Presenter
Bradley Love, UCL
Bradley is a Professor of Cognitive and Decision Sciences in Experimental Psychology at UCL and a fellow at The Alan Turing Institute for data science and AI, as well as the European Lab for Learning & Intelligent Systems (ELLIS). His lab's research centers around human learning and decision making, integrating behavioural, computational, and neuroscience perspectives. Currently, they are focused on large-scale modelling of brain and behaviour using deep learning approaches, as well as using large language models to create BrainGPT, a tool to assist neuroscience researchers.
Summary:
Potentially disruptive findings are overlooked due to the rapid expansion of the scientific literature. We propose a human-machine teaming solution in which machines assist humans in integrating vast scientific literatures. While we focus on neuroscience, our approach applies broadly and we encourage its adoption across science. Our tool, BrainGPT, will be trained to capture data patterns in the neuroscience literature, taking advantage of recent advances in large-language models. Researchers can prompt BrainGPT with proposed study designs for which BrainGPT will generate likely data patterns reflecting its current synthesis of the scientific literature. Other uses include instant meta-analysis and identifying anomalous findings. Beyond its use as a tool, BrainGPT can shed light on the structure of neuroscience as a field. Importantly, BrainGPT will not summarise nor retrieve articles. In such cases, large-language models often confabulate, which is potentially harmful. BrainGPT is an open-source community effort with 1337+ volunteers.
Challenges:
People are limited in their ability to integrate vast quantities of information. Unfortunately, knowledge intensive endeavours, such as neuroscience, require that capacity. AI solutions are needed to assist humans. Human-machine teams can uncover connections and make discoveries not otherwise possible.
