Our coaches will use models of argument and dialogue to underpin their interactions with the user. The arguments used by the coaches won’t just be to help justify their advice (e.g. “you should go for a run because it is healthy”), but will also be used to help select the appropriate thing to say at a given time.
For example, when a coach is permitted to ask a question, precisely what question should it ask? “Where did you go on holiday?”, “Why not take your dog for a walk?”, “Have you considered phoning a friend?”,…
Each of these questions fulfills a different purpose: asking about a holiday is a friendly conversation-starter; suggesting the user takes their dog for a walk is encouraging physical activity; suggesting to phone a friend encourages social interaction. Also, suggesting they take their dog for a walk relies on them actually having a dog in the first place.
Similarly, the coach itself will have an idea of what it wants to achieve in the dialogue, both overall (as a final goal), and at a specific point in time. For instance, the coach might be focused on improving the user’s social interactions but is less concerned about the amount of exercise they are getting. We can use the coach’s goals combined with the purposes served by each question to determine which of those question(s) would be the most suitable; so in the example above, the coach would suggest phoning a friend.
In many cases, some questions will fulfill more than one purpose and the coach might have more than one goal. This is where theories of argumentation come in: we can reason about which question(s) are most appropriate by considering all the relevant purposes and goals, and any preconditions (e.g. owning a dog is a precondition on walking a dog).
Consider the problem like this: a coach needs to choose one and only one question to ask.
If we frame the ability to choose each question as an argument, it follows that those arguments contradict each other – after all, if we choose the question “Where did you go on holiday?” we can’t also choose the question “Why not take your dog for a walk?”. We’re not saying the questions themselves contradict each other: just the coach’s ability to choose a question.
That presents a problem, however: because all the possible choices contradict each other, no argument “wins”. This is where we bring in the purposes fulfilled by each question, and the coach’s goals and determine a set of preferences over those arguments.
Preferences help make a choice between two contradictory outcomes: we choose the one that is most preferred, regardless of any conflict between them. We determine preferences based on which of the questions, in terms of their purposes, is most closely aligned to the coach’s goals.
In our example, if the coach’s goals are more closely aligned to “Where did you go on holiday?” than “Why not take your dog for a walk?”, the conflict from the latter to the former is ignored, leaving the holiday question as the coach’s choice.