Experiment
Organisational Memory AI
Move from finding records to reconstructing why Defra believes something and why a decision was made.
01
Problem to address
Records may survive while context does not. Years later, staff and users may know what outcome was recorded but not why, which evidence was considered, which rules applied or what changed afterwards.
02
What the evidence tells us
Farmers keep their own evidence in case of a challenge years later. Officers need a record they can defend. When a decision is a decade old, it is often hard to reconstruct why it was made.
03
The proposition to explore
The idea
Use connected evidence and AI to reconstruct what happened, why a decision was made, which rules applied and what changed afterwards.
04
Hypothesis
If important decisions can be reconstructed with the evidence, policy version and chronology attached, Defra can reduce defensive record-keeping, improve fairness and make institutional knowledge resilient to staff and system change.
05
How it could work
Connect decision records to evidence, actors, applicable rules, effective dates and later changes. AI provides a natural-language reconstruction while always linking back to the source record and stating where evidence is missing.
06
Example: bring it to life
Imagine asking:
“Why did we approve this intervention in 2028?”
AI reconstructs the evidence chain based on policy/rules applicable at the time:
- application
- land condition
- officer assessment
- decision
- reasoning
- supporting evidence
- subsequent changes
And answers:
“The intervention was approved under version 4.2 of the scheme rules because conditions A, B and C were met. The decision relied on the following evidence. The rule was subsequently changed in 2030.”
This takes REMEMBER much further than document storage. It's institutional memory with provenance.
The dairy farmer's defensive archive becomes much less necessary if both farmer and RPA can reliably reconstruct:
What did we know? What was agreed? What rules applied? Why did we decide that?
07
AI opportunity
AI retrieves and assembles distributed evidence, compares policy/rule versions, explains a decision history and cites the underlying sources. It should not invent reasoning that was never recorded.
08
Potential value
- Faster case review and escalation
- More transparent and fair decisions
- Reduced reliance on personal archives
- Safer use of AI over historical cases
09
Reduce uncertainty without overcommitting
Smallest meaningful experiment
Prototype on a small set of historic cases where the real decision trail is known. Ask staff to reconstruct the case manually and with the prototype; compare completeness, time and confidence.
10
Open questions and risks
What we still need to learn
- AI cannot recover reasoning that was never recorded
- Retention must align to liability and policy
- Sensitive information may require fine-grained access