Second-Order Cybernetics: What Happens When HR Watches Itself

60-Second Summary

Heinz von Foerster (1974): second-order cybernetics is 'the cybernetics of observing systems' — the observer cannot be removed from the system being observed.
Applied to HR: every measurement is also an intervention. Engagement surveys change engagement. Nine-box grids reshape talk about people. Ratings redefine what performance means. There is no clean measurement.
Consequence 1: instruments become norms. Whatever HR measures, employees optimise for. This is Goodhart's Law fed by second-order cybernetics.
Consequence 2: the HR function is inside the culture it reports on. HR cannot report an 'objective' culture score; it can only report what it sees given its own position, incentives, and history in that culture.
Design implication: choose measurements you'd be happy for people to optimise for; state the observer effect openly; and build feedback loops where the observed shape the instrument.

A company runs its annual engagement survey. Scores drop. Leadership responds with a 'listening tour', a new intranet, and a mandate that every manager improve their team's score by 5 points next year. The following year scores go up. Actual engagement — measured indirectly through voluntary attrition, referral rates, and internal mobility — gets worse. The instrument moved. The thing being measured didn't. This is a classic second-order cybernetics failure: the observer forgot they were inside the system, and their measurement became an intervention that reshaped what it measured.

First-order vs second-order cybernetics

“A brain is required to write a theory of a brain. From this follows that a theory of the brain, that has any aspirations for completeness, has to account for the writing of this theory. And even more fascinating, the writer of this theory has to account for himself.”
— Heinz von Foerster, 'Cybernetics of Cybernetics' (1974)

First-order cybernetics — Wiener, Ashby, early Beer — studies controlled systems from an assumed outside vantage point. The thermostat regulates temperature; the engineer stands outside the loop. Most HR analytics implicitly assumes this: HR observes employees, culture, engagement, performance from a neutral vantage.

Second-order cybernetics, developed by Heinz von Foerster, Humberto Maturana, Francisco Varela, Margaret Mead, and Ranulph Glanville in the 1970s, says that stance is a fiction for any social system. The observer is inside the system. The act of observing changes it. The categories the observer uses (engagement, performance, potential, culture) don't discover reality — they construct it. Any adequate theory of the observed system must account for the observer's position in it.

This isn't philosophical flourish. It has concrete operational consequences the moment HR starts measuring anything.

Where second-order effects hit HR

Six places the observer is inside the system

1
Engagement surveys
The survey signals what leadership cares about. Employees learn to answer strategically (manager will read this), managers coach teams before surveys, scores become a KPI managers optimise for. The instrument shapes the phenomenon. What you're now measuring is 'survey behaviour under a KPI regime', not engagement.
2
Performance ratings
A 5-point scale doesn't discover a distribution of performance — it forces the population into 5 buckets. Once ratings drive comp and promotion, they define what performance means to employees. The scale is the definition.
3
9-box grids and talent reviews
The grid is not a passive readout of talent — it's a set of categories that shape how leaders talk, remember, and decide about people. The category 'high potential' begins to describe the ways someone shows up in meetings after they're placed there.
4
Culture assessments
HR is part of the culture it assesses. What HR notices, ignores, or asks about is shaped by HR's own position, history, incentives, and relationships. A fully external assessor is closer to first-order, but the moment their findings are shared, they too become internal.
5
DEI dashboards
Which categories are tracked (gender/ethnicity/tenure/disability/etc) and at what granularity shape what problems become visible and which stay invisible. The dashboard is the ethical frame, not a neutral mirror.
6
Manager effectiveness scores
Rating managers by upward feedback changes what managers do — often toward being pleasant rather than being effective. The instrument's presence changes managerial behaviour before it measures it.

Goodhart

'When a measure becomes a target, it ceases to be a good measure'

Marilyn Strathern's 1997 formulation of Goodhart's Law — the operational face of the second-order problem in orgs

~2×

predictive value from open-ended survey questions vs Likert-scale ratings

Common finding in HR analytics practice — Likert items train people to answer strategically

Designing measurements that survive being observed

First-order HR vs second-order HR

First-order HR (unaware observer)

Reports engagement/culture/performance as 'facts'
Adds measures whenever a question arises
Sets numerical targets on soft measures ('lift eNPS by 10')
Treats decline in a metric as decline in the thing
Denies the observer effect ('we run it identically each year')

Second-order HR (aware observer)

Reports metrics as jointly produced by observer + observed
Adds new measures reluctantly — each is an intervention
Avoids numerical targets on constructs; uses direction + narrative
Cross-checks metric moves against behavioural signals (attrition, referrals, mobility)
Names its own position in the culture it reports on

Six design rules for second-order-aware HR

1
Ask: what will people do if they know we measure this?
If the honest answer is 'they'll game it', either don't measure it, don't publish it, or accept the gaming as part of what you'll now be measuring.
2
Triangulate constructs with behavioural signals
Never let a single Likert-based construct drive a decision. Engagement scores + attrition + referrals + internal mobility together resist gaming better than any one.
3
Prefer measures where the incentive to game equals the incentive to improve
Attrition, time-to-fill, offer-accept rates, internal mobility — you can only 'game' them by actually being better. Prioritise these over pure perception scales.
4
State the observer effect openly
When reporting a survey, say 'this reflects how people responded, knowing the survey exists and who reads it'. That framing itself lowers strategic-answer behaviour over time.
5
Let the observed shape the instrument
Involve employees in what to measure and how — not once, but as an ongoing loop. This is the second-order move: closing the loop between observer and observed.
6
Retire measures on a schedule
Every measure decays. Sunset instruments after 2–3 cycles unless they've earned renewal. Long-lived instruments become the most gamed and the most decoupled from what they claim to measure.

The practical litmus test

Before adopting any new HR metric, write two sentences: (1) what people will do if they learn we measure this, and (2) whether those behaviours would be a good outcome even if the underlying construct didn't move. If (2) is yes, adopt. If (2) is no, redesign or drop.

FAQ

Frequently asked questions

Isn't this just a fancy way of saying 'Goodhart's Law'?

Goodhart is a symptom; second-order cybernetics is the underlying mechanism. Goodhart tells you the measure decays under pressure. Second-order tells you the observer is inside the system so the measure was never independent to begin with — which is why Goodhart is inevitable rather than avoidable.

So should HR stop measuring anything?

No. It should measure carefully, prefer behavioural over perceptual measures, involve the observed in shaping instruments, triangulate, and admit the observer effect openly. The goal isn't to escape second-order effects — you can't — it's to design instruments that survive them.

How does this relate to Reflexivity (Soros)?

Reflexivity is Soros's economics-flavoured articulation of the same second-order insight: the market is an observing system whose observations change what it observes. Von Foerster's frame is broader but the mechanism is the same.

What's the connection to VSM?

VSM tells you which functions must exist for an org to be viable. Second-order cybernetics tells you those functions are performed by people who are inside the system they observe. The two go together — VSM is the structure, second-order is the epistemology.

Takeaways

There is no HR measurement that stands outside the culture it measures. Every instrument is also an intervention.
Whatever HR chooses to count becomes the definition employees optimise against.
Prefer behavioural signals (attrition, referrals, mobility) over perceptual scales — they resist gaming better because gaming them requires actually improving.
Involve the observed in shaping the instrument; retire instruments on a schedule; report metrics with the observer effect stated openly.
Second-order cybernetics is not academic — it is the difference between HR that improves culture and HR that improves scores while culture decays.