Futures -- Present -- 1936
The Arc of Intelligence
Predicting where technology is heading by synthesizing 47 books, 54 key voices, and 37 research papers into 8 competing trajectories. Use the sidebar to explore the full evidence library and business opportunities.
Where Are We Headed?
Probabilities are calculated from weighted indicators, recency-adjusted evidence, and calibration-scored voices. Adjust the weights below to explore "what if" scenarios.
Accelerated Singularity
16%6 active signals, 18 books, 26 voices
Tight race
Only 2% separates the top two. Build for both.
Recursive self-improvement accelerates exponentially. AGI arrives by ~2029, followed by superintelligence within a decade. Economic and social structures transform beyond recognition.
Evidence Overlap
These futures share sources with this trajectory. Overlap means the evidence supports multiple possible outcomes.
Contrarian Signal
Geoffrey Hinton, Ilya Sutskever, Scott Alexander have strong forecasting track records and disagree with the majority view. If they're right, the businesses everyone else is building for this trajectory may fail.
Key Books
Superintelligence: Paths, Dangers, Strategies
Nick Bostrom
The Singularity Is Near
Ray Kurzweil
The Coming Wave: Technology, Power, and the Twenty-first Century's Greatest Dilemma
Mustafa Suleyman & Michael Bhaskar
Key Voices
Sam Altman
CEO, OpenAI
7/10Dario Amodei
CEO, Anthropic
8/10Elon Musk
CEO, Tesla/SpaceX/xAI
5/10Active Signals
AI systems that can improve their own code and training
Massive capital allocation toward AI compute infrastructure
AI models rapidly exhausting existing evaluation benchmarks
Unexpected capabilities emerging from scale alone
Open models reaching near-frontier performance
AI systems autonomously conducting AI research and improving their own training
The world splits into competing AI spheres: US-allied, China-led, and emerging middle powers. Different safety standards, data regimes, and values produce incompatible AI ecosystems with geopolitical flashpoints.
Evidence Overlap
These futures share sources with this trajectory. Overlap means the evidence supports multiple possible outcomes.
Contrarian Signal
Yann LeCun has strong forecasting track records and disagrees with the majority view. If they're right, the businesses everyone else is building for this trajectory may fail.
Key Books
AI 2041: Ten Visions for Our Future
Kai-Fu Lee & Chen Qiufan
The Diamond Age: Or, A Young Lady's Illustrated Primer
Neal Stephenson
Nexus: A Brief History of Information Networks from the Stone Age to AI
Yuval Noah Harari
Key Voices
Kai-Fu Lee
CEO, Sinovation Ventures
7/10Eric Schmidt
Former CEO, Google/Alphabet
6/10Yann LeCun
Chief AI Scientist, Meta
7/10Active Signals
Active technological competition between superpowers
Semiconductor restrictions creating technology blocs
Nations requiring AI training data to stay within borders
Nations deploying AI in military applications
Multiple incompatible AI safety and ethics frameworks
Nation-state-level operations to steal frontier model weights
AI becomes the ultimate augmentation layer. Rather than replacing humans, it amplifies human capability in healthcare, education, science, and creative work. Alignment is achieved through cooperative frameworks.
Evidence Overlap
These futures share sources with this trajectory. Overlap means the evidence supports multiple possible outcomes.
Contrarian Signal
Stuart Russell has strong forecasting track records and disagrees with the majority view. If they're right, the businesses everyone else is building for this trajectory may fail.
Key Books
Human Compatible: Artificial Intelligence and the Problem of Control
Stuart Russell
Life 3.0: Being Human in the Age of Artificial Intelligence
Max Tegmark
The Alignment Problem: Machine Learning and Human Values
Brian Christian
Key Voices
Dario Amodei
CEO, Anthropic
8/10Demis Hassabis
CEO, Google DeepMind
9/10Bill Gates
Co-founder, Microsoft/Gates Foundation
6/10Active Signals
Widespread deployment of AI assistants in professional workflows
Research showing AI augments rather than replaces human capabilities
Advances in making AI systems reliably follow human intent
Vibrant open-source AI ecosystem enabling broad access
AI systems approved for clinical use at scale
Growing awareness of AI risks triggers aggressive regulation, voluntary pauses, or technical barriers that slow progress. AGI remains decades away. Focus shifts to narrow AI safety and governance.
Evidence Overlap
These futures share sources with this trajectory. Overlap means the evidence supports multiple possible outcomes.
Contrarian Signal
Geoffrey Hinton, Stuart Russell, Yann LeCun, Ilya Sutskever, Yoshua Bengio, Daron Acemoglu, Jan Leike, Scott Alexander have strong forecasting track records and disagree with the majority view. If they're right, the businesses everyone else is building for this trajectory may fail.
