Simple reward-based learning suits adolescents best

Adolescents focus on rewards and are less able to learn to avoid punishment or consider the consequences of alternative actions, finds a new UCL-led study.The study, published in PLOS Computational Biology, compared how adolescents and adults learn to make choices based on the available information.

18 volunteers aged 12-17 and 20 volunteers aged 18-32 completed tasks in which they had to choose between abstract symbols. Each symbol was consistently associated with a fixed chance of a reward, punishment or no outcome. As the trial progressed, participants learnt which symbols were likely to lead to each outcome and adjusted their choices accordingly.

Adolescents and adults were equally good at learning to choose symbols associated with reward, but adolescents were less good at avoiding symbols associated with punishment. Adults also performed significantly better when they were told what would have happened if they had chosen the other symbol after each choice, whereas adolescents did not appear to take this information into account.

"From this experimental lab study we can draw conclusions about learning during adolescence. We find that adolescents and adults learn in different ways, something that might be relevant to education," explains lead author Dr Stefano Palminteri, who conducted the study at the UCL Institute of Cognitive Neuroscience and now works at the École Normale Supérieure in Paris. "Unlike adults, adolescents are not so good at learning to modify their choices to avoid punishment. This suggests that incentive systems based on reward rather than punishment may be more effective for this age group. Additionally, we found that adolescents did not learn from being shown what would have happened if they made alternative choices."

To interpret the results, the researchers developed computational models of learning and ran simulations applying them to the results of the study. The first was a simple model one that learnt from rewards, and the second model added to this by also learning from the option that was not chosen. The third model was the most complete and took the full context into account, with equal weighting given to punishment avoidance and reward seeking. For example, obtaining no outcome rather than losing a point is weighted equally to gaining a point rather than having no outcome.

Comparing the experimental data to the models, the team found that adolescents' behaviour followed the simple reward-based model whereas adults' behaviour matched the complete, contextual model.

"Our study suggests that adolescents are more receptive to rewards than they are to punishments of equal value," explains senior author Professor Sarah-Jayne Blakemore (UCL Institute of Cognitive Neuroscience). "As a result, it may be useful for parents and teachers to frame things in more positive terms. For example, saying 'I will give you a pound to do the dishes' might work better than saying 'I will take a pound from your pocket money if you don't do the dishes'. In either case they will be a pound better off if they choose to do the dishes, but our study suggests that the reward-based approach is more likely to be effective."

###

In your coverage please use this URL to provide access to the freely available article in PLOS Computational Biology: http://dx.plos.org/10.1371/journal.pcbi. pcbi.1004953

Press-only preview: http://blogs.plos.org/everyone/files/2016/06/pcbi.1004953-press-preview-2.pdf

Contact: Name: Stefano Palminteri

Email: Stefano.Palminteri@gmail.com

Citation: Palminteri S, Kilford EJ, Coricelli G, Blakemore S-J (2016) The Computational Development of Reinforcement Learning during Adolescence. PLoS Comput Biol 12(6): e1004953. doi:10.1371/journal.pcbi.1004953

Image Caption: Simple reward-based learning suits adolescents best

Image Credit: ZEISS Microscopy / Flickr

Image Link: http://blogs.plos.org/everyone/files/2016/06/Palminteri-image-1.jpg

Funding: SP is supported by a Marie Sklodowska- Curie Individual European Fellowship (PIEF-GA-2012 Grant 328822). EJK is supported by a Medical Research Council studentship. GC is funded by the European Research Council (ERC Consolidator Grant 617629). SJB is funded by a Royal Society University Research Fellowship. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing Interests: The authors have declared that no competing interests exist.

About PLOS Computational Biology

PLOS Computational Biology (http://www.ploscompbiol.org) features works of exceptional significance that further our understanding of living systems at all scales through the application of computational methods. For more information follow @PLOSCompBiol on Twitter or contact ploscompbiol@plos.org.

Media and Copyright Information

For information about PLOS Computational Biology relevant to journalists, bloggers and press officers, including details of our press release process and embargo policy, visit http://journals.plos.org/ploscompbiol/s/press-and-media .

PLOS Journals publish under a Creative Commons Attribution License, which permits free reuse of all materials published with the article, so long as the work is cited.

About the Public Library of Science

The Public Library of Science (PLOS) PLOS is a nonprofit publisher and advocacy organization founded to accelerate progress in science and medicine by leading a transformation in research communication. For more information, visit http://www.plos.org.

Disclaimer

This press release refers to upcoming articles in PLOS Computational Biology. The releases have been provided by the article authors and/or journal staff. Any opinions expressed in these are the personal views of the contributors, and do not necessarily represent the views or policies of PLOS. PLOS expressly disclaims any and all warranties and liability in connection with the information found in the release and article and your use of such information.

---

---