Interaction Design for Human-Centered AI
Pic CC Hal 9000
Usability is a fundamental principle of good UX.
“Good design is actually a lot harder to notice than poor design, in part because good designs fit our needs so well that the design is invisible.” - Don Norman
As we build AI that becomes more complex (more than simple dyadic human-system relationships), pay extra attention to usability design and the role of humans in those systems. If we really think about it, ML models manifest themselves in the form of features trying to mimic rich human interaction.
Musings on AI Interfaces:
People are forgiven but systems are not- To err is to be human. Would you afford machines the same benefit?
Design for human sociability: Would you want people to interact with AI in the same manner they interact with other people?
Human-AI delegation: Can delegation between humans and AI outperform humans or AI alone? How can humans learn to delegate to an AI?
Timing Humans in the loop is not universal: Some interactions warrant faster humans in the loop intervention than others. The timing also depends on the end-user.
When to use AI:
Content Personalization: Understanding likes, dislikes to recommend content
Image Classification: The ability to recognize patterns in images.
Prediction of future events based on historical training data
Personalization of UX: Understanding user preferences to improve UX.
Natural language understanding: picking up on diction, intonations, and other aspects of human speech disfluency
Detection of the small variances in low occurrence events over time. Learn evolving patterns to detect anomalous events as they emerge.
Completing basic routine tasks using agents or bot: Automating mundane, domain-specific tasks to save human time.
When not to use AI:
Providing static or limited information: For example, a credit card entry form is simple, standard, and doesn’t have highly varied information requirements for different users.
Minimizing costly errors. If the cost of errors is very high and outweighs the benefits of a small increase in success rate, such as a navigation guide that suggests an off-road route to save a few seconds of travel time.
Optimizing for high speed and low cost. If the speed of development and go to market is more important than anything else to the business, including the value that adding AI would provide.
Automating high-value tasks. If people explicitly tell not to automate or augment a task with AI, that’s a good task not to try to disrupt.
Augmentation vs Automation
How do we play to the strengths of both humans and machines?
One large consideration is if you should use AI to automate a task or to augment a person’s ability to do that task themselves.
Human augmentation vs machine delegation.
When designing usability AI there are 2 main approaches:
Approach 1- Taking existing human-human interactions and replacing them with human-machine interactions. This is complicated because one must work through all the ideological, political, and educational palliatives or first-order approximations to arrive at something that even has a chance of being “human-like”.
Approach 2- Take inspiration from rich human interaction but contextualize 2 key interaction qualities:
Attention (foreground vs background)
System Initiatives (reactive vs proactive)
The second method is also complicated since we must account for the needs of a single user and their (often unknowing/implicit) delegation to a system to help fulfill a task/goal.
Products are often designed based on certain user expectations that end up being perceived and used in a completely different way. As creators, you must consider nth order effects as much as possible. Acknowledge your biases and design inclusively.