the original discrimination.
This approach revealed a gradual build-up of activity akin to a process of EA in both social and non-social trials (Fig. 4c;
top: Social, bottom: Non-social). Similar to the classifier performance, the neural activity began to rise around 400 ms after
stimulus presentation in both the social and non-social trials, with the build-up rate being proportional to the amount of decision
difficulty. Note that the build-up rate from medium difficulty trials was situated between the two extreme conditions used to
originally train the classifier, thereby establishing a fully parametric effect across the three levels of decision difficulty (F(2,
5
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted October 18, 2021. ; https://doi.org/10.1101/2021.10.18.464762doi: bioRxiv preprint

References
More filters
Book
01 Jan 1944
TL;DR: Theory of games and economic behavior as mentioned in this paper is the classic work upon which modern-day game theory is based, and it has been widely used to analyze a host of real-world phenomena from arms races to optimal policy choices of presidential candidates, from vaccination policy to major league baseball salary negotiations.
Abstract: This is the classic work upon which modern-day game theory is based. What began more than sixty years ago as a modest proposal that a mathematician and an economist write a short paper together blossomed, in 1944, when Princeton University Press published "Theory of Games and Economic Behavior." In it, John von Neumann and Oskar Morgenstern conceived a groundbreaking mathematical theory of economic and social organization, based on a theory of games of strategy. Not only would this revolutionize economics, but the entirely new field of scientific inquiry it yielded--game theory--has since been widely used to analyze a host of real-world phenomena from arms races to optimal policy choices of presidential candidates, from vaccination policy to major league baseball salary negotiations. And it is today established throughout both the social sciences and a wide range of other sciences.

19,337 citations

Journal ArticleDOI
TL;DR: In this article, the authors designed an experiment to study trust and reciprocity in an investment setting and found that observed decisions suggest that reciprocity exists as a basic element of human behavior and that this is accounted for in the trust extended to an anonymous counterpart.

5,033 citations

Journal ArticleDOI
15 Jun 2006-Nature
TL;DR: It is shown, in a gambling task, that human subjects' choices can be characterized by a computationally well-regarded strategy for addressing the explore/exploit dilemma, and a model of action selection under uncertainty that involves switching between exploratory and exploitative behavioural modes is suggested.
Abstract: Humans are remarkably curious, and that is useful in helping us to learn about new environments and possibilities. But curiosity killed the cat, they say, and it also carries with it substantial potential risks and costs for us. Statisticians, engineers and economists have long considered ways of balancing the costs and benefits of exploration. Tests involving a gambling task and an fMRI brain scanner now show that humans appear to obey similar principles when considering their options. The players had to balance the desire to select the richest option based on accumulated experience against the desire to choose a less familiar option that might have a larger payoff. The frontopolar cortex, a brain area known to be involved in cognitive control, was preferentially active during exploratory decisions. The results suggest a neurobiological account of human exploration and point to a new area for behavioural and neural investigations.

2,003 citations

Journal ArticleDOI
TL;DR: In this paper, the authors consider optimal decision making in two-alternative forced-choice (TAFC) tasks and show that all but one can be reduced to the drift diffusion model, implementing the statistically optimal algorithm.
Abstract: In this article, the authors consider optimal decision making in two-alternative forced-choice (TAFC) tasks. They begin by analyzing 6 models of TAFC decision making and show that all but one can be reduced to the drift diffusion model, implementing the statistically optimal algorithm (most accurate for a given speed or fastest for a given accuracy). They prove further that there is always an optimal trade-off between speed and accuracy that maximizes various reward functions, including reward rate (percentage of correct responses per unit time), as well as several other objective functions, including ones weighted for accuracy. They use these findings to address empirical data and make novel predictions about performance under optimality.

1,693 citations

Journal ArticleDOI
TL;DR: The supplementary motor complex consists of the supplementary motor area, the supplementary eye field and the pre-supplementary motor area and theories regarding their function vary widely.
Abstract: The supplementary motor complex consists of the supplementary motor area, the supplementary eye field and the pre-supplementary motor area. In recent years, these areas have come under increasing scrutiny from cognitive neuroscientists, motor physiologists and clinicians because they seem to be crucial for linking cognition to action. However, theories regarding their function vary widely. This Review brings together the data regarding the supplementary motor regions, highlighting outstanding issues and providing new perspectives for understanding their functions.

1,492 citations

Frequently Asked Questions (1)
Q1. What are the contributions mentioned in the paper "A common neural currency account for social and non-social decisions" ?

To date, social and non-social decisions have been studied in isolation. The authors show that a region of the posterior-medial frontal cortex ( pMFC ) uniquely explains trial-wise variability in the process of evidence accumulation in both the social and non-social contexts. The authors further demonstrate a task-dependent coupling between the pMFC and regions of the human valuation system in dorsoand ventro-medial prefrontal cortex ( dmPFC/vmPFC ) across both contexts. Finally, the authors report domain-specific representations in regions known to encode the early decision evidence for each context. These results are suggestive of a domain-general decision-making architecture, whereupon domain-specific information is likely converted into a “ common currency ” in the dmPFC/vmPFC and accumulated for the decision in the pMFC.