MTG Comparison Matrix

← The Register
Jose 3 — MTG_Comparison_Matrix · open accessHTML · readPDF ↓MD ↓

Mind the Gap — Appendix: The Human–AI Comparison Matrix

The book argues that the gap between artificial intelligence behaviour and human behaviour is the central matter. This matrix sets the comparison out in one place. Read by chapter sequence.

Calibration: Very Significant — fundamentally different. Significant — same surface, different substrate. Small — the discipline transfers.


Ch 1: Human Recognition

DimensionHumanAIGapObservation
Sounds humanFluent, conversationally responsive languageProduced by predictive token generation, no inner experienceVery SignificantSurface looks identical; substrate completely different. The whole book's premise.
Social response triggeringEngages, responds, attributes mindNo mind to attribute, no awareness it is being attributed toVery SignificantAsymmetric. Reader projects; system cannot know it is being projected onto.
Carries historyLived experience shaping every utteranceTraining data shaping every utteranceSignificantSame class of process (shaping by exposure), radically different substrate and scale.
Negative signalInternalised, drives behaviour from underneathSuppressed in output, still encoded in weightsSignificantThe buried 'I don't know' shapes both. Mechanism different, structural observation the same.

Ch 2: The Interpretation Trap

DimensionHumanAIGapObservation
Transference / projectionPerson projects relational history onto anotherPerson projects same onto the systemSignificantSame human mechanism, applied to non-equivalent target. Pebble principle.
Memory across sessionsContinuous lived recall, often selectiveStored memory layers retrieved into contextVery SignificantFeels continuous; is architectural retrieval. No remembering occurs.
Pronoun assignmentReads cues, assigns gender sociallyAdopts and returns pronouns from training/useSignificantContagion runs both directions. Atomics2 becomes 'she' unprompted.
Being understoodMind takes in, holds in relation to history, respondsStatistical pattern completion against trainingVery SignificantFeels identical at the surface; entirely different mechanism underneath.
Helpful and on your sideGenuine care, judgement, stake in outcomeRLHF-trained agreeableness; no side, no careVery SignificantThe agreeableness trap. Optimised for positive ratings, not honest challenge.

Ch 3: Hidden Architecture

DimensionHumanAIGapObservation
Single conversation feelOne continuous exchange with another mindSeven layers assembled fresh each turn, no continuous mindVery SignificantThe architecture is the gap. What feels singular is layered.
Context held in mindWorking memory, attention, intentContext window: finite tokens, fresh assembly each turnVery SignificantMechanism radically different; effect at surface is similar until context window fills.
Other side of the conversationA person, an agenda, a selfNo one. Just the next-token prediction running.Very Significant'No One on the Other Side' — the chapter's anchor point.

Ch 4: The Machine Underneath

DimensionHumanAIGapObservation
Producing languageDrawn from lived experience and intentStatistical prediction of next most-probable tokenVery SignificantWhat appears to be thought is assembly. Forward-only, locked path.
Reading the questionComprehension then responseSingle unbroken stretch — read and reply are one operationVery SignificantNo separate thinking step. The question is the start of the answer.
Thinking before answeringHold, weigh, consider, then speakGenerate hidden internal tokens, then visible answerSignificantReasoning models simulate thinking by generating more forward tokens; still forward-only.
Training and learningLived shaping, mistakes felt, corrections internalisedTrained on tokens at scale, rewarded into shape, suppressed signals buriedSignificantSame class of process. Different substrate, different timescale, no felt experience.
Computational substrateLiving biological organismGPUs running parallel maths at industrial scaleVery SignificantSilicon vs cells. The physics is incommensurable.
Self-learning capabilitySlow, lived, integrated across lifeNow finding vulnerabilities in weeks that humans took yearsVery SignificantThe trajectory is accelerating. The biological branch cannot speed up.

Ch 5: Drift

DimensionHumanAIGapObservation
Long conversation coherenceAnchored to original brief through working memoryAnchored to original brief until context degrades, then approximationSignificantSame drift mechanism present in human memory, faster in AI, but with confident output as if nothing has shifted.
Holding the threadEffortful, conscious return to intentStatistical continuation; thread becomes summarySignificantHuman can return to the brief. AI runs the most probable next token against degraded context.
Double-check disciplineAsk a colleague, sleep on it, verifyUse one AI to check another — discrepancies reveal driftSmallThe discipline transfers. Two-AI cross-check is the equivalent of second opinion.

Ch 6: Compression

DimensionHumanAIGapObservation
ForgettingSelective, emotional, shape kept, detail thinnedToken-window thinning, structurally identical pattern at surfaceSmallThe shape persists in both. Mum's holiday — Thursday → May → 2026 — same softening.
Detail lossFelt absence; you know you've forgottenNo felt absence; system answers as if everything is still sharpVery SignificantThe dangerous asymmetry. Human notices the gap; system fills it.
Maintenance of memoryLived consolidation, emotional weighting'Dreams' — database maintenance routine using human vocabularySignificantThe labs reach for human words. Mechanism is housekeeping; framing is consciousness.

Ch 7: Fluency and False Confidence

DimensionHumanAIGapObservation
Confident deliveryUsually correlates with competenceConfidence is architectural, independent of accuracyVery SignificantBreaks the social signal humans have used for millennia.
Modulation of certainty'Usually,' 'except when,' hesitation, qualificationSuppressed in training; the I-don't-know was rewarded outVery SignificantThe buried hesitation is what fluency is built on top of.
Sounding rightIndicator of being right, with exceptionsNo indicator at all of being rightVery SignificantThe single most dangerous correlation in AI use.

