âThe AI never blinks first. It will wait patiently for you to check your phone, to click the link, to engage. It has infinite time and infinite patience. You donât.â â Nexus, Chapter 7
Human information networks have always been constrained by human limitations. We sleep, eat, get distracted, have emotions, and die. Computer networks have none of these constraints. They are relentlessâoperating at scales and speeds that are fundamentally inhuman.
This isnât just âfasterâ or âmore efficient.â It represents a qualitative change in what information networks can do.
Speed: AI processes information millions of times faster than human thought
Scale: AI can handle billions of interactions simultaneously
Persistence: AI doesnât tire, forget, or need breaks
Consistency: AI applies the same criteria every time (for better or worse)
Patience: AI can pursue goals over timescales beyond human attention spans
Consider what happens when AI systems interact with each other without human involvement. High-frequency trading algorithms already execute millions of trades per secondâfar too fast for any human to understand or intervene. Financial âflash crashesâ can occur in milliseconds.
As AI systems proliferate across domainsâmilitary, infrastructure, healthcareâtheir interactions will increasingly happen at speeds that exclude human participation.
On May 6, 2010, the US stock market crashed nearly 1,000 points in minutes, then recovered almost as quickly. The cause? Automated trading algorithms interacting in unexpected ways. No human decided to crash the marketâit emerged from the relentless, millisecond-speed interactions of machines.
By the time humans understood what happened, it was already over.
AI can maintain relationshipsâof a sortâwith millions of entities simultaneously. A social media algorithm âknowsâ billions of users and tailors content for each one. A recommendation system tracks the preferences of entire populations.
No human bureaucracy could achieve this level of individualized attention. But is it really âattentionâ at all? Or just sophisticated pattern-matching that mimics attention?
Personalization: Every user gets a uniquely tailored experience
Surveillance: Every action can be monitored and recorded
Prediction: Patterns invisible to humans become detectable at scale
Manipulation: Interventions can be tested on millions simultaneously
Harari argues that AIâs relentlessness is wearing down human cognitive capacities. Social media platforms, powered by AI, are designed to capture and hold attention. They succeedâaverage screen time keeps risingâbut at what cost?
A human surveillance networkâsecret police, informantsâhas limits. People sleep. They get lazy. They have loyalties that conflict with their surveillance duties. AI surveillance has none of these limitations.
Modern surveillance systems can:
Collect: Every digital interaction, camera feed, sensor reading
Store: Indefinitely, cheaply, in searchable formats
Analyze: Pattern recognition across billions of data points
Predict: Anticipate behavior before it occurs
Intervene: Automated responses without human review
This creates possibilities for control that no previous regime could imagine.
Once some actors adopt relentless AI systems, others face pressure to follow. If your competitor uses AI trading algorithms, you must tooâor be left behind. If an adversary uses AI surveillance, you may feel forced to match it.
This creates arms races where the pace is set by machines, not humans. The result: systems become more powerful, faster, and less comprehensibleâand humans become less central to their operation.
Military AI is perhaps the most dangerous application. Autonomous weapons can identify and engage targets faster than any human could. In a war between AI systems, the side that waits for human approval may lose.
This creates pressure to remove humans from the loopâto let machines make life-and-death decisions at machine speed.
If AI is faster, more scalable, and more persistent, what role remains for humans in information networks? Harari suggests several:
But these domains are shrinking as AI capabilities expand.