At Rise of AI 2017 in Berlin, Martin Spindler, describing himself as a historian rather than a futurist, argues that looking at past visions of the future is the best way to calibrate forecasts about AI and 2030. He opens with a 1970 artwork imagining a global information network and early computers used for intercontinental missile calculations, to show that people overestimate technology's short-term impact and underestimate its long-term impact. He traces Moore's Law and notes computing power grew by roughly a billion times between the 1956 Dartmouth conference, where the term artificial intelligence was coined, and today. He tests the claim that AI will transform the world like electricity did by revisiting electrification's history of platform battles, military applications and today's collapsing coal stock values, then lays out four scenarios for 2030: a runaway technological singularity, sigmoid-curve disruption as seen in self-driving cars and media, data-driven centralization, and a volatile 'world of asymmetries' created by cheap, widely accessible AI, illustrated with consumer drones used as weapons.
Key takeaways
- Spindler argues societies severely overestimate technology's short-term impact and underestimate its long-term impact, and the two errors differ in kind.
- He notes computing power grew by roughly a billion times between the 1956 Dartmouth conference and today.
- He tests the claim that 'AI is the next electricity' against electrification's real history of platform battles, military misuse and industry concentration.
- He cites the top four US coal stocks losing about 99 percent of their value over four years as a sign of the disruption implied by that analogy.
- He describes consumer drones used as weapons in Mosul costing about 200 dollars against 1.3 million dollar Patriot missiles as a warning about cheap, decentralized AI enabling asymmetric conflict.
Detailed notes
Learning from past futures
Spindler frames his approach as historical rather than predictive, using a 1970 artwork by the Italian collective Superstudio, imagining a globe-spanning information network, and period computers built to calculate intercontinental missile flight paths, to show how differently the future of computing actually unfolded. His central claim: societies overestimate the short-term impact of new technologies and severely underestimate their long-term impact, and the two errors differ in kind, not just degree.
Moore's Law and the scale of exponential change
He revisits Moore's Law, framed by Intel co-founder Gordon Moore, describing processors doubling in speed and capacity at the same cost roughly every two years. He contrasts a 1960s missile-guidance computer costing an inflation-adjusted 35 million dollars with a modern smartwatch, and notes that from the 1956 Dartmouth conference, where the term artificial intelligence was coined and researchers expected to model the brain within a few years, to today, compute power has grown by roughly one billion times.
Testing the 'AI is the next electricity' claim
Responding to a prominent AI researcher's claim, aired earlier at the conference, that AI will transform the world the way electricity did, Spindler revisits electrification's actual history: the rivalry between Thomas Edison and Nikola Tesla and George Westinghouse over electrical standards, the political ideology built around electrification, its military applications such as the 'demon core' criticality accidents, and the industry concentration that followed. He points to today's collapsing coal industry, saying the top four US coal stocks lost about 99 percent of their value over four years, and notes the Bank of England has warned about the macroeconomic risk of a similar bubble in fossil-fuel valuations, as the implied endpoint of treating AI like electricity.
Four scenarios for 2030
Spindler sketches four plausible, non-exhaustive scenarios. The first is the technological singularity, where AI improves faster than humans can control it, which he calls possible but not preferable. The second applies a sigmoid diffusion curve to innovation, as in self-driving cars and the earlier disruption of media and music, arguing that meaningful innovation in outputs only happens once the underlying technical infrastructure is settled. The third is centralization, where treating 'data as the new oil' implies heavy upfront investment concentrating power on a few platforms, eroding the middle class, an idea he links to Thomas Piketty's work. The fourth, his favorite, is a 'world of asymmetries', where falling costs for renewable energy, batteries and computation make technology cheap and accessible, fueling both creativity and volatility, illustrated by consumer drones used as weapons in Mosul that cost about 200 dollars against Patriot interceptor missiles priced at roughly 1.3 million dollars each.
Quotes
We tend to overestimate the impact of technologies in the short term and tend to severely underestimate the impact of technologies in the long term.
Martin Spindler
Top four US coal stocks of last four years lost ninety-nine percent of their value.
Martin Spindler
Questions answered in this talk
Why does Martin Spindler compare AI to the history of electricity?
According to Martin Spindler at Rise of AI 2017, an AI researcher claimed AI will transform the world like electricity did; Spindler revisits electrification's platform battles, ideology, military misuse and industry concentration to show what that comparison actually implies for AI.
How much has computing power grown since AI was first defined?
According to Martin Spindler at Rise of AI 2017, computing power has grown by roughly one billion times between the 1956 Dartmouth conference, where the term artificial intelligence was coined, and today.
What scenarios does Martin Spindler outline for AI and society by 2030?
According to Martin Spindler at Rise of AI 2017, the four scenarios are a runaway technological singularity, sigmoid-curve industry disruption like self-driving cars, data-driven centralization of power, and a volatile 'world of asymmetries' created by cheap, widely accessible AI.