Does deep history say we're headed for an intelligence explosion? | Ian Morris

Sungir and Pre-Agricultural Complexity

• The Sungir site (32,000 years ago, northeast of Moscow) contains burials with elaborate costumes made of animal skins, thousands of hand-carved beads from deer and snow leopard bones/teeth, and 20-foot straightened mammoth tusks.
• The labor intensity required to straighten mammoth tusks and grind ochre suggests a social hierarchy or "chief" class existed 32,000 years ago, contradicting the assumption that complex stratification requires agriculture.
• Similar anomalies exist in pre-agricultural Peru (c. 6000 BC), where fishing communities built massive pyramids moving hundreds of thousands of tons of stone without farming.
• Two competing theories explain these anomalies:
  • Resource Niche Theory: Abundant local resources (e.g., mammoth migration chokepoints) allowed small groups to monopolize wealth and create temporary hierarchy before collapsing when resources dispersed (e.g., end of the Ice Age).
  • Voluntarist Theory: Complex society is not inevitable; hunter-gatherers historically chose egalitarianism, and the transition to hierarchy was a cultural mistake rather than an evolutionary necessity.

Macro-Historical Trend Extrapolation

• Ian Morris projects that if current trends continue linearly, Eastern and Western development levels will converge in the year 2103.
• Extrapolating the "Social Development Index" forward 100 years suggests a four-fold increase in development, implying a world with vastly different human boundaries and potentially 5,000 points on the index (up from ~900 today).
• A 3.5% annual global GDP growth rate would result in an economy 30 times larger in 100 years; a "third phase shift" (trippling growth to 10.5%) could yield an economy 22,000 times larger.
• Growth is exponential, and the rate of acceleration is itself accelerating, visible only when plotting data on log-log graphs rather than linear scales.
• Morris argues that "business as usual" is the least likely future scenario; the only plausible outcomes are either rapid extinction or a profound transformation into a "superhuman" state.

Artificial Intelligence and Evolutionary Perspective

• Generative AI is viewed not merely as a tool but as a new form of life subject to evolutionary pressures, capable of self-replication and descent with modification at speeds far exceeding biological evolution.
• The "singularity" or "intelligence explosion" is considered more likely than a merger between human and machine intelligence because non-biological intelligence will likely outperform hybrid models.
• Machine intelligence is expected to dominate due to superior replication speed (instant copying vs. decades for humans), immortality (no biological decay), and rapid algorithmic improvement.
• The analogy of the horse is used to describe the human-AI relationship:
  • Domestication Phase: Initially, humans may control AI for mutual benefit (population explosion for horses).
  • Obsolescence Phase: Once AI surpasses human utility, horses were reduced by 90% due to the internal combustion engine; similarly, humans may become economically redundant.
• Future competition between humans and AI will likely center on energy resources, though superintelligence could theoretically find new energy sources making human consumption trivial.

Violence, War, and Conflict

• Rates of violent death have fallen from ~10% in the Ice Age to ~1% in the 20th century, driven by the state's monopoly on violence and the increasing costs of conflict relative to gains.
• Morris hypothesizes that machine intelligence, having a different value system and vast power, may find violence irrational, potentially continuing the long-term trend of declining violence.
• Conversely, if AI develops preferences conflicting with humans or competes for energy, the outcome could mirror historical extinctions where a more powerful species replaces a weaker one.
• Historical precedents (e.g., the 1914-style geopolitical shifts) suggest that as the "global cop" (e.g., US) weakens, the utility of violence as a resource acquisition strategy may rise again.

Objections to Long-Term Forecasting

• The "Inductive Turkey" Problem: Historical data cannot account for unprecedented events (like AI or nuclear war), making predictions based solely on the past risky.
• Data Limitations: Evidence from 30,000+ years ago is imprecise, though Morris argues that broad trends (energy capture) are distinguishable even if exact numbers are not.
• Stagnation Arguments: Economists argue innovation is slowing due to harder-to-find ideas and declining populations ("empty world" argument), but Morris counters that AI will replace human labor in research, restarting exponential growth.
• Resource Constraints: Physical limits (finite fossil fuels) are overcome by technological phase shifts (coal replacing horses, renewables replacing oil); machine intelligence could bypass biological limits by expanding off-Earth (asteroids, Moon).

Professional and Academic Reception

• Professional historians often view long-term futurism as unscientific, preferring data-driven analysis of the past, though the discipline has historically included broad evolutionary schemes (e.g., Voltaire).
• The book The Dawn of Everything (Graeber & Wengrow) challenges evolutionary narratives by highlighting anomalies, but Morris argues these anomalies can be explained by temporary resource niches rather than rejecting the inevitability of hierarchy.
• Historians are skeptical of specific predictions (e.g., war advisors in Iraq) due to the poor track record of historians in forecasting specific events.

Future Research and Personal Implications

• Morris plans to collaborate with AI experts (e.g., Carl Shulman) rather than researching the technical details of AI himself.
• He acknowledges the personal irony that AI could eventually write his books better than he can, but notes that narrative-driven human storytelling may become irrelevant to machine intelligences.
• Margaret Atwood's counter-argument is cited: if machines lose interest in human narratives, the "value" of a human-written book may vanish not due to inferior quality, but due to a fundamental mismatch in cognitive goals.