← Every exhibit
Why must human-AI workflow design begin with discrete decisions rather than tools?
| Work field | Design question | Possible human contribution | Possible machine contribution |
|---|---|---|---|
| Task | What is the unit of work being assigned? | Define the decision and its boundary. | Process a specified search or comparison. |
| Contribution | Which strength is scarce at this step? | Context, intuition, judgment, empathy, and responsibility. | Analysis, pattern recognition, speed, and repeatability. |
| Handoff | When does one actor's output become another's input? | Interpret uncertainty and decide whether the output is usable. | Return a traceable result in the agreed format. |
| Coordination | How are disagreement and timing handled? | Ask for clarification, override, or escalation. | Surface alternatives, anomalies, or confidence information. |
| Accountability | Who can explain and own the action? | Accept responsibility for the decision and its effects. | Provide evidence of the processing step and its limits. |
| Failure mode | What happens when the output is wrong or incomplete? | Detect contextual mismatch and correct course. | Flag low coverage, uncertainty, or a pattern outside scope. |
Swipe or scroll horizontally if the table is wider than your screen.
Reference & Evidence
Source: Table from this essay. Sources and interpretation are given in the article.
Related exhibits
-
The human-machine division-of-labor card
From the essay Human and machine should divide the work
-
Why must algorithmic evaluation begin with task structure rather than system capabilities?
From the essay Algorithm trust changes with the task
-
Why must advertising measurement design begin with commercial decision thresholds?
From the essay More advertising observations do not guarantee better decisions