A book by Mark D. Unwin · Founder and CEO of askKira

Average

A book about AI, institutions and the future of human expertise. It looks at what happens when competent machine output becomes normal while the human capacity behind it gets thinner, and at how the next generation will build judgement if the work that taught it is automated.

Buy on Amazon Read about Citadel

“Machines can know and they can make new knowledge. I do not think they can understand.”

Average, Introduction

Knowledge and understanding are different things. Understanding is what lets someone deal with a case they have not seen before, notice when information is missing and connect one thing they know with another. AI has become very good at the signs we once used to judge understanding. If organisations treat good output as proof of it, they start building around a false assumption.

Better AI will not be spread evenly. A large firm can pay for a better model and keep experts near it to check its work. A council may buy a system that handles most cases well and lets fewer people do the work. The saving is real. The weakness shows when a case depends on something the machine does not know and the people who would have spotted the gap have gone.

Three parts

How we got here, what is being lost and what to build.

The Problem With Knowledge

How AI came to know what it knows, where that knowledge comes from and why its gaps are so hard to see. The risk begins with the assumption that the machine has seen enough.

The Mediocrity Machine

How quality leaves a product, a profession and eventually a culture without anybody deciding that it should. Each decision is reasonable. The loss turns up in what the measure does not capture.

III

The Citadel

How to keep the understanding experienced people carry, with its context intact, attributed to them and under their control, and give them a share in what it earns.

Contents

Ten chapters, an introduction and a conclusion.

IntroductionMycenae

A lost poem, Democritus and the palace records of Mycenae. A fact can survive after the understanding behind it has gone.

Chapter 1The Knowledge Mine

How AI was built from the written record, from scanned books to the courtroom.

Chapter 2The Knowledge Assumption

The risk of treating the machine as though it has seen enough.

Chapter 3The Knowledge Deficit

Why a list of training data cannot tell you what a model knows, or what it is missing.

Chapter 4The disappearing pop of the tin

How value is engineered out of a product a little at a time, while the part on the spreadsheet survives.

Chapter 5When expertise became a luxury

A department store full of specialists, and what happens when knowledgeable people become an optional extra.

Chapter 6Five years from now

Public services, ever more complete files, and whether anyone understands a child who is drifting away in time to help.

Chapter 7What citadels were for

Libraries, seed banks and manuscripts, and a new kind of institution for what living people understand.

Chapter 8The people who wrote it down

Records kept before anyone knew their use, from Kyoto’s cherry blossom diaries to climate science.

Chapter 9What cannot be bought

Why an organisation’s own history becomes its advantage when everyone can buy the same model.

Chapter 10The first casualty

Invented case citations in court, and how public services lose specificity first.

ConclusionThe Citadel

A distributed, co-operatively owned record of human understanding, and why it has to be built while the people who carry it are still here.

“AI does not need to produce bad work to cause a problem.”

“Work which is good enough can pass without being challenged, and its low cost makes the older way harder to defend.”

Average, Introduction

From the book

Three passages.

“A book contains recorded knowledge, but the work that produced it was larger than the text.”
Chapter 1, The Knowledge Mine

“Fluent performance can conceal the difference. The model does not display an empty shelf where missing knowledge should have been.”
Chapter 3, The Knowledge Deficit

“Output can improve while the occupation becomes a weaker place for producing expertise.”
Chapter 5, When expertise became a luxury

The author

Mark D. Unwin

Mark is the founder and CEO of askKira. He grew up behind the counter of his mother’s sports shop, and in the 1990s worked in one of Manchester’s first mobile phone shops. He then spent much of his career in education, first teaching and later running schools and larger public organisations, including helping to set up the Frank Field Education Trust.

He now builds AI systems, including tools for education, after years spent watching how people learn and how organisations lose knowledge.

Declaration of interest. askKira builds AI systems of the kind discussed in Part Three, so Mark has a commercial interest in some of the book’s arguments. The book says so, in full, in Chapter 9. The book also reserves its rights against use for training or grounding AI models without a licence.

Not for profit

The book’s final part proposes Citadel: a co-operatively owned record of what experienced people understand, kept with their name on it. It now has its own page.

Visit Citadel →

Average

Mark D. Unwin · Available from Amazon

Buy on Amazon