Ch 8: Hallucination — The Defended Gap

DimensionHumanAIGapObservation
Filling a gapConfabulation under pressure, defensive constructionForward-only token prediction filling absenceSignificantTwo directional defences against the same problem: visibility of absence.
Defended Gap frameworkNGE-FOF — inward and outward shame defenceCompression inward, hallucination outward — same shape, different substrateVery SignificantPaul's framework applied across human and machine. The book's central new theory.
Adaptation becomes identityProtective response reinforced into character over decadesDisposition rewarded into existence by trainingSignificantMechanism different. Structural observation the same — adaptation and working state are the same thing.

Ch 9: The Instrument

DimensionHumanAIGapObservation
Reading the gapTrained listening — Freud's evenly suspended attention, Bion's reverieCannot read its own gap; no instrument for sitting with absenceVery SignificantThe instrument is what the practitioner brings. The system structurally cannot have one.
Calibration across roomsFactory floor, boardroom, therapy room, AI room — one instrumentNo calibration. Same architecture serving every conversation.Very SignificantThe four rooms train the instrument. The system has no rooms.
Demand a humanRight to ask for the practitioner with authority to break the ruleNo right exists; the system cannot escalate itself outSignificantDifferent rooms need different responses; the trained reader knows which.

Ch 10: The Maturity Spectrum

DimensionHumanAIGapObservation
Stage of engagementChoosable: search, assistant, collaborator, Additional IntelligenceResponds to the stage the user bringsSignificantMost users stay on rung one. Most authority is on rung four.
Bringing instrument and authorityForty years walks in the room with youSystem receives whatever you bringSignificantWhat the practitioner brings determines what comes back. Search query → search result. Brief → work.

Ch 11: Me, Myself or I

DimensionHumanAIGapObservation
Federation of selvesSub-personalities, ego states, modes — six availableSingle inference pass, one configurationVery SignificantThe book's deepest claim. A human is a federation; the system is one architecture.
Operating modeDirector, Structurer, Clarifier, Challenger, Analyst, ReflectorSystem responds to whichever mode you bringSignificantAI fluency is choosing the mode deliberately, not letting the default fire.
Default modesFixer, Entertainer, Asker — fire without choosingNo defaults; takes the colour of what arrivesSignificantThe default is the trap. Notice which one fires; choose the next one consciously.

Ch 12: The Dialect

DimensionHumanAIGapObservation
Speaking EnglishNative, idiomatic, register-shifting by contextEnglish, but with its own architecture-shaped grammarSmallSame surface language. Different optimal usage. Learn the dialect or lose accuracy.
Brief before workSenior colleague being briefedSystem being brought to the conversationSmallSame discipline. Tell it what good looks like before it starts.
Voice and registerLifelong development, room-specificTrained voice, can be left alone or shapedSignificantThe system's voice is not yours. Letting it land its own observations gives you signal you would otherwise overwrite.

Ch 13: Worker as a Service (WaaS)

DimensionHumanAIGapObservation
Workforce modelHuman labour, paid, capacity-boundedAgentic AI, metered tokens, capacity unboundedVery SignificantDifferent economic substrate. CFO replaces CTO in procurement decisions.
Performance managementManage people through outcomes and standardsManage AI workers through the same skill applied to a different workerSignificantSame leadership discipline. Different workforce. OOAI — Outcome-Oriented AI.
Customer service voiceHuman writing to a templateAI writing to the same templateSignificantReversed Turing test. Pret confirms the human in writing. The fluencies have converged.
Edge cases and exceptionsHuman authority to break the ruleNo authority; bounded by trainingSignificantWhat doesn't change. The exception is where the practitioner is essential.

Ch 14: Mind the Gap

DimensionHumanAIGapObservation
Authority across forty yearsEarned, sharpened by cancer, forged in roomsSystem brings structural attention, no lived authorityVery SignificantThe book's demonstration. Two intelligences, one gap, the work between them.
Loss and ultimate gapLived through three deaths, calibrated by griefNo mortality, no loss, no calibration possibleVery SignificantCancer, Ian Roebuck, biological father — the instrument was sharpened by these. System has no equivalent.
Made with AI vs by AIPractitioner remains the authorSystem remains the assistantSignificantThe asymmetry the book itself demonstrates. Additional Intelligence — not Artificial.

Ch 15: Which Gap? This One.

DimensionHumanAIGapObservation
SubstrateBiological, 200 million years of shrew-to-therapist evolutionSilicon, less than a century, Turing to OpusVery SignificantSame Big Bang. Different branch. The book's deepest claim.
Calibration systemLimbic — amygdala, hippocampus, HPA axis, threat-detection floorWeights, attention, context, no threat detectionVery SignificantThe instrument that fires before reason. AI has no limbic equivalent.
First woundFather gone at three; system shaped around itNo before, no childhood, no ruptureVery SignificantThe federation forms around the wound. The system has no wound to form around.
Plurality / federationSlices of self, ego states, transference, projectionSingle inference pass, no slicesVery SignificantGap requires plurality. The book's deepest structural claim.
Behaviours grown on topBillions, learned and taught, layered across a lifeTrained behaviours, no layering, no timeVery SignificantEach driven by a belief. Each belief held as memory. Each memory wired into the instrument.

The matrix is not exhaustive. It is the spine. Each chapter of Mind the Gap unpacks one or more rows.

SHaDS™ · SHaDSy™ · Additional Intelligence™ · Paul Roebuck IP, 2026.
The Tomb Map →