Thammasat University students interested in sociology, computer science, demography, climate change, public administration, and related subjects may find it useful to participate in a free 29 August Zoom webinar on What can an Open, Computational Approach Bring to the Social Sciences?
The event, on Tuesday, 29 August 2023 at 2:30pm Bangkok time, is organized by the Faculty of Social Sciences, University of Hong Kong (HKU).
The TU Library collection includes several books about different aspects of computation and the social sciences.
The speaker will be Dr. Charles Rahal, Senior Departmental Research Lecturer at the Leverhulme Centre for Demographic Science, University of Oxford, the United Kingdom.
The event webpage explains:
Abstract:
Dr Charles Rahal will firstly introduce key concepts related to Open, Reproducible, and Responsible Research. Such concepts – such as the use of free and open-source software (FOSS) – make substantial contributions to ensuring the maintenance and validity of the scientific record. He will then discuss a range of different topics and projects which use original, unstructured, and freely available data sources in conjunction with open and custom-built software libraries to generate new insights – both inferential and predictive – into old and new problems across the (social) sciences. The subdomains of interest are varied, but typically span civil society, social care, climate change, public administration, and the scientometric evaluation of research (and research policy) at scale. The seminar will discuss the development of exciting new tools built in FOSS which allow better visualisations of uncertainty, and conclude with discussions of a replication project regarding the use of pseudo-random number generation, all in the interest of enhancing the scientific record.
Students are invited to register for the event at this link:
https://hkuems1.hku.hk/hkuems/ec_regform.aspx?guest=Y&UEID=89945
With any questions or for further information, please write to
tcfa99@hku.hk
In an article he coauthored last year, The rise of machine learning in the academic social sciences, Dr. Rahal and colleagues noted:
Machine Learning (ML) is gradually revolutionizing the social sciences as it has done for subjects like genomics and medicine. The new millennium brought an ambition to find the ‘Signal and the Noise’, followed by funding initiatives such as the creation of a working group in Computational Social Science by the Russell Sage Foundation. All aim to capitalize on ML’s ability to find intricate patterns; patterns which might have otherwise been missed in the traditional approach to model building. Figure 1 quantifies the ‘rise of machine learning’ via a regular-expression based search across all social science abstracts hosted on Scopus at the time of writing, calculating the prevalence of key words pertaining to ML over time. Growth in the use of (and discussion and debate around) ML methods in the immediate past has been remarkable; from 0.63% between 1960-2017, to nearly quadruple since (2.34%). We provide three explanations for this recent trend, and rationales for an even more optimistic view of the future:
- Historical Ideologies: Social Scientists have previously had a preoccupation with parsimonious explanation and inferential ‘beta-hat’, as opposed to predictive ‘y-hat’ questions. However, the value of predictive algorithms is increasingly appreciated. The Fragile Families Challenge aimed to generate a better understanding of social determinism, but not every emergent application need be survey based. The use of optical character recognition (OCR) for digitizing archival population records and the prediction of history are prime examples of other recent and exciting applications of what ML makes possible. There are substantial public policy applications and opportunities for intervention based upon prediction, too; if we can more accurately predict rain tomorrow, we can better plan to bring an umbrella. There is also the essential realisation that ML can help with causal questions, and complement and improve classical tools designed for inference, especially important given the rise of ‘Explainable Artificial Intelligence’ (XAI). The meticulous focus within ML on limiting over-fitting of the data also provides welcome encouragement for a renewed emphasis on reproducibility.
2. Training and Accessibility: Comparatively less attention has been paid to the development of ML skills for graduate social science candidates. Most degree-granting institutions – with exceptions such as the Oxford Internet Institute’s ‘MSc in Social Data Science’, and the University of Chicago’s ‘Masters in Computational Social Science’ – maintain little emphasis on the training of ML skills. However, global initiatives like the Summer Institute in Computational Social Sciences and the data and software ‘Carpentries’ have emerged. Combined with the proliferation of ever increasing accessible ML libraries, this partially resolves concerns that such courses in advanced analytics (to overcome the ‘epistemological challenges’ of finding small patterns in ‘Big Data’) were ‘not exactly your standard degree at the university’.
- Data and Computing: Constraints due to small-scale datasets and the ‘curse of dimensionality’ that have hampered social scientists in the past are rapidly changing, too. This is due to the enormous growth in large longitudinal surveys, long-term biobanks, and the availability of other administrative and unstructured ‘hidden’ data. Combined with substantial advances in high performance computing capacity (and the prospects of quantum computing more generally), this will allow social scientists to go beyond classical methods which were – in part – designed with computational limitations in mind.
However, the social science community still has an important role to play. We must acknowledge that many of the ground-breaking yet, by now, more ‘classical’ methodological advances that occurred across the 20th century were made with wholly different restrictions in place: we should embrace new methodological trajectories accordingly. Social scientists need to actively ensure that ambitions which have been central to our discipline are maintained in our further development of ML methods, such as through a continued emphasis on explainability and causal reasoning. Immense care also needs to be taken to ensure that the algorithms which we develop are fair and unbiased. Unacceptable levels of bias have already been observed in criminal justice and healthcare, and are quickly emerging in the area of recruitment, all acting in a way which amplifies existing biases and inequalities within society. Indeed, there have already been more than reasonable high profile arguments ‘Against Prediction’ in certain settings, unless it can be done in a socially responsible way. Alongside all relevant ethical concerns regarding individual-level prediction, we call for further theoretical work that attempts to understand what the ‘predictive ceiling’ of social variables substantively represents as we further eliminate reducible error. If we take these steps, we might postulate that the use of ML in the academic social sciences is at the beginning of a sharp incline across the technologist’s S-Curve. Indeed, social scientists may be beginning a wholesale change in the nature of the research process, or – at the very least – are moving from a ‘peak of inflated expectations’ to a ‘plateau of productivity’.
(All images courtesy of Wikimedia Commons)