← Back to Autonomy

A Plain-Language Field Guide · Part 6 of 7

← Part V · The Health of a Thinking System

Logic and Reasoning

Part VI · The Dynamics of Thinking

Every part so far treated thinking as a structure. This one treats it as a motion: a process that unfolds in time, settling into conclusions, tipping into new ones, and sometimes looping. The lens is dynamical systems, and it applies at every scale the series has built.

Section 44

Thinking is a motion, not a snapshot

Every part so far has treated thinking as a structure: kinds of inference, the parts of an agent, a society, the states of knowledge. But thinking is also a motion. It unfolds in time, and the unfolding has a shape.

Borrow the language you would use for any moving system. A line of thought is a trajectory through a space of mental states. A conclusion is where that trajectory settles, a resting point the dynamics fall into, what a physicist calls an attractor or a fixed point. Conviction is how stable that resting point is: nudge a firm belief and it returns; nudge a shaky one and it slides somewhere else. Indecision is a trajectory that never settles, and obsession is one that settles into a place it cannot leave.

This is not a metaphor borrowed for color. It is the same lens that describes a feedback loop, a control system, or a cooling metal, and it applies to thinking at every scale this series has built: one chain of reasoning, one agent's loop, and a whole society of them. Naming the motion lets us see why some thinking converges, some oscillates, and some runs away, and what decides which.

attractor: a conclusion never settles space of mental states
Figure 39. A thought as a trajectory. One path spirals into a stable attractor, a conclusion held; another wanders without ever settling, the motion of indecision.

The rest of this part follows the motion: first the rhythm that drives it, then how it settles, then how it tips, then how it fails, and finally the same dynamics seen at every scale at once.

Section 45

The rhythm: generate and select

Thought does not move in a straight line. It breathes, alternating between two opposite motions, and good thinking is largely a matter of timing that alternation well.

The two motions are divergence and convergence. Divergent thinking widens the space, generating options and associations without judging them. Convergent thinking narrows it, pruning and selecting toward one answer. This is the cognitive face of the explore-and-exploit tension from earlier, now a rhythm rather than a single choice. Run them at the wrong time and you get the two classic failures: judging while you generate kills ideas before they form, and generating while you should be deciding never lets you finish.

Interactive · the two motions

Zoom out and a hard problem moves through a known sequence, named long ago by Graham Wallas: preparation, loading the problem; incubation, stepping away and letting it run in the background; illumination, the answer surfacing; and verification, checking it. Incubation is the surprising one and the most dynamical: stepping away lets the trajectory leave a stuck region it could not escape under direct, convergent pressure.

PREPARE INCUBATE ILLUMINATE VERIFY divergent convergent
Figure 40. The phases of a hard piece of thinking. Incubation is the dynamical heart of it: only by releasing convergent pressure can the trajectory leave a basin it was stuck in.
Practical note Separate generation from selection in time, not just in principle, and respect incubation. A process held under constant convergent pressure often cannot reach the state where the answer actually lives.
Section 46

Settling: attractors, stability, and conviction

The moment a thought clicks is the moment a moving trajectory falls into a resting point and stays. What kind of resting point it is decides how much you should trust it.

Picture a landscape of valleys. The deeper the valley, the more strongly the thought is pulled into it and the harder it is to dislodge. A conclusion reached and held is a ball that has rolled into a valley, a stable attractor; a deep wide valley is a firm conviction, a shallow one a tentative belief a small push will tip. Recall works the same way: a partial cue starts the ball rolling and the dynamics complete the pattern, which is why a fragment of a memory drags the whole thing back. Roll the ball, then perturb it.

Interactive · settling into a basin
Figure 41. Conviction as a basin. The ball settles into a valley and stays unless a large enough perturbation knocks it out; a deeper valley is a firmer belief.

This picture also explains two failures you met earlier. Settling too easily into the first available valley is fixation, the premature convergence from problem-solving. Settling into a valley you cannot climb out of, even when it is wrong, is the stubborn belief that survives contrary evidence. Stability is a virtue for a correct conclusion and a trap for an incorrect one, so the depth of your conviction should match the strength of your evidence, not exceed it.

Why it matters Distinguish a belief that is stable because it is well supported from one that is merely stuck. The test is whether new evidence can still move it. A mind whose attractors are too deep has stopped learning.
Section 47

Tipping and insight: bifurcations

Sometimes the landscape itself changes, and a valley that held a conclusion flattens and disappears. The thought does not drift then, it jumps, and the jump is what we feel as a sudden change of mind or a flash of insight.

As evidence accumulates it tilts the landscape. For a while the old conclusion holds, its valley merely shallower. Then at a threshold the valley vanishes and the ball rolls abruptly into a new one. A physicist calls that a bifurcation; we call it a tipping point, a change of mind, or, when the new valley is a better way to see the whole problem, insight. The aha moment is not slow accumulation but fast restructuring, the trajectory snapping from one organization of the problem to another. Slide the evidence and watch the ball tip.

Interactive · the tipping point
evidence for B
Figure 42. The tipping point. As evidence tilts the landscape the old valley shallows and then vanishes, and the ball jumps rather than drifting across, a bifurcation. Hysteresis means the return trip needs more evidence than the crossing did.

Two features of tipping are worth knowing. It is abrupt, so a long stretch of no apparent change can end in a sudden flip, which makes a system on the edge of tipping easy to mistake for a stable one. And it shows hysteresis: the evidence needed to tip you out of a belief is usually more than the evidence that would have kept you out of it in the first place, which is one honest source of confirmation bias rather than mere stubbornness. A system tuned too rigidly never tips when it should; tuned too loosely it tips at every passing input. Healthy cognition sits between, near what dynamics calls the edge between order and chaos.

