01What "intelligence explosion" means
The term comes from statistician I. J. Good in 1965. The idea: "A machine smarter than humans designs a machine smarter than itself, which designs another even smarter, and so on —" When that chain starts, intelligence grows at an accelerating rate.
For decades, this was a thought experiment. Since the 2023 generative-AI wave, it has been reassessed as "a near-future reality":
- 2024: OpenAI o1 — AI gained "time to think"
- 2025: Devin / Claude Code — AI handles long autonomous coding sessions
- 2026: AI now co-authors parts of AI research papers
When AI can accelerate AI research itself, development reaches a speed that does not require human-hand intervention. That is the threshold.
02"When?" — what serious people actually predict
Major industry estimates:
- ~ 2028: AGI predictions — many leaders at Anthropic, OpenAI, DeepMind say "AI surpassing humans on most cognitive tasks" arrives around 2027–2028
- ~ 2030: AI automating AI research — the threshold Dario Amodei (Anthropic) names; humans become "supervisors"
- 2030 ~ 2035: Rapid capability explosion — possibility of "a year of progress per month". Leopold Aschenbrenner's "Situational Awareness" (2024) maps this in detail
- Disputed: Yann LeCun (Meta) argues the current LLM trajectory won't reach AGI; timelines may delay by 10–20 years
What matters is that the industry's median estimate is "sooner than people expect". Living between "definitely will" and "definitely won't" — i.e., on the assumption that "there's a non-trivial chance, and possibly sooner than thought" — is the rational stance.
03Five structural changes to society
If intelligence explosion happens, the structure of society may shift in the following ways.
- Essential change in labor — most cognitive work (clerical, analytical, junior development, content generation) becomes replaceable. What remains for humans: not "what to do" but "for whom, and why"
- Concentration of power — countries, companies, and individuals with frontier AI gain unprecedented advantage. Those without lose competitiveness rapidly
- Collapse of timescales — progress outpaces the speed at which humans can adapt. Ethics, law, and social consensus cannot keep up
- Erosion of truth — high-quality fake videos, audio, and documents are mass-produced. The social ground for "what is real" wobbles
- Redefinition of wealth — material scarcity may be rapidly solvable. But "who benefits" becomes the new class question
04Daily life — imagine it concretely
Not abstractly. Picture a morning a few years after the threshold:
An AI assistant wakes you instead of the alarm. From last night's sleep data, it reports: "I've cut two of your meetings today."
The kids' homework is personalized by an AI tutor. The school report now includes "AI literacy" as a graded item.
At work, your "AI team" of five receives the morning brief. You confirm finished tasks and issue new instructions.
Lunch break, your mother calls: "I asked the AI to explain my health check-up — it said the following." Medical consultation through AI has become daily.
Evening news. Over half the articles carry an "AI-generated" label. The rest are the premium "human-written" tier.
In that landscape, people will probably hold both "things easier than now" and "things more restless than now". Materially richer, yet a constant pressure of one's role being redefined. Everyone confronts their own replaceability, every day.
05Family and relationships
Less discussed but important — how this shifts the family:
- Generational gaps widen: younger people using AI fluently vs. older people unable — a wider divide than before
- "Raising" changes meaning: if AI personalizes children's learning, parents' "teaching" role shrinks. What replaces it requires social redesign
- Partnership: if only one partner masters AI, household information and decision-making balance breaks
- "Conversation" becomes scarce: more time talking to AI necessarily means less talking with humans. Deep human relationships require deliberate effort to maintain
06Politics and democracy
More serious still: the impact on politics.
- Elections become AI-generated content battlegrounds — fake videos, fake audio, impersonation posts. Visible already in the 2024 US election
- Sophisticated opinion manipulation — "microtargeting 2.0" where AI sends person-specific messages tuned to individual psychology
- Regulation cannot keep pace — legislative processes can't match AI's progress speed. "Post-hoc regulation" becomes the norm
- Military applications — autonomous weapons, cyber-attack AI accelerate. "AI arms-control treaty" becomes a real agenda
- Democracy vs. authoritarianism — transparency-prioritizing democracies and speed-prioritizing authoritarian states diverge in AI strategy
07What will happen in pharma and medicine
For our readers, the direct implications:
- Dramatic drug-discovery acceleration: AlphaFold-class evolution may bring new drug-candidate discovery from "years" to "weeks"
- Clinical trial redesign: synthetic control arms and digital twins reduce required patient numbers substantially
- Real personalized medicine: treatment design based on individual genomic and history data, at viable cost
- Diagnostic AI assistance as norm: the era of physicians diagnosing without AI input ends
- Regulators' new role: redesign of AI-driven drug development and approval. FDA, PMDA, EMA each exploring
- "Access inequality": wealthy countries / patients benefiting from AI discovery vs. those who don't — a new medical-ethics issue
08What humanity should prepare — seven priorities
Invest in AI safety research
Technology to guarantee AI "works correctly" (interpretability, alignment) lags behind capability research. Human intellect must concentrate here.
International AI governance agreement
A framework like the NPT for AI. Full agreement is unlikely, but a minimum "must-not-do" list is essential.
Full redesign of education
From memorization to AI-fluency and AI-resistant judgment. School and university curricula rewritten within 5 years.
Rebuilt social safety net
When employment shifts abruptly, livelihood guarantees during transitions, retraining programs, and serious UBI discussion become necessary.
Individual AI literacy
Every individual gaining the capacity to use, understand, judge, and ethically handle AI. Not optional — a precondition of social participation.
Rediscover "what only humans can do"
Empathy, creative leaps, physicality, art, religion, philosophy. Deliberately cultivate domains AI cannot replace.
Preserve "the culture of waiting"
In an instant-everything era, deliberately preserve cultures that think in long time horizons (philosophy, long-range planning, traditional rituals).
09"Not preparing" is no longer an option
What I most want this essay to convey: "This doesn't concern me" / "Maybe it'll happen, but not in my lifetime" — I understand the wish to think this way. I myself, when healthy, considered patient-rights protection somebody else's problem.
But the intelligence-explosion question is of the kind where "by the time you notice, the structure has already changed". Inn towns before railroads, painters before photography, bookstores before the internet — only those who began preparing right before the structural change kept a place after it.
The exact probability of intelligence explosion, I don't know. Maybe tomorrow, maybe ten years, maybe never.
But "the probability is high enough that it can't be ignored" — that consensus in the industry is no longer movable. Given that, living on that premise is what rational adults do.
What AI news will say next week or month, I don't know. But for myself, my family, my workplace, "if it changes suddenly, what would I do?" — thinking about that now ── that, I believe, is the responsibility of those of us standing at the threshold.