Three layers of the same story: the machines, the models, and what they do once they are pointed at people. The last two entries are the ethics reading, and they are here rather than on a shelf of their own because the consequences are not a separate subject from the technology.
Chip War
Chris MillerEssentialBookStart here
What to notice. One company in the Netherlands is a single point of failure for everything.
How semiconductors became the chokepoint of the world economy. The most useful single book for anyone covering technology, and increasingly for anyone covering anything.
Where it shows up in a roomWhere it shows up in a room
Export controls and supply concentration now appear in technology, macro and credit interviews alike, and this is the one book that covers all three.
Genius Makers
Cade MetzRecommendedBookStart here
What to notice. Deep learning won a decades-long argument, then was bought in a weekend.
The history of how neural networks went from an academic backwater to the centre of the industry, told through the people. The best available context for why the field looks the way it does.
Where it shows up in a roomWhere it shows up in a room
Knowing why the field ignored this approach for thirty years is a better answer to why now than anything about compute alone.
The Coming Wave
Mustafa Suleyman and Michael BhaskarRecommendedBookIntermediate
What to notice. Containment is the problem, and it has no precedent that worked.
Artificial intelligence and synthetic biology framed as a containment problem rather than a product one. The most frequently cited book in technology interviews this cycle.
Where it shows up in a roomWhere it shows up in a room
It is the reference everyone reaches for, so having a specific disagreement with it is worth more than having read it.
The Alignment Problem
Brian ChristianEssentialBookIntermediate
What to notice. A system optimises the objective you wrote, not the one you meant.
The clearest non-technical account of why specifying what we want is the hard part. It is a research history as much as an ethics book, which is what makes it credible.
Where it shows up in a roomWhere it shows up in a room
Asked what could go wrong with a model you have proposed, reward misspecification is a concrete failure mode rather than a worry.
Weapons of Math Destruction
Cathy O’NeilRecommendedBookStart here
What to notice. A model trained on a biased outcome will reproduce it at scale.
What happens when opaque scoring systems are pointed at sentencing, hiring and credit. Written by a quant who left finance for exactly this reason, which is why it lands harder than most of the genre.
Where it shows up in a roomWhere it shows up in a room
Any model that touches people invites a fairness question, and naming proxy variables is a specific answer where most are vague.