By Karin Grasenick

The Mind and the Machine

Image courtesy of Pixabay

Neuroscience has led to important insights into how we learn and make decisions. It has inspired machine learning and robotics. Applications already outperform us on classification and logical tasks like playing chess or Go. Machine learning has also been applied in medicine, to diagnose diseases based on large digital data sets. While underlying algorithms are often presented as neutral, when trained with biased data sets, they have been proven to reproduce social biases which might even lead to life-threatening consequences (e.g. when failing to recognize cancer on darker skin accurately) These findings have led to critical discourse revealing unquestioned norms as well as over-simplified models and operationalisation of variables.

Unlike with the brain, machines commonly learn from relatively narrow training data, without the ability to find and understand additional features in the data presented, to reason or to justify decisions made. Although inspired by biology, artificial neural networks differ substantially from our brain, amongst other features of topology and the ability to build new connections enabling unique learning experiences and recovering from injuries. Artificial neurons are not exposed to chemicals and do not respond differently depending on their sex¹.

For neuroscientists, sex can be a key variable that should be addressed not only when studying humans but also when working with animals and cell cultures. For example, studies have demonstrated that cells and the central nervous systems of male and female mice react differently to chemical stressors with potentially severe consequences for drug design and treatments for brain diseases like Parkinson stroke². Neurodegenerative diseases such as multiple scleroses take different courses for women and men³.

Despite the growing body of knowledge on such consequences, which has proven that sex differences are significant from functional brain organization down to the level of cells, science still often fails to consider the sex of cells, animals and humans, not to mention further intersecting diversity traits relevant for humans like ethnicity, age, and gender, to name a few.

This neglect raises questions on scientific excellence and ethical issues. The implications of poor or incomplete neuroscience are particularly high as properties of lived experience and identity arise from the brain. As stated in a recent Neuron article from the HBP: “The analysis of the human brain will provide new insights into the uniqueness of each human being and into the common biological basis of humankind embedded into the evolution of our species.” (Salles et. al., 2019 p. 380)

The “Merit-Go-Round”

Image courtesy of Pixabay

To ignore variables that have proven to be significant is only possible if reviewers and publishers accept research that does so. Publications and citation practices are inherited in scientific communities, reproducing standards and practices. Reviewers are part of these communities and often continue to neglect feedback on the differentiation of variables in terms of sex, gender and further diversity traits⁴. Funding Agencies like the Canadian CIHR, the US American NIH, and the European Commission have therefore developed clear guidelines how to integrate sex and gender in research proposals and publications. Guidelines like SAGER (Sex and Gender Equity in Research), developed by the European Association of Science Editors (EASE) were formulated to improve the quality of science editing accordingly.

Time is Money

One reason to carry out studies without considering sex difference but yet to apply the findings for all sexes, which has been brought forward continuously are costs.

Generally speaking, introducing and testing for new variables leads to the need to increase sample sizes in order to obtain statistical reliable results.

Studying the brain takes place on multiple scales, from the level of brain regions, down to the level of single neurons. The development of complex, multilevel models such as computational brain models are per se time consuming and expensive. They are derived from high resolution images and experiments with the same individual for many sessions, leading to a low number of involved subjects with a large number of experimental trials.

Bias In = Bias Out

Image courtesy of Randall Munroe, xkcd

When theories and models are derived from neuroscience data, we have to reflect that biases have been built in the production and usage of these data, impacting the applications which are based on them. The underlying data sets should therefore be examined carefully. Brain data from minority populations might not be included or adequately represented in the data set to guarantee generalizability to wider populations, which has for example been proven for Alzheimer’s disease⁵. Algorithms have been developed to identify discrimination as a result of biased data sets, even though it has been proven difficult to consider the complex interplay of variables.

The Data should be FAIR

What should be done? Publications should clearly indicate some basic information on the data that has been used to enhance the reproducibility of results. The Human Brain Project offers with EBRAINS a brain atlas service with an interface and workflows for finding, extracting and analysing of neuroscientific data in an anatomical context provided by the reference atlases. Metadata are included according to the FAIR principles. It enables the users to identify key elements in disease onset and progression. Since the data used for models is tracked, it is possible to build model variants based on gender, age, etc., for brain health diagnosis. Detailed information must be included to enable the reuse of produced experimental data for future analysis which might need information not used in the original study.

But how do we come up with novel questions and insights if they were not considered at first hand?

Diversity Trumps Ability

Even though data are available, as well as funding opportunities and a variety of examples which demonstrate that considering inclusivity and diversity in research leads to novel approaches, scientists might get trapped in the well beaten tracks of their communities like an over-trained artificial neural net.

Not surprisingly, the one of the best strategies for taking diversity into account in research is team diversity.

Image courtesy of Pixabay

The diversity of a scientific team might be based on culture, gender, educational background, scientific discipline, and expertise. Most importantly team members should differ in the way they perceive problems and apply different methods and processes to solve them. Groups of such diverse problem solvers can outperform homogenous groups of high-ability problem solvers⁶.

There is only one hook in it: having an inclusive project with diverse perspective means time and effort. Collaboration across different backgrounds and disciplines requires mutual understanding, the ability to bridge the different terminologies, methods and perceptions. Fortunately, toolkits for such inter- and trans- disciplinary endeavors have been developed – and must be applied for diverse teams to be successful, for example when developing a terminology and methodological approach which is understood, shared, and applicable by all disciplines.

Incentives to go

Neuroscience is inherently interdisciplinary, involving biology, physics, mathematics and computer science. The Human Brain Project comprises neuroscience, medical informatics, AI, robotics and more. Each discipline offers examples how interdisciplinary collaboration, the differentiation of models and variables lead to new insights, e.g. when analysing human-robot interactions, individualising medical treatment and ensuring that ethical, social and philosophical issues are an integral part of brain research and the infrastructure provided⁷.

Word Cloud created with the HBP call text Image courtesy of Karin Grasenick

The HBP encourages scientist to explore the benefits of diversity and of an interdisciplinary discourse. It has therefore initiated a workshop held on 26th and 27th of Sept. at the Technical University in Graz Austria for scientists who are interested to further develop their research approach by gaining insights from within and outside their own discipline. In hands-on parts of the workshop, tools for interdisciplinary research and diversity in research will be applied while encouraging a critical discourse on ethics and data protection regulations, which are of specific relevance in neuroscience, protecting the privacy and integrity of patients and test persons.

The HBP further increases the visibility of researchers, who are developing concepts that differentiate gender and further diversity traits via an open call. The most outstanding and promising concepts will be presented at the HBP Summit in February 2019 in Athens.

________________

Karin Grasenick, founder and CEO of convelop holds a PhD in Computer Science and Biomedical Engineering. Her thesis has led to her growing interest and research in inter- and transdisciplinary research, diversity, and equal opportunities in science. She lectures, coaches and supports teams, universities and international projects. In the HBP, she coordinates equal opportunities and diversity measures as success factor for research and innovation.

References