Bayesian modelling of decision-making in the human brain.

Lead Research Organisation: University of Oxford
Department Name: Experimental Psychology

Abstract

Whenever we make a decision, such as where to invest our money, our brains have to weigh up the available options and decide what is best to do. How do we know which option to choose? How do we keep making the best decisions in ever-changing environments? Despite the difficulties presented by competing desires, most humans and other animals are good decision-makers, performing almost as well as the best computers in laboratory experiments. My training in information engineering allows me to understand the best way for computers to make decisions. By using these computer algorithms and comparing their decisions to those made by humans, I will be able to learn how humans understand the available options, and choose which one to take. I will use a technique called FMRI to watch different parts of people‘s brains while they are making decisions. This will let me discover what information people store to help them decide on a course of action.
The outcome of this research will be a description of the human mechanism for weighing options and making decisions. This will facilitate further research into how this decision-making process is altered in disorders such as obsessive-compulsive disorder, psychopathy and depression.

Technical Summary

We aim to describe the mechanism by which humans make decisions when choosing between competing actions. Previous studies have found behaviour to follow rules that generate the most profitable actions. However, these rules are specific to the study at hand, and only apply to a stable environment. We hypothesize the existence of a general mechanism for making decisions, independent of the study in question and applicable also to rapidly changing, or volatile, environments. The objective of this research is to develop a model of this general mechanism. In order to produce optimal behaviour the model should have two separate layers: the perception layer provides a strategy for gathering information and forming an estimate of the current state of the reward environment; the action layer provides a strategy for making decisions on the basis of this estimate. It is by paying attention to the details of the information gathering strategy that we are able to account for decision-making in a volatile or changing environment. We will use Bayesian models to compute the optimal perception of the reward environment and the optimal actions required both to build and to profit from these perceptions. We will observe human subjects performing a decision-making task in both stable and volatile reward environments. With respect to this data, we will compare the performance of our perception and action model against that of simpler rule-based models.
The proposed two-level model requires the brain explicitly to code for perceived features of the environment in the perception layer. For example, sometimes a subject should choose an action not for immediate gain but out of curiosity to increase their knowledge of the environment. In order to judge when this is appropriate, the subject must maintain an estimate of the environment‘s volatility. By modelling the perception layer, we will have access to such hidden parameters, which are reflected only indirectly in the subject‘s actions. We will use predictions from our model to test for the neural coding of such parameters with FMRI. We expect to see BOLD signal correlates of both information-gathering and decision-making processes. Finding neural correlates of the first of these, that have only an indirect influence on action, would support a general perception and action model of decision-making.
The research proposed would provide a set of models that could be used in studies of pathologically altered decision-making in conditions such as obsessive-compulsive disorder, psychopathy, and depression.

Publications

10 25 50
 
Description CONNECT Consortium Of Neuroimagers for the Non-invasive Exploration of Brain Connectivity and Tracts (3m euros across 9 participants)
Amount £140,000 (GBP)
Funding ID 238292 
Organisation European Commission 
Sector Public
Country European Union (EU)
Start  
 
Description Centre grant to establish Wellcome Centre for Integrative Neuroimaging (Behrens, Co-I)
Amount £11,463,085 (GBP)
Organisation Wellcome Trust 
Sector Charity/Non Profit
Country United Kingdom
Start 04/2017 
End 04/2022
 
Description HUMAN CONNECTOME PROJECT ($30m across 9 centres)
Amount £2,100,000 (GBP)
Organisation National Institutes of Health (NIH) 
Sector Public
Country United States
Start  
 
Description LIFESPAN HUMAN CONNECTOME PROJECT in DEVELOPMENT and AGING ($2x14m across 9 centres)
Amount $1,000,000 (USD)
Organisation National Institutes of Health (NIH) 
Sector Public
Country United States
Start 05/2016 
End 05/2020
 
Description MRC Research Grant (Biophysical models of white matter)
Amount £350,000 (GBP)
Funding ID G0800578 
Organisation Medical Research Council (MRC) 
Sector Public
Country United Kingdom
Start  
 
Description Mechanisms of behavioural control (Behrens)
Amount $600,000 (USD)
Organisation James S. McDonnell Foundation 
Sector Charity/Non Profit
Country United States
Start 10/2013 
End 09/2019
 
Description Senior Research Fellowship - Neural mechanisms of behavioural control (Behrens)
Amount £1,968,459 (GBP)
Funding ID 104765/Z/14/Z 
Organisation Wellcome Trust 
Sector Charity/Non Profit
Country United Kingdom
Start 10/2015 
End 09/2019
 
Description Strategic award (Behrens, Co-I)
Amount £2,000,000 (GBP)
Funding ID 098369/Z/12/Z 
Organisation Wellcome Trust 
Sector Charity/Non Profit
Country United Kingdom
Start 09/2012 
End 08/2018
 
Description Wellcome Trust Research Career Development Fellowship (Biophysical mechanisms of goal-based decision-making)
Amount £736,658 (GBP)
Funding ID 088312 
Organisation Wellcome Trust 
Sector Charity/Non Profit
Country United Kingdom
Start  
 
Title Bayesian learning model 
Description It is a model that describes human learning in volatile environments. It underlies at least 5 of my most high profile publications (Nature x1, Sciencex1 Nature Neuroscience x2, Neuronx1) 
Type Of Material Model of mechanisms or symptoms - human 
Year Produced 2007 
Provided To Others? Yes  
Impact Other labs are using the same model +extensions to describe their behavioural and imaging data 
 
Title Diffusion Tractography Software 
Description We have developed software for revealing brain connections using diffusion imaging 
Type Of Material Data analysis technique 
Year Produced 2006 
Provided To Others? Yes  
Impact The software has been downloaded by over 1000 labs and has been cited in over 600 peer-reviewed publications. It is the world's most popular software for analyzing diffusion-weighted MR images 
 
Description Anatomy of the primate brain 
Organisation Emory University
Country United States 
Sector Academic/University 
PI Contribution Ongoing extensive collaboration to analyze white matter anatomy in humans,chimpanzees and macaques. My expertise in diffusion image and tractgraphy is a central part of this collaboration
Collaborator Contribution Diffusion imaging data on Chimpanzees and Macaques
Impact Paper in Nature Neuroscience Jim Rilling Sabbatical in Oxford 2009 leading to pending publications.
 
Description FIL 
Organisation Wellcome Trust
Department Wellcome Trust Cente for Neuroimaging
Country United Kingdom 
Sector Academic/University 
PI Contribution We have been running MEG experiments in London, using to gain from the experiences of the London MEG team
Collaborator Contribution I am running new studies at UCL collaborating with Karl Friston, Ray Dolan and the MEG team.
Impact None as yet
Start Year 2009
 
Description Monkey Electrophysiology 
Organisation University College London
Department Institute of Neurology
Country United Kingdom 
Sector Academic/University 
PI Contribution I have been analysing Steve Kennerley's single unit data with my models. I have been advising on the design of new experiments
Collaborator Contribution It has given me access to single unit electrophysiology in monkeys.
Impact We have 2 papers in submission
Start Year 2008
 
Description Radio 
Form Of Engagement Activity A press release, press conference or response to a media enquiry/interview
Part Of Official Scheme? No
Primary Audience Media (as a channel to the public)
Results and Impact I have been interviewed on BBC Radio 4 three times and on the BBC world service. I have performed written interviews for Reuters international Media group. My research has appeared on the BBC news website. I have written articles for the new scientist

Interest
Year(s) Of Engagement Activity 2008,2009,2010