AI assistants: does the right to edit encourage adoption?
In SHR — Swiss Human Resources’ programme “Creating your own AI assistants and agents without coding”, design also concerns the user’s role in shaping the final output. The management question is specific: does allowing users to amend a proposal help sustain use of an assistant after a visible error?
What the literature establishes
Berkeley Dietvorst, Joseph Simmons and Cade Massey (2015, Journal of Experimental Psychology: General) show, in forecasting experiments, that participants may abandon an algorithm after seeing it make mistakes, even when it still outperforms the human forecaster. Observing its errors undermines confidence more than observing human errors. Comparative performance alone therefore cannot explain whether people choose to use a tool.
Scope for editing as a mechanism
Berkeley Dietvorst, Joseph Simmons and Cade Massey (2018, Management Science) show that allowing participants to adjust an algorithm’s forecasts, even to a limited extent, increases their willingness to use it. Their findings suggest that having some control over the output can matter more than having complete freedom to change it. For a no-code assistant, this invites a distinction between an editable proposal and an output that can only be accepted or rejected.
The premature conclusion
It might seem to follow that all outputs should be freely editable. Yet these studies demonstrate neither that every human correction improves the output nor that greater adoption guarantees better decisions. Scope for editing is a design hypothesis to test, not evidence of quality. (our executive and employee training programmes)
What the evidence does not allow us to transfer
These experiments concern algorithmic forecasts, not generative assistants integrated into business processes. Adjusting a forecast value is not equivalent to rewriting a case summary or a customer response. In financial services and insurance in particular, the technical ability to edit a text does not establish the authority required to approve it.
A practical check in Zurich
In a Zurich team working in financial services, insurance, tech or a European headquarters, a programme exercise could compare two interfaces for the same assistant drafting internal briefing notes, using fictional or anonymised cases. After showing the same proposal containing a flagged error, randomly assign an interface offering only acceptance or rejection, or one also allowing edits. Then measure voluntary reuse on the next case and have the final notes assessed blindly against a common quality rubric. A higher reuse rate without a decline in quality would provide local support for allowing edits, without establishing its value across every profession. To go further: explore the Building AI Assistants and Agents Without Coding training in Zurich, or browse our executive and employee training programmes in Switzerland.
In pictures: Building AI Assistants and Agents Without Coding in Zurich
- building ai assistants and agents without coding
- Zurich
- research
- training zurich