Framework for Computational Persuasion

We have recruited two research associates to work on our new EPSRC funded project that started in March 2016 and runs for 36 months.

Project summary

Persuasion is an activity that involves one party trying to induce another party to believe something or to do something. It is an important and multifaceted human facility. Obviously, sales and marketing is heavily dependent on persuasion. But many other activities involve persuasion such as a doctor persuading a patient to drink less alcohol, a road safety expert persuading drivers to not text while driving, or an online safety expert persuading users of social media sites to not reveal too much personal information online. As computing becomes involved in every sphere of life, so too is persuasion a target for applying computer-based solutions.

Many of the current persuasion technologies for behaviour change (e.g. for encouraging healthier life styles) are based on some combination of questionnaires for finding out information from users, provision of information for directing the users to better behaviour, computer games to enable users to explore different scenario concerning their behaviour, provision of diaries for getting users to record ongoing behaviour, and messages to remind the user to continue with the better behaviour.

Interestingly, argumentation is not central to the current manifestations of persuasion technologies. The arguments for good behaviour seem either to be assumed before the user accesses the persuasion technology (e.g. when using diaries, or receiving email reminders), or arguments are provided implicitly in the persuasion technology (e.g. through provision of information, or through game playing).

So explicit consideration of arguments and counterarguments are not supported with existing persuasion technologies. Yet in real-world persuasion, in particular in applications such as behaviour change, presenting convincing arguments, and presenting counterarguments to the user's arguments, is critically important. For example, for a doctor to persuade a patient to drink less alcohol, the doctor has to give good arguments why it is better for the patient to drink less, and for how it is possible.

In this project, we intend to bring argumentation into a new generation of persuasion technologies. An automated persuasion system (APS) is a system that can engage in a dialogue with a user (the persuadee) in order to persuade the persuadee to do (or not do) some action or to believe (or not believe) something. To do this, an APS aims to use convincing arguments in order to persuade the persuadee.

The dialogue may involve moves including queries, claims, and importantly, arguments that are presented according to some protocol. The dialogue may be asymmetric since the kinds of moves that the APS can present may be different to the moves that the persuadee may make. For instance, the persuadee might be restricted to only making arguments by selecting them from a menu (in order to obviate the need for natural language processing of arguments being entered). In the extreme, it may be that only the APS can make moves. Whether an argument is convincing depends on the context and on the characteristics of the persuadee. An APS maintains a model of the persuadee, and this is harnessed by the strategy of the APS in order to choose good moves to make in the dialogue.

Computational persuasion is the study of formal models of dialogues involving arguments and counterarguments, of user models, and strategies, for APSs. The overall goal of this project is to develop a formal framework for computational persuasion. This framework will extend recent developments in computational models of argument. The emphasis will be on APSs that will help users in changing behaviour (e.g. to persuade the user to drink less, or to not text while driving).

Project hypothesis

The hypothesis of this project is that we can implement APSs for entering into dialogues with users on specific topics, and that through argumentation, they can be shown to have a reasonable success rate in some persuasion goal (i.e. that a reasonable proportion of the users are persuaded by the arguments and therefore do the desired action or accept the belief).

Project approach

In order to understand the nature of persuasion sufficiently, and in order to ensure that our APSs are well-behaved, computationally viable, etc, and to ensure that they can be implemented via well-understood and efficient processes, we need to develop a formal framework for computational persuasion. The project builds on developments in computational models of argument, but it will go significantly beyond the state of the art in computational models of argument by developing protocols, persuadee models, and strategies for argument-based systems that would be appropriate for behaviour change.

In particular, the current state of the literature does not adequately offer the following (and for which we will develop solutions): (1) A formalization of domain knowledge for behaviour change; (2) Protocols for that take account of human unable to make rich input (since we are not supporting free text input for the persuadee); (3) Persuadee models that allow the persuasion system to construct a model of the users beliefs and preferences, to qualify the probabilistic uncertainty of that model, and to update that model and the associated uncertainty as the dialogue progresses; (4) Strategies for persuasion that harness the persuadee model to find optimal moves to make at each stage (trading the increase in probability of successfully persuading the persuadee against the raised risk that the persuadee disengages from the dialogue as it progresses).

Therefore there are a number of theoretical developments to be produced in the project. The project will also deliver a prototype workbench for engineering APSs for running on web sites and as mobile applications. This will be a very significant development for the field, as it will then enable us to undertake field studies with APSs.

Project objectives

The objectives of the project can be summarised as follows: (1) Develop a theoretical framework for computational persuasion; (2) Develop a range of specific options for specifying APSs; (3) Implement a prototype workbench for specifying and evaluating APSs; and (4) Evaluate APSs in specific domains in behaviour change for healthcare.

Timeliness of the project

Research in computational models of argument has resulted in a promising range of formalizations. Furthermore, there is a valuable opportunity to develop persuasion technology that undertakes argumentation for a wide range of applications such as in behaviour change. However, in order to harness computational models of argument for APSs, there is a need to further develop the field (as outlined in the Current State of the Art section) in order to ensure the persuasion is well-behaved and computationally viable. Now is an ideal time to bridge this gap.

Measuring success of the project

The success of the project (at 36 months) will be measured in terms of whether we have developed a theoretical and computational framework for persuasion that will allow us to produce prototype APSs that can be used in systematic trials in the year after the project has finished. We therefore need to develop a sufficiently robust and rich framework for use in our case study APSs such that through the evaluation within the project we can convince our collaborators that our technology is ready for more comprehensive evaluation.

Project scope

The project will draw on, and contribute to, developments in computational models of argument, knowledge representation and reasoning (using logic and probability theory), and multi-agent systems, for incorporation in artificial systems that support rational decision making.

Project collaboration

The principal investigator for the project is Anthony Hunter, and the project will be based in the Intelligent Systems Group (an internationally leading group in artificial intelligence including computational models of argument) in the UCL Department of Computer Science.

The project involves a close collaboration with other parts of UCL including:

  1. UCL eHealth Unit which develops and evaluates web and mobile interventions for public and patients for problems such as alcohol, sexual health, and type 2 diabetes (Dr Fiona Hamilton)
  2. UCL Centre for Health Informatics & Multiprofessional Education which has considerable expertise in the development and evaluation of technologies for healthcare including behaviour change technologies (Dr Henry Potts)
  3. UCL Department of Infection and Population Health which has considerable expertise in epidemiology, prevention and treatment of infections (Prof Andrew Hayward)
  4. UCL Centre for Behavioural Change which is a new initiative for bringing together researchers from across the college interested in developing and evaluating methods for behavioural change (Prof Susan Michie)

Publications

More information

For more information, contact Anthony Hunter (anthony.hunter@ucl.ac.uk)

www.computationalpersuasion.com