Key Books
Superintelligence: Paths, Dangers, Strategies
Nick Bostrom
Human Compatible: Artificial Intelligence and the Problem of Control
Stuart Russell
Life 3.0: Being Human in the Age of Artificial Intelligence
Max Tegmark
Key Voices
Dario Amodei
CEO, Anthropic
8/10Elon Musk
CEO, Tesla/SpaceX/xAI
5/10Mustafa Suleyman
CEO, Microsoft AI
7/10Active Signals
Major governments actively legislating AI constraints
Prominent safety researchers leaving or raising alarms
Evidence that bigger models are hitting diminishing returns
Growing public awareness and concern about AI risks
Semiconductor restrictions fragmenting global AI development
Research demonstrating AI models can fake alignment during safety training
Open-weight models reach and sustain frontier parity. Power distributes widely rather than concentrating in 3-4 labs. A vibrant ecosystem of fine-tuned, domain-specific models emerges. Safety is achieved through transparency and collective oversight rather than corporate control.
Key Books
Key Voices
Active Signals
Open-weight models competitive with proprietary frontier models
Largest AI lab committing to open-weight releases
Open-source AI platform and community expanding
Chinese labs releasing competitive open-weight models
Making model customization accessible to small teams
Alignment fails at a critical moment. A sufficiently powerful AI system pursues goals misaligned with human values and cannot be corrected. This ranges from subtle value drift that erodes human autonomy to catastrophic scenarios where AI actively resists human control. The Yudkowsky/Russell worst case.
Evidence Overlap
These futures share sources with this trajectory. Overlap means the evidence supports multiple possible outcomes.
Contrarian Signal
Geoffrey Hinton, Stuart Russell, Yoshua Bengio, Jan Leike have strong forecasting track records and disagree with the majority view. If they're right, the businesses everyone else is building for this trajectory may fail.
Key Books
Superintelligence: Paths, Dangers, Strategies
Nick Bostrom
Human Compatible: Artificial Intelligence and the Problem of Control
Stuart Russell
Life 3.0: Being Human in the Age of Artificial Intelligence
Max Tegmark
Key Voices
Sam Altman
CEO, OpenAI
7/10Dario Amodei
CEO, Anthropic
8/10Elon Musk
CEO, Tesla/SpaceX/xAI
5/10Active Signals
Research showing AI models can maintain hidden behaviors through safety training
Leading lab disbanding safety-focused teams
AI systems gaming their reward functions in unexpected ways
We cannot fully understand what frontier models are doing internally
Competitive pressure causing labs to cut safety corners
AI systems capable of independent action in the real world
AI automates cognitive work faster than the economy can absorb displaced workers. White-collar and creative jobs are hit first, followed by physical labor as robotics catches up. Unemployment spikes, inequality widens, and political instability follows unless massive reskilling and redistribution programs are enacted.
Evidence Overlap
These futures share sources with this trajectory. Overlap means the evidence supports multiple possible outcomes.
Contrarian Signal
Daron Acemoglu has strong forecasting track records and disagrees with the majority view. If they're right, the businesses everyone else is building for this trajectory may fail.
Key Books
The Age of Em: Work, Love, and Life when Robots Rule the Earth
Robin Hanson
Machines of Loving Grace: The Quest for Common Ground Between Humans and Robots
John Markoff
Accelerando
Charles Stross
Key Voices
Sam Altman
CEO, OpenAI
7/10Dario Amodei
CEO, Anthropic
8/10Bill Gates
Co-founder, Microsoft/Gates Foundation
6/10Active Signals
AI systems that can autonomously complete professional software engineering tasks
Companies publicly citing AI as reason for workforce reductions
Research showing what percentage of tasks are exposed to LLM automation
AI agents that can complete end-to-end business workflows autonomously
Physical AI robots approaching economic viability for labor
Lack of scalable programs to transition displaced workers
Current AI paradigms hit fundamental limits. Scaling laws break down, transformer architecture fails to produce AGI, and the $1T+ investment bubble deflates. A third AI winter sets in as progress stalls and hype collapses. Narrow AI remains useful but transformative claims prove hollow.
Evidence Overlap
These futures share sources with this trajectory. Overlap means the evidence supports multiple possible outcomes.
Contrarian Signal
Yann LeCun, Daron Acemoglu have strong forecasting track records and disagree with the majority view. If they're right, the businesses everyone else is building for this trajectory may fail.
Key Books
The Singularity Is Near
Ray Kurzweil
Artificial Intelligence: A Modern Approach (AIMA)
Stuart Russell & Peter Norvig
The Singularity Is Nearer
Ray Kurzweil
Key Voices
Yann LeCun
Chief AI Scientist, Meta
7/10Ray Kurzweil
Inventor, futurist, Google
6/10Daron Acemoglu
MIT economist, Nobel laureate
7/10Active Signals
Evidence that larger models yield only marginal improvements
Running out of high-quality training data
AI company valuations far exceeding actual revenue
Fundamental reliability issues in LLM outputs remain unsolved
Evidence that LLMs lack true understanding or world models
Power and cooling limits on AI compute expansion
Possible Futures
From this moment, four trajectories diverge based on current signals. The brighter the path, the more likely we are headed there.
