Production
Sustaining a defined level of new knowledge production.
Research programme · Epistemic K-point
Keep the gains from AI.
Understand what sustains them.
AI can help a team investigate, compare and produce more. Can the team keep doing that work if AI assistance is withdrawn? What would it take to maintain or recover its performance? These are the questions behind K-point.
Research in development
Doing more, and being able to keep it up
Imagine a team that used to complete ten usable reports a week. With AI it completes twenty; during an outage it still manages twelve. Some of the extra output is lost, but the team is doing more than before. That does not show that people have lost their previous skills. If the team can barely complete a sound report during the outage, we need to investigate what has changed and how it can recover.
Such a comparison also needs to account for differences between tasks, practice and the time it takes to switch ways of working.
We want to understand how AI helps people produce more useful work, and how to sustain that improvement. That means counting the time spent checking, correcting and handing work over.
The working definition
A K-point is a measurable transition at which removing continued AI participation causes a defined knowledge function to fall below its previously sustainable level of performance, except at materially higher cost, delay, loss of capability, or loss of recoverability.
This is our working definition. We still need to investigate whether such a threshold exists and how it could be measured.
The focus is a team’s or organisation’s ability to produce, assess and use knowledge.
Four aspects to investigate
We provisionally distinguish four functions. They may develop differently; K1 to K4 do not indicate a sequence or a score.
Validation
Sustaining checking, comparison, synthesis and assessment of knowledge.
Agenda formation
Investigating AI’s influence on which questions are selected and investigated.
Recovery and reconstitution
Rebuilding a knowledge function within agreed limits after prolonged dependence.
What do we want to investigate?
Before each study, we specify the work being assessed, the required quality and the acceptable time and cost of recovery. We use four reference points:
| Reference | Meaning |
|---|---|
| B · Before AI | The earlier level on comparable tasks without AI. |
| A · With AI | The level with the selected AI workflow, including human checking. |
| R · After withdrawing further AI assistance | The level after withdrawing further AI assistance, with sources and notes specified in advance. |
| T · Specified minimum | A threshold for the knowledge function specified before measurement. |
B describes what the team could do before; T is the minimum set in advance. That minimum may refer to the old level or to a new level that the team has sustained. The study must make this clear beforehand. A drop after withdrawing AI is not enough, on its own, to establish a K-point.
Switching to another AI provider may reduce dependence on one supplier. That is a different comparison from continuing without AI.
What might challenge this explanation?
- The change is gradual rather than a meaningful threshold.
- The comparison mainly measures task differences, practice or temporary switching costs.
- Previous capability is retained or improves; only additional scale is lost.
- Recovery succeeds within the agreed limits, or existing concepts adequately explain the result.
Research status
An initial literature review and a proposed study design are available. The dry walkthrough on 12 September 2026 used input constructed by ChatGPT, without participants or human timing measurements. It identified four ambiguities in the exercise material. This is protocol preparation, not an empirical participant study. No participant findings are recorded in the available dossier.
The next step is to develop the measures and specify and approve a pilot protocol. We also need to test whether K1–K4 are useful distinctions and what K-point adds to existing research.
Find the K-Spot
At KopKen, the programme goes by the playful name Find the K-Spot. Stories and discussions bring its questions to a wider audience. ArtiProof provides the research background.
K-point is a separate research programme alongside the ArtiProof method. The cases below offer starting points for exploring its questions.
Visit KopKen →Read further and contribute
Do you work on AI dependence, knowledge production, checking or recovery? Sources, counterexamples and research proposals are welcome.
Get in touch about K-pointContent basis: Epistemic K-point — Research Programme v0.1 (11 September 2026), supplemented by the brainstorm, v0.2 follow-up and dry run of 12 September.