The Complex Perspective · 2036
The Decade Test
The 2016 first edition graded against 2026 — mechanism claims held, capability claims broke.
Why should you believe a site about 2036? You shouldn’t — not on trust. But this site has something most futurism lacks: a prior run of the experiment. In 2016, the first edition of Die komplexe Perspektive made claims about technology, prediction, and AI. One decade and one AI revolution later, those claims can be graded — including, prominently, the ones the author got wrong. The misses are graded in detail in the 2025 retrospective “AI Took a Different Path Than I Expected”; this page adds the half that post left implicit: what held.
The verdicts sort themselves: claims argued from mechanism held; claims argued from capability mutated or broke. That sorting is the entire epistemic basis of this site.
The grading
„Verlässliche Vorhersagen über einen langen Zeitraum sind in komplexen Systemen sehr unwahrscheinlich.“ “Reliable long-range predictions are highly unlikely in complex systems.”
Survived untouched — and 2016–2026 was the harshest possible test decade. A claim about the structure of complex systems: agents adapt, so the future is not an extrapolation of the present. It even predicted the failure of the chapter’s own capability forecasts, below.
2016 edition · ch. 10 · Die Zukunft
„Man bekommt kein korrektes Bild von der Zukunft, wenn man die Gegenwart einfach linear weiterführt und mögliche Änderungen nicht berücksichtigt.“ “You do not get a correct picture of the future by simply extending the present linearly and ignoring possible changes.”
The extrapolation-error mechanism, illustrated in 2016 with Harry Harrison’s 1966 vision of a starving 35-million-person New York of 1999. It held in both directions: linear pessimism about resources kept failing, and — with some irony — linear confidence about which AI capabilities were far away failed too.
2016 edition · ch. 10 · Die Zukunft
„Aber es wird keinen ‚Wundercomputer‘ geben, der ganz plötzlich intelligenter als ein Mensch ist und die Weltherrschaft übernehmen will.“ “But there will be no ‘miracle computer’ that is suddenly more intelligent than a human and wants to take over the world.”
No takeover, no discontinuous miracle — that part held through 2026. But the claim’s framing assumed capability would stay narrow and legible, and instead a single architecture became unexpectedly broad. Half mechanism (no sudden discontinuity), half capability bet (breadth is far away) — and each half aged true to type.
2016 edition · ch. 10 · Die Zukunft · graded ⚠ contested in the 2025 retrospective
„Man kann davon ausgehen, dass es 2020 selbstfahrende Autos gibt, die allerdings noch nicht in allen Ländern erlaubt sein werden.“ “One can assume that self-driving cars will exist by 2020, though not yet permitted in every country.”
Directionally right — driverless taxis carried public passengers by 2020, in one or two cities, and regulation did lag. But the schedule was optimistic, the rollout narrower than the sentence implies, and the decade’s actual AI story arrived from a direction this list never mentions. A capability claim with a date attached, and it aged like one.
2016 edition · ch. 10 · Die Zukunft
„Computer können nur Symbole verarbeiten, deren Bedeutung sie nicht wirklich verstehen. Sie verarbeiten nur die Syntax und nicht die Semantik.“ “Computers can only process symbols whose meaning they do not truly understand. They process only syntax, not semantics.”
The book’s central capability argument — and the most instructive miss. Large language models extracted usable semantic structure from text statistically, with no hand-built meaning layer. Whatever one holds about “true” understanding philosophically, the economic content of this claim — machines cannot act on meaning — dissolved. It felt like a mechanism claim; it was a capability claim wearing a philosopher’s coat.
2016 edition · ch. 10 · Die Zukunft · graded ✗ refuted in the 2025 retrospective
„Aber dieses Wissen zu formulieren, diese ‚Ontologien‘ zu erstellen, ist mit sehr viel Arbeit verbunden […] Daher wird es vorerst keine Computer mit ‚Common Sense‘ geben.“ “But formulating this knowledge — building these ‘ontologies’ — takes enormous effort […] So for now there will be no computers with ‘common sense.’”
Right premise, wrong conclusion: hand-built ontologies were indeed a dead end — RDF and OWL never scaled to common sense. The miss was assuming they were the only path. The bottleneck wasn’t real; it was an artifact of the then-current method. Broken bottleneck arguments are a recurring species of failed forecast, worth learning to recognize.
2016 edition · ch. 10 · Die Zukunft
What this buys you
If mechanism claims survive decades that shred capability claims, then a site about 2036 should be built exclusively from mechanism claims — and that is the baseline’s discipline and every module’s structure. It also buys calibrated humility: the author’s own best philosophical argument broke inside ten years. So when a module here says mechanism, it means the boring kind — arithmetic about relationships, cost asymmetries, incentive structures — not philosophy in a mechanism costume. And when this site wants to dream anyway, it uses the Cinema register, where being wrong is the point.