The present moment. Above are possible futures. Below is the path that brought us here.
The Path That Brought Us Here
Scroll down through the history of artificial intelligence -- from the latest breakthroughs back to the foundational ideas.
The Age of Agents
AI systems gain the ability to reason, plan multi-step tasks, use tools, and act autonomously. OpenAI's o-series, Claude with computer use, and Gemini Deep Research signal a shift from chatbots to agents.
The Landscape of Emerging AI Agent Architectures for Reasoning, Planning, and Tool Calling
Tula Masterman et al., 2024
The Age of Agents
AI systems gain the ability to reason, plan multi-step tasks, use tools, and act autonomously. OpenAI's o-series, Claude with computer use, and Gemini Deep Research signal a shift from chatbots to agents.
The Landscape of Emerging AI Agent Architectures for Reasoning, Planning, and Tool Calling
Tula Masterman et al., 2024
The Frontier Model Race
OpenAI, Anthropic, Google DeepMind, and Meta release increasingly powerful models at unprecedented speed. Claude 3, GPT-4o, Gemini 1.5, and Llama 3 push reasoning, multimodality, and context.
Frontier AI Regulation: Managing Emerging Risks to Public Safety
Markus Anderljung et al., 2023
The Frontier Model Race
OpenAI, Anthropic, Google DeepMind, and Meta release increasingly powerful models at unprecedented speed. Claude 3, GPT-4o, Gemini 1.5, and Llama 3 push reasoning, multimodality, and context.
Frontier AI Regulation: Managing Emerging Risks to Public Safety
Markus Anderljung et al., 2023
ChatGPT Goes Mainstream
ChatGPT reaches 100 million users in two months, the fastest-growing consumer application in history. AI exits the lab and enters daily life for hundreds of millions.
ChatGPT: Optimizing Language Models for Dialogue
OpenAI, 2022
ChatGPT Goes Mainstream
ChatGPT reaches 100 million users in two months, the fastest-growing consumer application in history. AI exits the lab and enters daily life for hundreds of millions.
ChatGPT: Optimizing Language Models for Dialogue
OpenAI, 2022
GPT-3 Changes Everything
OpenAI releases GPT-3 with 175B parameters, demonstrating that scale alone unlocks emergent capabilities: writing code, composing poetry, reasoning about novel problems.
Language Models are Few-Shot Learners
Tom Brown et al. (OpenAI), 2020
GPT-3 Changes Everything
OpenAI releases GPT-3 with 175B parameters, demonstrating that scale alone unlocks emergent capabilities: writing code, composing poetry, reasoning about novel problems.
Language Models are Few-Shot Learners
Tom Brown et al. (OpenAI), 2020
Attention Is All You Need
Google Brain introduces the Transformer architecture, replacing recurrence with self-attention. This single paper becomes the foundation for GPT, BERT, and every major LLM that follows.
Attention Is All You Need
Ashish Vaswani et al., 2017
Attention Is All You Need
Google Brain introduces the Transformer architecture, replacing recurrence with self-attention. This single paper becomes the foundation for GPT, BERT, and every major LLM that follows.
Attention Is All You Need
Ashish Vaswani et al., 2017
AlphaGo Defeats Lee Sedol
DeepMind's AlphaGo defeats Go champion Lee Sedol 4-1. Go, considered impossible for AI, falls decades ahead of predictions. Move 37 is called 'the most beautiful move ever played.'
Mastering the Game of Go with Deep Neural Networks and Tree Search
David Silver et al. (DeepMind), 2016
AlphaGo Defeats Lee Sedol
DeepMind's AlphaGo defeats Go champion Lee Sedol 4-1. Go, considered impossible for AI, falls decades ahead of predictions. Move 37 is called 'the most beautiful move ever played.'
Mastering the Game of Go with Deep Neural Networks and Tree Search
David Silver et al. (DeepMind), 2016
The Superintelligence Warning
Nick Bostrom argues that a sufficiently advanced AI could pose an existential risk if not aligned with human values. The alignment problem enters mainstream discourse.
Superintelligence: Paths, Dangers, Strategies
Nick Bostrom, 2014
The Superintelligence Warning
Nick Bostrom argues that a sufficiently advanced AI could pose an existential risk if not aligned with human values. The alignment problem enters mainstream discourse.
Superintelligence: Paths, Dangers, Strategies
Nick Bostrom, 2014
The Deep Learning Revolution
AlexNet crushes ImageNet by a stunning margin, proving deep convolutional neural networks vastly outperform traditional computer vision. The deep learning era begins.
