P R E D I C T I O N E R R O R S
One System
How one computation unifies neuroscience,
accounting, finance, and artificial intelligence
Spencer Nash
predictionerrors.com
What if there's only one computation?
What if the way neurons process reality,
the way accountants measure performance,
and the way emotions drive behaviour
are all the same operation?
What Neurons Do
Expected − Actual = Prediction Error
Zero error
As expected.
Nothing to learn.
Negative error
Worse than expected.
Update the model.
Positive error
Better than expected.
Update again.
What Accountants Do
Budget − Actual = Variance
A budget is an expectation.
An actual result is an outcome.
The variance is a prediction error.
Profit is a positive prediction error.
Loss is a negative prediction error.
Every financial report ever written is a prediction error document.
Same Computation
Neuroscience
Expected reward − Actual reward
= Prediction Error
→ Update the model
Accounting
Budget − Actual
= Variance
→ Update the forecast
Five Independent Channels
The Emotional Comparator Framework
Resource
Depletion
→
Abundance
−10 to +10
Status
Dismissal
→
Recognition
−10 to +10
Belonging
Rejection
→
Connection
−10 to +10
Values
Violation
→
Integrity
−10 to +10
Novelty
Boredom
→
Fascination
−10 to +10
Two Outputs, Not One
Output 1: The Result
Emotion → Mood
Finance → Return
Output 2: Reliability
Trust the mood?
Trust the return?
A 15% return on stable foundations is a fundamentally different thing from a 15% return on shaky ones. Both systems know this.
The Reliability Function
Operationalises Friston's precision
Reliability = f ( Volatility, Age, Sample Size, Trend )
Volatility
How erratic the signal is.
High volatility → low reliability.
Age
How recent the data is.
Stale data degrades reliability.
Sample Size
How many observations.
More data → higher reliability.
Trend
Direction of change.
Clear trend → higher reliability.
The Same Five Channels in Finance
Resource
Available markets, capacity, funding
Constrained
→
Abundant
Status
Competitive position, pricing power, brand
Weak
→
Strong
Belonging
Customer, employee, supplier loyalty
Churn
→
Loyalty
Values
Governance, compliance, integrity
Failure
→
Strength
Novelty
Market maturity vs emergence
Saturated
→
New
The Channel Mapping
| Channel | Emotional Domain | Financial Domain | Shared Dimension |
| Resource | Depletion → Abundance | Constrained → Abundant | Available inputs |
| Status | Dismissal → Recognition | Weakness → Strength | Capability recognition |
| Belonging | Rejection → Connection | Churn → Loyalty | Relational security |
| Values | Violation → Integrity | Failure → Strength | Standards maintenance |
| Novelty | Boredom → Fascination | Saturation → Newness | Environmental newness |
| Output | Emotional | Financial |
| Result | Mood | Return |
| Reliability | Trust the mood? | Trust the return? |
Neural Grounding
The four variables are directly observable on a spike train
Volatility
Random variation in spike rate
Age
Duration of the spike train
Sample Size
Number of spikes
Trend
Rate increasing or decreasing
Phasic firing encodes the prediction error — the event-driven signal.
Tonic firing encodes the standing expectation — the baseline.
The contrast between them is expected minus actual.
The Large Accounting Model
Period Entry replaces double-entry bookkeeping.
- Transactions tracked over temporal intervals, not arbitrary points
- Co-signature technology, not blockchain
- Reconciliation eliminated — payments at email speed
Financial Assets
Measured in £ / $ / €
Traditional currency on a ledger
Knowledge Assets
Measured as Competence (−10 to +10)
Bankable on the same ledger
Finance & Process Engineering
In an AI world, accountants don't do the books.
They develop the systems that do the books.
Green Belt
Foundations — build your first system
↔ Diploma
Black Belt
Advanced — architect complex systems
↔ Degree
Master Black Belt
Expert — lead, teach, innovate
↔ Masters
What This Means for AI
A nervous system for LLMs
ECF doesn't add emotion. It makes visible what's already there. Prediction errors across five channels, weighted by reliability, accumulated on a ledger.
Dreaming
A persistent memory ledger enables offline processing of accumulated prediction errors. The system updates its own model. Not being trained — learning.
Alignment through interpretability
Every decision traceable to specific channels, weightings, and reliability scores. The ledger is the explanation. No black box.
One Computation
Neuroscience
Prediction errors drive learning
Psychology
Emotions as reliability-weighted PE
Finance
Return + reliability from five channels
Accounting
The original prediction error system
AI
A nervous system with a memory ledger
P R E D I C T I O N E R R O R S
predictionerrors.com
Read the papers
Watch the course
Join the discipline
Spencer Nash
Master Black Belt: Financial, Process & System Development