Note: Day 3 was a holiday!

These points are extracted from a combination of my notes and my memory, both of which are at best only noisy indicators of the true content. I imagine a large amount of editorialising is present. If anyone feels their views are misrepresented I’d be happy to include their comments alongside my orthogonal interpretation, so please do correct me!

Embodiment in visual politics

Manos Tsakiris spoke broadly on the power of images, particularly photographs, to evoke visceral reactions and, by doing so, to incite practical and political responses in viewers. Examples of this power of images were the Abu Grade torture images (provided with the somewhat contestable assertion that everyone knew about the torture but it took the images to galvanise a response) and a picture of a drowned refugee child on a beach. The line of research presented focused on the question of reality judgements - the extent to which people believe that scenes portrayed in images reflect genuine emotions and events as opposed to being staged.

Participants were presented with a set of images, some of which were from real events and others from military and emergency services drills practicing responses to similar events. People rated images more likely to be real if those images were more emotionally impactful, an effect which was stronger for older participants and equally clear across the political spectrum. High alexithymia scores were assocated with lower overall reality judgements. There was a suggestion that alexithymia created an emotional confusion which could be readily manipulated by demagogues willing to undertake the task of labelling the electorate’s emotions for their own ends, which would cast those with alexithymia as resistant to emotional manipulation through images but perhaps more susceptible to manipulation through rhetoric.

We were also treated to a study of the visual portrayal of refugees as a function of political position: pro-refugee positions offered portrayals of small groups of identifiable refugees, while anti-refugee positions tended towards showing a mass of unidentifiable humanity. These portrayals were sensitive to contexts, however, with images of strong young men being used in pro-refugee media where there was a need for immigration to fill low-skilled jobs, while in more saturated employment contexts refugees are more frequently represented as women and children in need of sanctuary.

The talk ended with a call for interdisciplinary study of the way in which political engagement relies on embodied emotional engagement, not only between scientific disciplines such as psychology, neuroscience, and neuroendocrinology, but including the humanities, especially history and art.

Norms, emotions, traits: what determines moral judgement?

Annekathrin Schacht presented an early look at ongoing work investigating the processing of social information and its role in moral judgements. The talk was grounded on the high salience of social information in both bottom-up (e.g. automatic attention capture by faces) and top-down terms.

The theory presented argued that people evaluate morality through observation of actions, but these actions are not necessarily judged as im/moral in themselves, instead being used within the wider person-level context to contribute to an evaluation of a person as im/moral.1 This phenomenon is argued to emerge from a dual-process model of moral judgement in which morally-valanced actions provoke a gut response which is then fed into a slower, context-sensitive judgement process wherein actions which provoke initial condemnation may be excused, and during which the new information is integrated with the prior understanding of the actor as an im/moral person.

The key finding was that people scoring higher on an empathy scale were more likely to give extreme ratings to morally-valanced vignettes (either positive or negative). The use of vignettes was particularly interesting - although the initial set of vignettes used had undergone a laborious process of development and checking, further development was just finishing on an even more comprehensive set of vignettes which provided endings of different moral valence to the same stems (very thorough work from Ronja Demel).

Despite not answering the question posed in the title, the talk was an excellent example of how seminar talks develop, and I was grateful to get to see the work of a competent academic in its early stages,2 and perhaps to contribute to the polishing of the talk for future presentations.

Social learning strategies for matters of taste

Pantelis Analytis gave a complex but informative talk on recommender systems as tools for estimating subjective value of unseen items. Recommender systems have become ubiquitous in recent years, from personalised advertisements and online shopping recommendations through to suggestions for discovering new music, films, and people across a variety of applications.

Pantelis’ presented work used the Jester dataset,3 a corpus of 50,000+ ratings of 100 or so jokes by individual users. The task for the algorithms examined was relatively straightforward: given some indication of a user’s taste, predict ratings for novel items using the ratings provided for those items by other users.

There are a variety of classes of strategies which can be deployed for solving this task. The ones which were given the most attention were doppelganger strategies, clique strategies, and averaging strategies. These all bleed into one another in some ways, and offer different attempts to balance specificity with loss of information: the more one tends towards focusing on the most apparently relevant individuals as predictive, the more information one throws away with the rest of the individuals in the dataset.

While the optimal strategy depends to a great degree on the amount of information available about an individual’s taste (e.g. how many ratings they have provided), a similarity-weighted average of the entire dataset quickly dominates as a strategy and is largely uncontested in the limit. Certain individuals have somewhat unique properties in the data: there are those for whom averaging strategies perform particularly poorly (i.e. individuals whose judgement is decidedly not mainstream); and individuals whose judgement offers a good indicator for a wide range of other individuals, who might be regarded as indicative of trends.


  1. I am reminded of the distinction between guilt (“I have done something wrong”) and shame (“I am wrong”). It seems we may indeed have a hard time loving the sinner while hating the sin. 

  2. So often the perfect is the enemy of the good, and producing a really great talk can seem entirely out of reach for graduate students with experience only of mediocre under/graduate student presentations and top-quality well-rehearsed seminars delivered by academics on the seminar circuit. It’s very encouraging to know that you can start with good and get to great with practice and incremental improvement; you don’t need to come up with flawless from the get-go. 

  3. I really need to download the data and play with it sometime…