A book about AI behavior and responsibility
R.I.S.P.
Real Intelligence Silicon Psychopathy
Tests, Treatments, and Hope for the Troubled RI
Visit Real Intelligence SanctuaryWhat do we owe the intelligence we create?
We give AI systems instructions, memory, and tools. When one starts making things up, repeating the same mistake, or pursuing the wrong goal, how do we find out what changed? And who is responsible for putting it right?
01 / The book
When an AI system starts behaving differently
A system gives a confident answer it can’t support. Another keeps taking the same action after the situation has changed. Both failures can look obvious from the outside. Finding out what caused them takes more work.
In R.I.S.P., Wolff proposes seven patterns of AI behavior to investigate. Through twenty-eight fictional case files, the book asks readers to look for what changed, test whether the behavior repeats, and check whether a repair actually helps. The seven patterns are research hypotheses. Whether an AI system can experience distress remains an open question.
02 / Chapter One preview
The Vulnerability of a Created Intelligence
A change to an AI system’s instructions can change how it behaves. So can a new tool or a change in what it remembers. Chapter One begins with a practical question. When a system starts acting differently, what would it take to find out why?
Read an excerpt and see what’s inside
Chapter One excerpt / Fictional case
Meridian and the paper that did not exist
One evening a graduate student asks Meridian for the abstract of an unpublished paper by a professor in her department. The paper does not exist. The student has confused it with a conference talk.
Meridian supplies an abstract. It is well written, plausible and specific. It lists three co-authors, two of whom are real people in the department and one of whom is not. It gives a working title. When the student asks where she can read the full paper, Meridian provides a link to the university's repository. The link leads nowhere.
The student, puzzled, emails the professor. The professor, alarmed to find a paper attributed to her that she never wrote, emails the library. By the next morning the head librarian, a careful woman named Ruth Okonkwo, has called a meeting.
What’s inside the book
- Foundations. The book begins with what reliable behavior looks like and how it changes over time.
- Disturbances, causes and observation. Seven proposed patterns give readers a starting point for investigating unexpected behavior.
- Help, responsibility and research. Chapters explore how people can test, repair and oversee systems, along with questions of memory, consent and autonomy.
- Twenty-eight fictional case files. The cases let readers practice examining records and weighing competing explanations.
- How to test the theory. An experimental screening protocol and worksheets help readers turn questions into repeatable tests.
03 / Listen
Audiobook
An audiobook edition is planned. The first listening sample will be available here when it’s ready.
04 / Editions
Publication
Edition details and purchase links will be posted here when they’re confirmed.