design study of an agent-based financial self service system
michael svec (cartoonist)
cure is a design-study to show the possibilites of agent-based communication and interaction in the context of self-service banking terminals.
we developed several user interface prototypes for atms with different configurations [display sizes, touchscreen vs soft-buttons, keyboard vs numerical block, functional range].
the congenial cartoonesk interface-agent cureman allows to visualize the different system stati and to mediate technical latencies with the anthropomorphic possibilites of an animated character: his appearance and design, forms of expression, gestures, animations, reactions.
a user's interaction with the system partly takes place as communication with the interface-agent, and partly through context sensitive tools that cureman makes available to the user.
although cureman is mostly missing "intelligence" and own initiative — typical attributes of an "intelligent" agent — he uses some assumptions on and previous experiences with a user. as such cure is a comprehensible, predictable, and controllable interface that gives users the sense of power, mastery, control, and accomplishment ... and is truly enjoyable.
as the user interacts with a character [although a virtual one] the assumption that the system acts in her interests even in her absence becomes plausible, acceptable and desirable. this enables new services such as
- notification when a certain payment arrives
- delaying of a bank transfer until solvency
- monitoring of an "einziehungsauftrag"
in these ways the interface agent enables the user to make use of the extensive knowledge that the bank has always accumulated about her.
the filtering logic in cure: a complex filtering logic is elegantly realised with only two buttons ("show only ..." and "show all") and the area of focus
as any interface agent must pass a kind of anti-turing test in order to be effectively understood and employed by the user we conducted scientific and user research to isolate four major questions to be answered by cureman, our virtual clerk:
- understanding resp. competence: how can a user understand what the agent can do for her and who he is? how can the agent mediate his competences and limits?
- trust: where does the trust come from that the user should put into the agent?
- control resp. autonomy: how can a user control and manage the agent? how much autonomy should the agent have?
- personalizability: how can the agent adapt to a user's needs? how can a user communicate her needs? how [autonomously] can the agent learn from the user to assist her needs?
- personification: in which ways is the agent represented in the interface?
excessive work has gone into integrating error notifications into the interaction: event-driven alerts and non-fatal errors need not disrupt the user in her work immediatly. instead the agent unintrusively signals that he has something to say at the next best situation. only when the user finishes her current task or when some technical timeout demands it, the message is displayed.
fatal errors [which are not repairable by the user] on the other hand are instantly displayed by a seperate error-agent to visualize the concernment of both user and cureman.