There is increasing pressure on social media companies to reduce the spread of misinformation on their platforms. However, they would prefer not to be the arbiters of truth as the truth can be subjective or otherwise hard to determine. Instead, they would prefer that social media users themselves show better discernment when deciding which information to share. Here we show that allowing people to share only those social media posts that they have indicated are true significantly improves sharing discernment, as measured by the difference in the probability of sharing true information versus the probability of sharing false information. Because it doesn’t require social media companies to be the arbiters of truth, this self-censorship intervention can be employed in situations where social media companies suspect that individuals are propagating misinformation but are not sufficiently confident in their suspicions to directly censor the individuals involved. As such, self-censorship can usefully supplement externally imposed (i.e. traditional) censorship in reducing the propagation of false information on social media platforms.


Self-Censorship Appears to be an Effective Way of Reducing the Spread of Misinformation on Social Media

Sampling assumptions --- the assumptions people make about how an example of a category or concept has been chosen --- help us learn from examples efficiently. One context where sampling assumptions are particularly important is in social contexts, where a learner needs to infer the knowledge and intentions of the information provider and vice-versa. The pedagogical sampling assumptions model describes a Bayesian account of how learners and providers should behave given different assumptions they have about the other (e.g., is the provider trying to deceive or help me? Does the learner trust me?). In this study, we tested how well this model could describe learning behaviour in the rectangle game, where a fictional information provider revealed clues about the structure of a rectangle that the learner (a participant) needed to guess. Participants received clues from either a helpful information provider, a provider who was randomly sampling clues, or one of two kinds of unhelpful providers (who could mislead but could not lie). We found that people learned efficiently and in line with model predictions when the provider was helpful and that this was the case even when no cover story was provided. However, although participants could identify that unhelpful providers were not being helpful, they struggled to learn the strategy those providers were using.


Inferring the truth from deception: What can people learn from helpful and unhelpful information providers?

Core vocabulary is a topic of huge interest in linguistics and has been studied from a wide variety of perspectives, such as language learning, dictionary studies, and cross-linguistically. In many of these conceptions, word frequency is widely considered the conventional measure of a word's coreness; however, this approach overlooks important aspects of mental representation like centrality in an associative semantic network. In this experiment, we compare different approaches to defining core words in a task that involves predicting missing words in sentences. Results showed that core words (regardless of definition) were easier to guess than non-core words, but that frequency-defined ones did not perform as well as expected given their higher predictability and the nature of the task. Analysis of incorrect responses also showed that people preferred to guess core words, simple synonyms, and words that are taxonomically related to the target. The findings suggest that how core vocabulary is defined depends in part on the nature of the task and that aspects of both mental representation and the linguistic environment play an important role.

Word Prediction in Context: An Empirical Investigation of Core Vocabulary

Perceptual training has been shown to be an effective and rapid way of training people to make simple diagnoses using medical images. However, it appears to be less effective at training people to make more complex diagnoses that require non-binary judgements. In the present study, we investigated whether perceptual training could be augmented to make it more effective and what factors limited its effectiveness. In Experiment 1, we created artificial stimuli that were designed to simulate liver ultrasound images to assess perceptual learning for a complex task that involved judgements on a 7-point scale. Whilst performance improved somewhat with training, we found that incorporating annotations into the training provided no benefits. Additionally, contrary to our expectations, training that was structured in a stepped fashion was detrimental to learning. In Experiment 2, we found that perceptual learning in a simple task with shaded disks was most impacted by the extent to which the brightness levels of each disk were discriminable but that attending to multiple locations did not result in a significant cost to performance. Our findings show that augmenting perceptual training does not increase learning and that learning is less when the relevant features are harder to discriminate.

Testing the Effectiveness of Augmenting Perceptual Training With Annotations and Steps in a Difficult Visual Discrimination Task

How and why do people share opinions online? In research conducted offline, the social identity of the audience is a key factor: whether they are composed of one’s ingroup or outgroup affects what people share and why. Do people behave similarly and for similar reasons online? To test this, we put participants (N = 326) in imaginary forums belonging to their ingroup and outgroup. In each, people selected statements to share, along with reasons for doing so. The results showed a high degree of heterogeneity; people shared nearly all kinds of statements with both audiences, for a variety of reasons. However, there were also consistent patterns. Identity expression was the most common reason for sharing to both audiences, but this led to different things being shared to each. To the ingroup, people preferred to share statements expressing ingroup beliefs, while to the outgroup, they preferred statements expressing universal beliefs.

Online communication to the ingroup and the outgroup: the role of identity in the “what” and “why” of information sharing

In many situations where people communicate (e.g., Twitter, Facebook etc), people self-organise into ‘echo chambers’ of like-minded individuals, with different echo chambers espousing very different beliefs. Why does this occur? Previous work has demonstrated that such belief polarisation can emerge even when all agents are completely rational, as long as their initial beliefs are heterogeneous and they do not automatically know who to trust. In this work, we used agent-based simulations to further investigate the mechanisms for belief polarisation. Our work extended previous work by using a more realistic scenario. In this scenario, we found that previously proposed methods for reducing belief polarisation did not work but we were able to find a new method that did. However, this same method could be reversed by adversarial entities to increase belief polarisation. We discuss how this danger can be best mitigated and what theoretical conclusions be drawn from our findings.

What interventions can decrease or increase belief polarisation in a population of rational agents?

Perceptual training leads to improvements in a wide range of simple visual tasks. However, it is still unclear how effective it can be for more difficult visual tasks in real-world domains such as radiology. Is it possible to train people to the level of experts? If so, what method is best, and how much training is necessary? Over four training sessions, we trained medically naive participants to identify the degree of fatty liver tissue present in ultrasound images. We found that both COMPARISON and SINGLE-CASE perceptual training techniques resulted in significant post-training improvement, but that the SINGLE-CASE training was more effective. Whilst people showed rapid learning with less than one hour of training, they did not improve to the level of experts, and additional training sessions did not provide significant benefits beyond the initial session. This suggests that perceptual training could usefully augment, but not replace, the traditional rule-based training that medical students currently receive.

How effective is perceptual training? Evaluating two perceptual training methods on a difficult visual categorisation task

Reasoning beyond available data is a ubiquitous feature of human
cognition. But while the availability of first-hand data typically diminishes
as the concepts we reason about become more complex, our ability to draw
inferences seems not to. We may offset the sparsity of direct evidence by
observing the statements of others, but such social meta-inference comes with
challenges of its own. The strength of socially-provided evidence depends on
multiple factors which themselves must be inferred, like the knowledge, social
goals, and independence of the people providing the data. Here, we present the
results of an experiment aimed at examining how people draw conclusions from
information provided by others in the context of social media posts. By
systematically varying the degree of consensus along with the number of people
and distinct arguments involved we are able to assess how much each factor
affects the conclusions reasoners draw. Across a range of topics we find that
while people are influenced by the number of people on each side of an
argument, the number of posts is the dominant factor driving belief
revision. In contrast to well established findings in simpler domains, we find
that people are largely insensitive to the diversity of the arguments made.

Andrew Perfors

8

3

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES