Research

How applying basic psychological virtues could help advance science

Science and spirituality are often not compatible, because spirituality usually entails believing in God or a supernatural being that often does not reside in the materially oriented scientific way of thinking.

Image (From Canva @prerna-madans-team)

Religions were primarily developed to guide people in their everyday matters, and to pass down values of harmony and not division, such as compassion, acceptance, humility, kindness, forgiveness, and refraining from injury or revenge against others. Over time, the application of these virtues in the smallest matters of life has diminished. However, if we return to practising psychological virtues now, I believe great progress can be made, including in science. I will illustrate this using an example from my work.


Since 2018, together with my colleagues at UCL, Oxford University and the University of Edinburgh, I have been working on a meta-analysis (which in the words of Richard Dawkins – is “an analysis of analysis”) determining the association between oral hormonal contraceptives (OCs) and depression in healthy women. Following the strict criteria, we included 14 studies with a total of 2.5 million pill users. The results of the analysis showed that in users of contraceptive pills, the incidence of being diagnosed with clinical depression increased by 27%, and the incidence of being prescribed antidepressants increased by 25% compared to non-users. Depression is one of the most commonly reported side effects of any hormonal contraceptives (see the Lowdown), including pills, patches, intra-uterine devices, such as coils, injections, implants, and rings.


Now let’s look at what we found over so many years of delving deeper into the topic:


Very early in our research, we collated a list of prescribed contraceptive pills, and their patient information leaflets. We noted that depression or anxiety are reported in ‘1 in 10’ or ‘Very Common’ categories for almost all types of contraceptive pills. However, we also noted that the 2011 Cochrane Review on this topic has not mentioned depression a single time. The omission of this important data remains unclear and questionable, especially from a body such as Cochrane.
We also noticed how language around the reported side effects in more recent pills has been changed. When reviewing the patient information leaflet for the latest contraceptive pills (e.g., Opill), the terms depression or anxiety are not listed, but are replaced with the term “nervousness”. There are several plausible explanations for this. One is that the trials were conducted on women who already had these as pre-existing conditions (depression, anxiety), in which case, these don’t need to be categorised as a new side-effect. A second possible explanation is that all mental health effects were combined into a single descriptor “nervousness”, which is reductive of the border and more serious mental health side effects of contraception.


A lot could be said for the widespread push-back and lack of acknowledgment of women’s experiences regarding their health, even by the choice of scientific methods. Perhaps one of the most important discoveries while performing our meta-analysis was how little a layperson would know about the methods used in the studies. Almost all studies that showed no effect between OCs and depression used unclear methodological choices. For example, three (21%) of studies treated women as “non-users” if they have been on birth control for less than one month, three months, or six months. This is ambiguous because women become users of contraception the very day they first take the pill or a hormonal contraceptive. This is almost unimaginable in any other type of drug testing. Furthermore, these women were  grouped into “non-users”, which could vastly underestimate the incidence of depression in OC users.


In yet another large study, the methodological decisions became unclear when the authors excluded the category of women who had symptoms of depression combined with symptoms of anxiety. The reasoning seemed to be that women should show only one or the other form of the disorder, again something that is not entirely clear and runs counter to modern diagnostics, which treats the respective diagnoses as two sides of the same psychological coin. In adolescents who have the strongest association between depression and hormonal contraceptives (and as data suggests, persisting and irreversible), it is a common occurrence to experience depression coupled with anxiety. But again, why are such choices made in research that should be sensitive and not omissive of the very experience they are striving to study?


Our meta-analysis was not the first, but the second in the world on this topic. A group of researchers had already performed a meta-analysis in 2021 using a type of statistical method that, for reasons too complex to list here, is completely inappropriate for the data studied. For example, the authors in this meta-analysis made statistical conclusions, unsupported by the data – the authors claimed: “We provide quantitative evidence on experimental data that hormonal contraceptive use is relatively safe regarding the effect on depressive symptoms”, but the authors provided no evidence on the safety of hormonal contraceptives because they did not analyse any safety data! This sort of ambiguity should have been picked up by the reviewers at the least.


If we consistently apply the principles that are not favourable to human experience and wellbeing, and even negate them, we only drive ourselves deeper into confusion and to put it bluntly – “hell”. It has been over 60 years since the pill was first created, and the evidence for hormonal contraception causing a great deal of mental and physical suffering to many of its users is incontrovertible and overwhelming. In the words of Viktor Frankl, “The ultimate freedom given to human beings, at any moment in life, is one’s attitude in any given circumstances. Life is not primarily a quest for pleasure, as Freud believed, or a quest for power, as Alfred Adler taught, but a quest for meaning.” What, then, is the meaning of pretending we cannot see the extent of true suffering for contraceptive users? If we didn’t pretend up until now, we could have been at a different reality of contraceptive science. And this is how I see that “spirituality and science as one”.


Our meta-analysis is due for submission in The Lancet Obstetrics, Gynaecology, & Women’s Health in May 2025.

 

Intergenerational Fairness and its role in Research Culture

The Royal Society has led the way on research culture in recent years, establishing  the following definition, which is now widely adopted:

Research culture encompasses behaviours, values, expectations, attitudes and norms of our research communities.  It influences researchers’ career paths and determines the way that research is conducted and communicated.

You can find out more about the Society’s activities in this area. Recently they produced a video illustrating what research culture is and why it is important:

The University of Edinburgh published its own Research Culture Action Plan, and supporting Delivery Plan in 2023.

Both within the University and more widely much of the focus of research culture development is on early career researchers. In contrast, there has been less attention paid to those towards the end of their career and the role that they can play in creating a vibrant and inclusive research environment. In 2022 Professor Veronica van Heyningen, who has held leadership positions at UCL and the University of Edinburgh, led a Royal Society project on Changing Working Lives, which particularly looked at the roles and opportunities for researchers at different career stages. The project took into consideration general changes in the research environment and the impact of the pandemic on working practices.

We are all aware of demographic changes that are taking place in society, with rising numbers of older people and a falling birth rate. This project sought to understand the implications of these changes within academia, and in particular, on research culture. A key outcome of the project was the need for explicit consideration of  the responsibilities of older researchers and intergenerational fairness.

In the last decades there have been several changes that have led to longer active careers for older researchers such as the fact that there is now no fixed retirement age in most universities, flexible and part-time working has become more accepted and readily available for all, and the benefits of people staying healthy and active for longer are widely recognised. In the academic setting the continued participation of  experienced researchers has considerable benefits for the scientific community and our shared endeavours.

Nevertheless, sensitivity and awareness are needed to ensure that this “older” generation is supporting and generating opportunities for the “younger” generation, rather than becoming a block on their careers. This is particularly true when, as now, universities are working under situations of limited resource. Funding, PhD students, working space and positions of responsibility are all vital for early and mid-career researchers to establish themselves, but they may not have the credentials to compete directly with late career researchers.

The Royal Society Changing Working Lives project highlighted these issues and suggested that intergenerational fairness called for action from different stakeholders in research. Researchers themselves should consider how best to advance science. Particularly for those later in their career, this wider consideration should start to take precedence over advancing their own career. Funders should consider selection processes and the role of track record in funding decisions. Universities  should consider special support systems and resources for mid-career researchers to enable them to step up to the demands of senior leadership roles, then creating opportunities for early career researchers too.

Diverse teams have been shown to be more effective and creative in many circumstances, and diversity should include consideration of age and experience. But it should not become the default that the most senior member of the team should be the leader. These team members are likely to bring invaluable skills and experience, but in the long term these skills may be best used to focus on technical aspects of the problem whilst mentoring and supporting a less experienced colleague in the position of leader. It is essential that we maintain a flow of talent and provide sufficient resource for those in their early and mid-career for that talent to grow and flourish.



Equality, diversity and inclusion in AI research – why should we care, and what can we do about it?

Research in AI is an increasingly exciting and fast-paced environment, with many new interesting features and applications available at a wider scale. However, it is also the topic of heavy criticism for often failing to represent and serve minority groups, which have historically been underrepresented in conversations about technology. Being PhD students in the CDT in NLP, we think it is extremely important to keep up with issues regarding equality, diversity and inclusion (ED&I), both to improve our own work but also to be critical about new advancements in the field.  

Because of that, we are currently hosting a reading group in ED&I once a month, open to all postgraduate students and staff from the School of Informatics.  

Anyone involved can choose a paper which they think is of interest, no matter whether it is their own work or not. Although attendees are encouraged to read the paper beforehand, this is not a requirement as we start with a ~15 min presentation. Afterwards, an informal group discussion follows, which allows everyone to comfortably express their ideas and ask questions. For the past few months, the sessions have had a very friendly atmosphere and we have learned a lot from each other about how to be more mindful researchers.  

Through the ED&I reading group, we’re hoping to raise some awareness on how issues relating to equality, diversity and inclusion can impact current AI research, but also how AI research can have consequences in areas which have a direct or indirect impact on society. We also aim to foster a welcoming and inclusive environment where researchers can share and discuss their ideas on how AI research is impacting our society. We hope that attendees leave with thoughts on how their choices as a researcher can make a difference for people who have often been left out of the conversation about AI and how their choices can change that.  

From the past few sessions, we have learned a lot from all the people who have presented and whom we have shared a discussion with! Our past sessions have covered: 

Our next session will be on Tuesday 30th April and will be covering issues related to the use of deep learning to identify transgender and gender diverse patients from electronic health records (A deep learning approach for transgender and gender diverse patient identification in electronic health records).  

With AI being an exciting and constantly evolving area of research, we believe that issues of equality, diversity and inclusion are more important than ever for researchers to be aware of, even if their own topic of research is not directly linked to them.  

If you are a researcher at the School of Informatics, we hope you’ll join us the last Tuesday of every month from 1-2pm for engaging presentations and fruitful discussions. Let’s all learn from each other! We usually meet in G.03, with the exception of 30th April, where we will meet in IF 1.15. 

Artemis and Ariadna 

Subscribe to inf-edi-reading-group@mlist.is.ed.ac.uk for notifications on next sessions 


Reframing “Bias” in AI Research

As a PhD student in the School of Informatics, I’ve been researching gender bias in language and language technologies.  Time and again, I’m surprised by how simple people try to make biased language.  As Abeba Birhane stresses in her interview on the podcast The Good Robot, not everything can be conceptualized as a straightforward problem with a straightforward solution.  Our cultures are dynamic and complex.  Our languages evolve slowly over decades, but also change rapidly based on our relationships with the people we are speaking to or writing for.  Moreover, language does not exist in a vacuum.   

In the branch of Linguistics called Critical Discourse Analysis, language is studied in its context of use, considering how it legitimizes and maintains power, and how it incites social change. [1]  Nevertheless, AI research approaches bias as a problem to be fixed, as if bias is an error that can be removed from a dataset, or a mistake that a model can be taught to avoid.  In reality, however, bias is an ongoing challenge. 

Bias changes with time, place, and culture.  Bias will always be with us, because there is no universal, neutral, or objective perspective.  We all are shaped by our own unique viewpoint, by our own experiences of the world.  

We need to reframe research questions about bias in data and technology.  Rather than focusing on removing bias, we need to better understand bias.  We need to study how bias comes through in language and other types of data.  We need to consider the risks bias poses and the harms bias may cause.  Researchers such as Abeba Birhane and Kate Crawford are among a small but growing group of people in the computational research community trying to do this.  There is a wealth of research in the Humanities and Social Sciences that the computational research community can look to; people have been studying and theorizing about language and bias for much longer than the existence of AI as a field.  The School of Informatics has been an exciting place for me to research bias in language technologies because I’ve had the opportunity to talk about new ways to approach bias and ethical AI research with fellow PhD students like Nina Markl and Bhargavi Ganesh.

To reframe questions about bias in data and technology, we need a culture shift in the AI field.  Currently, efficiency, convenience, and quantity drive dataset curation and model creation.  Being the first to publish something is highly valued, so gathering data and building models quickly are done at the expense of critical approaches to dataset and model development.  To gather data quickly, language and images are taken from the Internet without consent from the people who own them or are represented in them.  Datasets are evaluated based on how large they are rather than how representative they are. 

Instead, we need accuracy over efficiency, balance and representativeness over convenience, and quality over quantity.  Then we will realize that bias comes not from the model or the data, but from us, people and society.  Then we can focus on changing the power structures that cause harmful biases. 

References

[1] For more on Critical Discourse Analysis, see Analysing Discourse: Textual Analysis for Social Research (Fairclough, 2003) and Uses of Heritage (Smith, 2006).