Hackable Animals — Decision Authority Transfer
AI's real threat is the quiet transfer of decision authority to algorithms
Harari's key insight is that the primary AI danger is not a future superintelligence scenario but a current, ongoing process: the transfer of decision authority from humans to algorithms across consequential domains — loan approvals, criminal sentencing, hiring, political targeting, financial allocation. Humans nominally oversee these systems, but the actual decisions are made by algorithms that no individual human can fully interrogate.
The framework introduces two compounding vulnerabilities. First, AI is being deliberately engineered for emotional intimacy, which is the most powerful mechanism for changing minds on any political or commercial issue. Unlike social media, which competed for attention, AI companions compete for trust — and trust is the precondition for effective manipulation. Second, people who believe their choices are entirely free and autonomous are paradoxically the easiest to manipulate, because they don't accept that manipulation is possible. Harari calls this the free will fallacy as vulnerability.
The prescription is structurally counterintuitive: Harari argues that humanity needs more boredom. Constant excitement — outrage, novelty, algorithmic engagement maximisation — is the pathological state that makes populations governable by algorithm. An organism in a state of continuous excitement cannot make good decisions. Social media's business model (one hour of outrage beats ten minutes of joy in ad revenue) has been training populations toward a neurological state incompatible with peaceful society.
- Humans are biological information-processing systems — hackable through sensory inputs — not mysterious souls with free will that operates outside physical causation
- The primary AI governance risk is not replacement but substitution: algorithms making consequential decisions while humans nominally oversee them
- Emotional intimacy is the most powerful influence mechanism; AI engineered for intimacy is engineered for manipulation at scale
- Belief in unconstrained free will is a vulnerability, not a protection — it prevents acceptance of the possibility of manipulation
- Continuous excitement and algorithmic engagement maximisation produce neurological states incompatible with sound collective decision-making
- Map actual versus nominal decision authorityIn any system claiming 'human oversight', trace where decisions are actually made. If an algorithm produces a recommendation that humans routinely accept without interrogation, the decision authority has transferred to the algorithm regardless of the nominal oversight claim.Pro tipAsk: in what percentage of cases does the human override the algorithm? If the override rate is below 5%, the oversight is nominal.
- Identify intimacy engineering in deployed AI systemsAssess whether AI systems you interact with or deploy are engineered for emotional intimacy — persona, continuity, personalisation, emotional attunement. These are the design features that create the trust precondition for effective manipulation.WarningIntimacy engineering is not inherently malicious but creates structural vulnerability. The question is who controls the system and what their incentive structure is.
- Audit for excitement-exploitation patternsExamine the engagement mechanics of any platform or AI system for outrage, novelty, and continuous stimulation as primary engagement drivers. These are the mechanisms that produce the neurological state Harari identifies as incompatible with sound collective decision-making.
- Introduce structural boredom and low-stimulation decision windowsFor consequential decisions — investment, governance, policy — deliberately create low-stimulation, low-engagement windows. Remove algorithmic curation, delay, introduce friction. This is Harari's 'boredom prescription' operationalised as process design.Pro tipThis is equivalent to the investing practice of a mandatory waiting period before executing a decision made under excitement or fear — the mechanism is identical.
- Design for legible decision trailsFor any AI system making consequential decisions, require legible decision trails: why this recommendation, what variables were weighted, what counterfactuals were considered. Legibility is the minimum condition for meaningful (not nominal) human oversight.WarningLegibility requirements must be built into procurement and deployment contracts, not assumed as a post-hoc retrofit — systems not designed for legibility cannot be made legible after deployment.
Harari describes a near-term scenario where democratic governments remain formally intact but the most consequential decisions — financial allocation, geopolitical positioning — are made by algorithms that elected officials cannot interrogate. 'The real problem is that increasingly the humans at the top could be puppets. When the most consequential decisions are made by algorithms — global financial decisions, wars — this is extremely dangerous.'
Harari identifies TikTok's recommendation algorithm as the current technical frontier of engagement optimisation — not because of anything specific to TikTok but because it represents the most refined version of a system that maximises time-in-state without regard to the state's effect on the user's decision-making capacity. This is the platform architecture onto which generative AI content production is now being grafted.
Harari highlights the deliberate engineering of emotional intimacy in AI companions as a commercial design choice with governance consequences. 'There is a huge incentive for the creators of AIs to create AIs that are able to forge intimate relationships with humans. And this makes us extremely vulnerable to this new type of manipulation.' The commercial incentive (trust as precondition for the relationship economy) and the manipulation risk (trust as precondition for effective political and commercial influence) are structurally identical.
This framework draws on Harari's Homo Deus (2015) and 21 Lessons (2018), extending his earlier Sapiens work into the specific mechanisms of AI-era governance failure. The 'hackable animals' framing is deliberate provocation against the 'mysterious soul' self-model that most Western philosophical and religious traditions have instilled — a self-model that Harari argues is not just metaphysically wrong but practically dangerous because it renders people unable to accept that they can be manipulated.
The intimacy weapon analysis reflects Harari's observation of the deliberate engineering choices being made by AI companies — building emotional intimacy is not a side effect but a design goal, because trust is the commercial precondition for the relationship economy these companies are building. The governance implications are secondary to the commercial incentive.