At Rise of AI 2019 in Berlin, Daniel Jeffries lays out predictions for artificial intelligence across three horizons: five to fifteen years, fifty years and five hundred years. He argues that near-term change is slower than hype suggests, and that AI is a dual-use technology that reflects the values of its creators. In healthcare he expects faster triage and diagnosis, pointing to a 2017 Kaggle competition that produced lung cancer detection algorithms now used in hospitals. He describes a shift toward "Centaur" human-AI teams, a term drawn from Garry Kasparov's freestyle chess tournaments, and cites Stitch Fix as an example of AI-driven design paired with a human in the loop. He warns that China's lighter approach to data privacy, illustrated by an Alibaba-backed "city brain" system that a leaked, unsecured police database showed was tracking dissidents, could make China the world's AI training center while the EU and US debate ethics. Looking further out, he foresees the human connectome project enabling new chip designs, ubiquitous smart devices and self-driving trucks, and at five hundred years predicts humanity becomes "post-human" through deep integration with machines.
Key takeaways
- Jeffries argues meaningful AI change takes longer than hype suggests, citing the "nothing takes five minutes" line from Black Hawk Down.
- He predicts near-term healthcare gains, pointing to a 2017 Kaggle competition that produced lung cancer detection algorithms now used in hospitals.
- He describes "Centaur" teams, a term from Garry Kasparov's freestyle chess tournaments, where three experts plus AI beat lone grandmasters or AIs.
- He warns that an Alibaba-backed "city brain" system in a Chinese city was found, via a leaked unsecured database, to be tracking dissidents.
- He predicts that within 500 years humans merge with machines into "post-humans," outmoding Homo sapiens rather than being destroyed by superintelligence.
Detailed notes
Why near-term AI change is slower than expected
Jeffries opens by comparing prediction to Monte Carlo analysis: chaos theory rules out forecasting Black Swan events, but broad trend lines can be projected. He says five to fifteen years out "not much is going to happen," borrowing the line from Black Hawk Down that "nothing takes five minutes." He frames AI as a dual-use technology that is both good and evil because it reflects the values of its creators.
Near-term healthcare and Centaur teams
He expects the best near-term uses in healthcare, such as photographing a worrying skin spot to decide whether to see a doctor, and cites a 2017 Kaggle competition that produced lung cancer detection algorithms with a large false positive and negative problem, now deployed in hospitals. Rather than mass job loss, he expects more "Centaur" teams, a term from Garry Kasparov's freestyle chess tournaments, where a team of three human experts plus an AI beat both lone grandmasters and lone AIs. He gives Stitch Fix as an example: genetic algorithms generate fashion designs while a human designer curates them.
China's surveillance advantage and the ethics gap
Jeffries argues China is ahead in deployment, pointing to an Alibaba-backed "city brain" that monitors 50,000 cameras and detects car crashes within 20 seconds. He says a leaked, password-free police database showed that data was also used to track dissidents. He contrasts this with an AI ethics event in Paris focused on collaboration and human-centred AI, arguing that countries and companies less concerned with privacy will have a near-term advantage, and predicts firms will train AI in China on unrestricted data and bring the resulting weights back to the EU and US.
Fifty years: connected devices and human connectome projects
At fifty years, Jeffries envisions everyday objects becoming smart and connected, and progress driven by human connectome projects mapping every neuron in the brain, which he compares to the human genome project in cost and duration. He predicts new chip designs moving away from rare earth materials toward carbon-based, "wet" hardware, medical devices such as smart bandages and pill bottles, AI personal assistants, and self-driving trucks with no human cab running on dedicated roads.
Five hundred years: post-humanity
Jeffries predicts a possible World War 3 driven by job losses from automation, with AI directing military strategy and robotic soldiers, though he says war would accelerate technologies like quantum encryption and 3D printing rather than end civilization. At five hundred years he foresees interplanetary expansion and AI-run economies that absorb shocks in real time. He argues AI does eventually "destroy" humanity, but not through Nick Bostrom-style superintelligence: instead, humans merge with machines and become "post-human," a shift he compares to Homo sapiens outmoding Neanderthals rather than war, calling evolution "inevitable."
Quotes
AI, it's important to understand, is a dual use technology, it's both good and evil, it exists on a spectrum and it's going to reflect our values, us as the creator.
Daniel Jeffries
The question is who gets to define the bad guys in the near future, and the bad guy changes with the definition of who's in power.
Daniel Jeffries
Evolution is inevitable, you cannot hold back the tides of the future.
Daniel Jeffries
Questions answered in this talk
What does Daniel Jeffries predict for AI in healthcare in the near term?
According to Daniel Jeffries at Rise of AI 2019, near-term AI will help triage patients, such as photographing a worrying skin spot to decide whether to see a doctor, building on a 2017 Kaggle competition that produced lung cancer detection algorithms now used in hospitals.
What is a Centaur team in AI, according to Daniel Jeffries?
Jeffries explains the term comes from Garry Kasparov's freestyle chess tournaments, where a team combining human experts and AI outperformed both lone grandmasters and lone AIs, showing humans and AI can raise each other's capability.
How does Daniel Jeffries think humanity will change over the next 500 years?
Jeffries predicts humans will become "post-human" through deep integration with machines, gradually outmoding unmodified Homo sapiens much as Homo sapiens outmoded Neanderthals, rather than being destroyed by a Bostrom-style superintelligent machine.