Description
An abstract wireframe of the moment a user invokes an AI function, used to generate AI feature ideas and let data scientists judge feasibility, backed by a quick difficulty check.
Designers and data scientists often struggle to discuss AI ideas: designers draw detailed screens, while data scientists need to know the input, context and expected output. Yang's notebook is a deliberately abstract wireframe of the moment of invocation: the user selects some content, chooses an intelligent function, and output appears somewhere, with optional extra inputs. Trigger and layout details are left vague so discussion stays on which functions are valuable. Overlaid on a wireframe of the product with its contexts marked, the notebook reads to data scientists as a model specification: the surrounding content as potential training data, the selection as runtime input, the output as the modelling goal. Ideas are then checked on quick difficulty axes (for language features: text length, whether the task is classifying, understanding or generating, labelling effort and the chance of finding matching training data), and infeasible ideas are reframed as other standard machine learning problems such as search or question answering.
Attributes
- Discipline
- Design & Product Practice
- Loss type
- Siloed
- Goal
- Generate ideas, Prioritise
- Stage
- Definition, Design
- Data
- Qualitative
- Effort
- Days
Steps
- Draw the abstract notebook: content selected, intelligent function chosen, output location and any extra inputs.
- Express each candidate idea as an instance of the notebook, marking input, context and expected output.
- Review the instances with data scientists, asking whether available or obtainable data are enough and what extra data would help.
- Rate each idea on a few difficulty axes agreed with the data scientists (for language features: text length, task type, labelling effort and availability of matching training pairs).
- Reframe infeasible ideas as other standard machine learning problems, such as search, recommendation or question answering, and generate new variants.
- Feed in user research about what data users' interactions could supply, and iterate the shortlist.
Sources
- Profiling Artificial Intelligence as a Material for User Experience Design (Yang, 2020)
Also known as
- The notebook
- Abstract interaction wireframe for AI
- NLP feasibility axes
- Reframing as canonical ML problems
Categories
- Design
- Design & Product Practice