Capstones

This is Unit Eleven of Artificial Intelligence: the capstones. It is the last one. Nothing new arrives. Every entry reassembles the course, and the assembly is organised around one idea stated in the second-to-last topic: what this course hands forward is not a snapshot of the field but a kit for auditing claims about it. That distinction is the point of the whole thing. A snapshot of capabilities goes stale within a year. Three clauses inside a capability claim, ten questions, reliabilities multiplied along a chain, and forecasts held to named mechanisms — those apply to claims that have not been made yet. The unit also settles accounts honestly. What the course skipped was skipped on purpose, and each omission has a name and somewhere to go for it. And there is an itemised bill of the mathematics standing between here and the field's literature. One entry addresses the belief that the real field belongs to other people. It takes the two reasons people give for it and shows both fail on the course's own evidence. By the end you will be able to do five things. One: place a real system on the design space by hand. Two: name the region the field cannot yet build for. Three: say what mathematics stands between here and the literature. Four: say what the course hands forward. Five: run the full kit against a published claim.

Three entries returning to the design space from Unit One and completing it. Every method family in the course occupies a home region — and one corner of the space is empty, which is where the field's genuine frontier sits.

The map, completed.

AI 0621 The Held Map

Every method family in the course has a region of the design space where it works. Unit One's table of environment properties, filled in by hand across ten units of material.

The empty corner.

AI 0622 The Corner Cell

The field still cannot build an agent acting well in indefinitely long, partially visible worlds shared with other autonomous agents. That is the cell where every dimension is difficult simultaneously.

The map, used.

AI 0623 Three Systems, Placed

Three everyday systems placed on the space by hand, one dimension at a time. This is the course's central diagram doing the job it was built for.

Two entries of honest accounting. Every omission was deliberate and has somewhere to go — and then an itemised statement of the mathematics between this course and the field's actual literature.

The omissions.

AI 0624 The Omissions Ledger

Everything skipped was skipped on purpose, and each omission carries a name, a reason, and a forwarding address in freely readable material. Nothing was left out silently.

The bill.

AI 0625 The Mathematics Invoice

Three bodies of mathematics stand between here and the full literature — calculus completing descent, linear algebra completing rows, and probability theory completing counting. Each is named rather than gestured at.

Two entries addressing why people stop. One takes the two reasons commonly given for believing the field belongs to somebody else and shows both fail on evidence this course produced. The other removes the practical obstacle.

The two proofs.

AI 0626 "The Real Field Is for Other People"

The belief rests on two arguments — not being a mathematics person, and having started too late. Both are invalid against what this course has already had its viewers do successfully.

And the obstacle.

AI 0627 The Reading Path Is Free

The path forward costs nothing — this course's verification chain doubles as its reading list, and the anchor text's exercises assume nothing the course has not built or explicitly invoiced.

Three entries on what survives the course. The reading habits, the field's own open problems now checkable against mechanisms you own, and the reason a kit outlives a snapshot.

The habits.

AI 0628 The Habits Transfer

No formula passes unread, every method is priced as a purchase with a guarantee and a cost, and every performance claim gets chased to its test set and its custody. Those transfer whole.

The open problems.

AI 0629 The Standing Open Problems

The field's named open problems — learning from scarce data, causality, embodied common sense, eliciting preferences, synthetic media — can each be checked against a mechanism this course explained.

And why it lasts.

AI 0630 The Kit Outlives the Snapshot

What the course hands over is an audit kit rather than a picture of the field — three clauses, ten questions, chains multiplied, forecasts held to named mechanisms. It applies to claims nobody has made yet.

Two entries returning to Unit One. The founding wager, seventy years on — and the word the whole field is named after, which still cannot carry what people load onto it.

The bet.

AI 0631 The Bet, Revisited

The wager that every feature of intelligence can be described precisely enough for a machine to simulate stands unsettled after seventy years — and magnificently productive throughout, which is a rare combination.

And the word.

AI 0632 The Word Carries Too Much

The word remains too ambiguous across humans and machines to bear the predictions built on it. The behavioural standard from Unit One was a working tool rather than a settled philosophy, and the course says so at both ends.

Two entries closing the course. A procedure that is really an assignment — take a claim published this month and run everything against it — and then a description of what a finished viewer is.

The capstone. One claim, this month.

AI 0633 The Final Audit

Take any capability claim published recently and run the whole kit against it, in writing, ending in a verdict stated in three clauses. The course closes by asking for the work rather than describing it.

And what you can do.

AI 0634 The Participant

Follow a claim to its test set, a mechanism to its arithmetic, a harm to the pile and objective that produced it, and a forecast to the assumption holding it up. That is what the course was for.

Five things you can now do, and the course is complete. You can place a real system on the design space one dimension at a time, using the diagram from the first unit as an instrument rather than an illustration. You can name the corner the field cannot yet fill — long-running, partially visible worlds shared with other autonomous agents. You can say what mathematics stands between here and the literature: calculus for descent, linear algebra for rows, probability theory for counting. You can say what the course hands forward — an audit kit rather than a snapshot, which is why it applies to claims nobody has made yet. And you can run the whole kit against a published claim, chasing it to its test set, its mechanism, and the assumption holding it up.