Research

[SUMMARY] AI is set to transform research systems, but turning its potential into practical, responsible use remains challenging. While AI has shown great promise in fields like biology and medicine, many applications in research struggle to deliver value and may introduce risks. To address this, RoRI launched the GRAIL project, partnering with 13 global funders to explore how AI is currently used in research funding and how shared knowledge can support more ethical and effective adoption. This handbook, Funding by Algorithm, offers insights from that work—not as a call for automated funding decisions, but as a critical resource to help funders navigate AI thoughtfully and responsibly.

[SUMMARY] Postdocs remain largely invisible in academia and the Science of Science field. I wrote a commentary on Duan et al.’s large-scale study, which highlights key factors for postdoc success, including high-impact publications, topic shifts, and international mobility. I argue that despite these insights, structural challenges like job insecurity, low pay, and immigration hurdles persist. Addressing these issues is essential to support postdocs as vital contributors to research.

[SUMMARY] Conjectures about the weak replicability in social sciences have made scholars eager to quantify the scale and scope of replication failure for a discipline. Yet small-scale manual replication methods alone are ill-suited to deal with this big data problem. Here, we conduct a discipline-wide replication census in science. Our sample (N = 14,126 papers) covers nearly all papers published in the six top-tier Psychology journals over the past 20 years.

[SUMMARY] Scientists have invested enormous resources into replicating existing findings. One replication study on average consumes 300 days. In response, we propose a new machine learning approach of estimating a study’s replicability from the texts in its original manuscript using algorithms. This method could assess the likelihood of replication success for any study with a manuscript within seconds. 

[Summary] This research investigates the implication of expressing one’s authentic vs. idealized personality on social media. The degree of authenticity is measured by the proximity between an individual’s self-reported personality (true self) and the personality inferred from Facebook Likes and status updates (expressed self). Combining a lab experiment and online big data, we found that authentic self-expression is associated with greater well-being. 

[SUMMARY] This project answers a classic question: Are romantic partners or friends similar in personality? Decades of research using self-reports of personality found no evidence. However, I identified a bias called the reference-group effect in the self-report method, which obscures similarity among close companions. To circumvent this bias, I assessed participants’ personalities based on their Facebook behaviours. The results showed that couples, or friends, are more similar in personality than previously thought.

[SUMMARY] This research presents machine learning models that could infer people’s personalities from their social media profiles. Our analysis, based on a large database of Facebook users, showed that personality inferred from digital behaviours is more accurate than human judgements. This work demonstrates the potential of machine algorithms as psychological assessment tools. Algorithm-based methods are automated and therefore cheaper, more reliable, and more scalable.