The exchange, announcement, of private information and opinions, which finally inspire action, is at the heart of human social interaction. Flow of information from the most informed party to the least informed one is not as straightforward as it seems. The transmission of information is strictly linked with reputation, influence, and personal traits and inclinations that could impact it.
In this project I turn the focus to the adviser. From an adviser point of view, influence on the advisee has a fundamental impact on the amount and type of information that will be conveyed (Bayarri and DeGroot, 1989). Here is an illustration of this situation, published also in Uta and Chris Frith's blog.
You can read our paper on advice giving here: Hertz U., Palminteri S., Brunetti S., Olesen C., Frith C. D., Bahrami B.; Neural computations underpinning the strategic management of influence in advice giving; Nature Communications, 2017. [Read]

To study this process we developed a novel task to evaluate the strategies that an adviser may utilize when transmitting information. On a trial by trial basis the participant has access to a probability of reward being in one of two locations. He is then asked to advise a client about the location of the reward using a confidence scale. The client has another adviser, and after each trial updates the influence weights assigned to each adviser according to the outcome and confidence reports. In the beginning of each trial, the advisor (i.e. participant) is informed if the client has chosen to follow him or his peer. Our behavioural results indicate that in line with computational predictions of Bayarri & Degroot (1989), participants employ a strategy aimed at maximising their influence on the client.

This projects concerns with a commonly experienced frustration of working with incompetent but overconfident colleagues. If only there was a magic potion for curing their hard-headed assurance and make them pay attention to those who might know better. This popular sentiment point to a very important question: what are the biochemical substrates of the brain processes involved in social interaction and collaboration? Is it possible to tweak and interfere with humans interactions by, for example, changing their hormonal states?
You can read our paper on the effect oxytocin on collective decisions here: Hertz U., Kelly M, Rutledge R. B., Winston J., Wrigh N., Dolan R. J., Bahrami B.; Oxytocin Promotes Collective Decision Making; PLoS ONE, 2016. [Read]

In a previous study we showed how testosterone have devastating effect on collective decision making, by increasing group members’ egocentric bias (Wright et al, 2012). Here we employed a recently developed laboratory model system for uncertainty-ridden collective decision making (Bahrami et al. 2010) and tested whether oxytocin promote collective decision making. We employed a double-blind placebo-controlled design to examine the collective benefit accrued by pairs of participants engaged in a perceptual detection task together. Most notably, we observed that under oxytocin, during disagreements the more competent dyad member insisted on their opinion more vigorously and the less competent partners showed greater propensity to change their mind. As a result, we observed an advantage in collective benefit under oxytocin. Importantly, this hormonal effect on social interaction was delayed and role-dependent. It was mediated by attenuation of the more competent dyad member’s performance over time which resulted in increased within dyad behavioural similarity and enhanced collective benefit. Furthermore, during disagreements oxytocin resulted in a greater tendency for the worse member to change his mind, while better members kept their original choice more often. We suggest that this role dependent effect is mediated by enhanced monitoring of own and other’s performance level under oxytocin, leading to efficient and beneficial collaboration.

A basic tenet of research on wisdom of the crowds is the independence of voters’ opinions before votes are aggregated. However, we often look for others’ opinions before casting our vote. Such social influence can push the groups towards herding, leading to “madness of the crowds”. To investigate the role of social influence in joint decision making, we had dyads of participants perform a visual odd-ball search task together. In the Independent (IND) condition participants initially made a private decision. If disagreeing, discussion and collective decision ensued. In the Influence (INF) condition no private decisions were made and collective decision was immediately negotiated. Dyads that did not accrue collective benefit under IND condition improved with added social influence under INF. In Experiment 2, covertly, we added noise to one of the dyad members’ visual search display. The resulting increased heterogeneity in dyad members’ performances impaired the dyadic performance under IND condition (Bahrami et al., 2010). Importantly, dyadic performance improved with social influence under INF, replicating Experiment 1. Further analyses revealed that under IND condition, dyads exercised equality bias (Mahmoodi et al., 2015) by granting undue credit to the less reliable partner. Under INF condition, however, the more reliable partner (correctly) dominated the joint decisions. While social influence may impede collective success under ideal conditions, it can help the group members overcome factors such as equality bias, which could potentially lead to catastrophic failure.
You can read our paper on the effect of social influence on collective decisions here: Hertz U., Romand-Monnier M., Kyriakopoulou K., Bahrami B.; Social Influence Protects Collective Decision Making from Equality Bias; JEP-HPP, 2016.PMID:26436525, or [here]

In this project I examine how the variances of actions’ outcomes affect decisions and decisions' confidence. People tend to avoid uncertainty, picking a more certain lottery even when its expected reward is low. Confidence in choices also follows uncertainty as people tend to be less confident when outcome uncertainty increases. However, in everyday situations such uncertainties are usually not explicitly available, and have to be tracked on a trial by trial basis, for example when choosing a route for work every morning.
You can read our paper on stochastic satisficing here: Hertz U., Bahrami B., Keramati M.; Stochastic satisficing account of choice and confidence in uncertain value-based decisions; PLoS ONE, 2018 (accepted), Bioarxiv: [Read]

Here we look choices and confidence ratings made by participants in a two armed bandit task. We examined four scenarios, in all of which mean expected reward of one option was constant and higher than the other. The variances of the rewards changed across our four conditions and could be high or low in a factorial 2x2 design. Participants were instructed to choose one of two doors in each trial and state how confident they were in choosing the door with the higher reward on a scale of 1-6. Participants’ probability of choosing the good option was highest when both options had low reward variances, and lowest when both options had high reward variances. When choosing the better option confidence ratings were dependent on the good option’s variance: lower ratings with high variance. When choosing the bad option confidence ratings were low and similar across all conditions. A threshold probability matching model was fitted to the data, showing that participants’ choices are driven by the probability of each option to yield high rewards, taking into account both mean and variance of rewards. Importantly, this model’s estimation best explained the confidence ratings pattern. We suggest that choices and confidence stem from the same probabilistic estimation process, and that this process is updated in a trial by trial basis to accommodate new samples and information.

Check out our recent poster presenting this project at RLDM 2015 here.

Publishing experiments online becomes more and more common in psychology. It has the advantage of not wasting time of the participants and experimenters on recruitment and travel, and the ability to reach diverse populations, beyond undergraduate psychology students. Building and running experiments using the browser also have the huge advantage of an ever-growing amount of tools developed by web designers and programmers, and highly active online community to provide you with tips and solutions to your frustrations. It is therefore extremely useful tool even for lab based experiments.
When I started building my online experiments, moving from other platforms like Cogent and Presentation, I collected bits and pieces from around the web and was finally able to get a functioning web based versions of my experiments. I thought that it will be useful to share what I learned by writing a tutorial. My tutorial on how to build an online two armed bandit experiment is here. You can also play with my fully functional (but not collecting data) demo of a two armed bandit task.