Why it matters Watch for a system sitting near a tipping point, where one small fact triggers a large revision. And budget more evidence to undo a belief than to form one, because the landscape resists leaving a valley more than it resisted entering it.
Section 48

When the dynamics go wrong

Most failures of thinking are not wrong facts but bad motion. The trajectory does one of three things it should not: it loops, it runs away, or it refuses to move.

The cleanest way to see all three is the one you would reach for with any feedback loop, by turning a single knob: how strongly each thought drives the next. Turn it up.

Interactive · loop gain
loop gain
Figure 43. One knob, three behaviours. Low gain converges to an answer, higher gain oscillates without settling, and too much gain diverges into a runaway loop.

At low gain the trajectory converges, damping out to a steady answer, which is healthy deliberation reaching a conclusion. Turn it up and it oscillates, swinging between two answers, which is healthy debate at first and a flip-flopping rumination if it never stops. Higher still and the loop runs away, each step amplifying the last into a diverging spiral, which is the agent doom loop, the escalating argument, the confirmation cascade growing more certain with every pass over the same evidence. The fourth failure is the opposite of all three: a trajectory so over-damped it freezes and never leaves the start, which is paralysis.

The cures are the cures of control. Add damping, which in a reasoning system is verification and the gate, so strong proposals are checked rather than amplified. Add a stopping rule, the metareasoning brake, so deliberation ends before it becomes rumination. And break the feedback path with an outside check, so a runaway loop cannot keep feeding on itself, which is exactly why the verifier and the human gate sit outside the loop they watch.

Practical note When an agent or a discussion will not settle, do not add more reasoning, add damping: an independent check, a stopping rule, or a break in the loop. More drive into an unstable loop makes it worse, not better.
Section 49

Dynamics at every scale

The same three shapes, settling, tipping, and looping, appear whether you watch one chain of thought, one agent, or a thousand of them. That is the payoff of the dynamical lens: one vocabulary for all of it.

At the smallest scale, a single chain of reasoning is a trajectory that should converge on an answer and sometimes spirals into nonsense instead. One level up, an agent's control loop is a feedback system that converges to a solution, loops when it gets stuck repeating itself, or diverges when its context fills with its own amplified output. And a society of agents is a coupled dynamical system, where consensus is convergence to a shared attractor and polarization is divergence into separate ones. Watch a population settle or split.

Interactive · opinion dynamics
ONE CHAIN converge to an answer ONE AGENT LOOP loop, or settle A SOCIETY consensus, or split
Figure 44. One vocabulary, three scales. Convergence, looping, and divergence describe a single chain of thought, one agent's loop, and a whole society alike.

In every case the same handful of controls shape the motion. Exploration, the temperature of the system, is the drive that lets it leave a basin. Verification and gating are the damping that lets it settle. A stopping rule is the brake. And an outside check is the break in the loop that prevents runaway. Tune them well and a thinking system explores enough to find good answers, settles firmly enough to commit, tips readily enough to change its mind on real evidence, and never spins forever.

Looking ahead Trace the arc once more. A single act of reasoning in Part I, its limits and its mechanization in Part II, one problem-solver in Part III, a society of them in Part IV, the health of their knowledge in Part V, and here the motion of the thinking itself. Underneath every part is one shape, a fast, fallible proposal disciplined by slower verification with something wiser holding the gate, and read dynamically that shape is simply a well-damped feedback loop: drive to explore, damping to settle, a gate for stability. Whether the thinker is a person, an agent, or a society, good thinking is good control. Let it move boldly, keep it damped enough to settle, and never mistake a deep rut for a firm conclusion. Part VII turns to what all of this motion is in service of: understanding, the model a mind builds so the world becomes predictable from the inside.
Practical note When you design or debug a thinking system, ask the questions you would ask of any dynamical system. Does it converge, and to what? Is it stable to a perturbation? Where are its tipping points? And is there enough damping in the loop to keep it from running away? Those four questions catch most of what goes wrong.
Appendix

References and further reading

Sources behind Part VI, on cognition as a dynamical system, attractors and bifurcations, insight, and the dynamics of collective opinion.

  1. van Gelder, T. (1998). The dynamical hypothesis in cognitive science. Behavioral and Brain Sciences, 21(5).
  2. Beer, R. D. (2000). Dynamical approaches to cognitive science. Trends in Cognitive Sciences, 4(3).
  3. Thelen, E. & Smith, L. B. (1994). A Dynamic Systems Approach to the Development of Cognition and Action. MIT Press.
  4. Spivey, M. (2007). The Continuity of Mind. Oxford University Press.
  5. Hopfield, J. J. (1982). Neural networks and physical systems with emergent collective computational abilities. PNAS, 79(8). Attractor dynamics.
  6. Strogatz, S. H. (1994). Nonlinear Dynamics and Chaos. Addison-Wesley. Attractors, bifurcations, and stability.
  7. Wallas, G. (1926). The Art of Thought. The four stages of creative problem-solving.
  8. Ohlsson, S. (1992). Information-processing explanations of insight and related phenomena. In Advances in the Psychology of Thinking.
  9. Hegselmann, R. & Krause, U. (2002). Opinion dynamics and bounded confidence. JASSS, 5(3).
  10. DeGroot, M. H. (1974). Reaching a consensus. Journal of the American Statistical Association, 69(345).
Continue · Part 7 of 7Understanding and Comprehension →What all the machinery is for: how a mind builds the model behind a conclusion, why it runs on prior knowledge, and whether a machine understands.← Back to Part 5The Health of a Thinking SystemPart V of the series.