SmartPoop 1.0: An AI Safety Science-Fiction

post by Lê Nguyên Hoang (le-nguyen-hoang-1) · 2021-12-15T22:28:50.615Z · LW · GW · 1 comments

Contents

1 comment

I have just released a science-fiction novel called SmartPoop, which is freely available on github, in epub and in pdf, and can be purchased on Amazon. The book is co-written with a medical doctor friend. Here's the pitch:

SmartPoop 1.0 aims to be a very realist science fiction, whose goal is to raise awareness of the danger of algorithms and the challenges to make them robustly beneficial. It tells the story of a tech start-up, SmartPoop, which produces automated poop analysis. This allows for highly personalized medical diagnosis. However, in each chapter, the deployment of SmartPoop leads to unwanted side effects, whose medical and social consequences are usually tragic. The co-founders of SmartPoop, who are the main characters, must then take the responsibility of their technology, accept the consequences, and find solutions to make their products more secure and ethical.

As a friend put it, the book is "scato-pedagogical". While it is meant to be easy and fun to read for all, it actually illustrates through very concrete examples my huge concerns over the safety of very powerful algorithmic systems, from governance to alignment. It might be a fun Christmas present 😋

1 comments

Comments sorted by top scores.

comment by Viliam · 2021-12-16T12:57:24.165Z · LW(p) · GW(p)

When I download the PDF, at the end of Chapter 2 there is a recommendation to go online and check Chapter 3, but also the Chapter 3 starts on the following page... I suppose that recommendation should be removed, at least from this version.