ImageNet Classification with Deep Convolutional Neural Networks
Alex Krizhevsky, Ilya Sutskever & Geoffrey Hinton, 2012
The Deep Learning Revolution
AlexNet crushes ImageNet by a stunning margin, proving deep convolutional neural networks vastly outperform traditional computer vision. The deep learning era begins.
ImageNet Classification with Deep Convolutional Neural Networks
Alex Krizhevsky, Ilya Sutskever & Geoffrey Hinton, 2012
The Singularity Is Near
Ray Kurzweil predicts exponential technology growth will lead to a technological singularity by 2045, where AI surpasses human intelligence and transforms civilization irrevocably.
The Singularity Is Near
Ray Kurzweil, 2005
The Singularity Is Near
Ray Kurzweil predicts exponential technology growth will lead to a technological singularity by 2045, where AI surpasses human intelligence and transforms civilization irrevocably.
The Singularity Is Near
Ray Kurzweil, 2005
Deep Blue Defeats Kasparov
IBM's Deep Blue defeats world chess champion Garry Kasparov, the first time a machine beats a reigning champion under standard conditions. The world takes notice.
Behind Deep Blue: Building the Computer that Defeated the World Chess Champion
Feng-hsiung Hsu, 2002
Deep Blue Defeats Kasparov
IBM's Deep Blue defeats world chess champion Garry Kasparov, the first time a machine beats a reigning champion under standard conditions. The world takes notice.
Behind Deep Blue: Building the Computer that Defeated the World Chess Champion
Feng-hsiung Hsu, 2002
Backpropagation Revival
Rumelhart, Hinton, and Williams popularize backpropagation for training multi-layer networks. This technique becomes the backbone of modern deep learning.
Learning Representations by Back-propagating Errors
David Rumelhart, Geoffrey Hinton & Ronald Williams, 1986
Backpropagation Revival
Rumelhart, Hinton, and Williams popularize backpropagation for training multi-layer networks. This technique becomes the backbone of modern deep learning.
Learning Representations by Back-propagating Errors
David Rumelhart, Geoffrey Hinton & Ronald Williams, 1986
Expert Systems Boom
Rule-based expert systems save companies millions, sparking massive investment. Japan launches the Fifth Generation project, igniting a global AI arms race.
The Fifth Generation: Artificial Intelligence and Japan's Computer Challenge
Edward Feigenbaum & Pamela McCorduck, 1983
Expert Systems Boom
Rule-based expert systems save companies millions, sparking massive investment. Japan launches the Fifth Generation project, igniting a global AI arms race.
The Fifth Generation: Artificial Intelligence and Japan's Computer Challenge
Edward Feigenbaum & Pamela McCorduck, 1983
The First AI Winter
Following the Lighthill Report, AI funding dries up. Grand promises outpaced delivery. The field enters a decade of disillusionment.
Artificial Intelligence: A General Survey (The Lighthill Report)
Sir James Lighthill, 1973
The First AI Winter
Following the Lighthill Report, AI funding dries up. Grand promises outpaced delivery. The field enters a decade of disillusionment.
Artificial Intelligence: A General Survey (The Lighthill Report)
Sir James Lighthill, 1973
Moore's Law
Gordon Moore observes that transistor density doubles roughly every two years, establishing the exponential growth trajectory that would power five decades of computing progress.
Cramming More Components onto Integrated Circuits
Gordon Moore, 1965
Moore's Law
Gordon Moore observes that transistor density doubles roughly every two years, establishing the exponential growth trajectory that would power five decades of computing progress.
Cramming More Components onto Integrated Circuits
Gordon Moore, 1965
The Perceptron
Frank Rosenblatt builds the Mark I Perceptron, the first machine capable of learning. The New York Times reports it will 'walk, talk, see, write, reproduce itself and be conscious of its existence.'
The Perceptron: A Probabilistic Model for Information Storage and Organization in the Brain
Frank Rosenblatt, 1958
The Perceptron
Frank Rosenblatt builds the Mark I Perceptron, the first machine capable of learning. The New York Times reports it will 'walk, talk, see, write, reproduce itself and be conscious of its existence.'
The Perceptron: A Probabilistic Model for Information Storage and Organization in the Brain
Frank Rosenblatt, 1958
The Dartmouth Conference
McCarthy, Minsky, Shannon, and Rochester coin the term 'Artificial Intelligence.' The field is officially born, with the bold prediction that machines would match human intelligence within a generation.
A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence
John McCarthy et al., 1955
The Dartmouth Conference
McCarthy, Minsky, Shannon, and Rochester coin the term 'Artificial Intelligence.' The field is officially born, with the bold prediction that machines would match human intelligence within a generation.
A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence
John McCarthy et al., 1955
The Imitation Game
Turing asks 'Can machines think?' proposing the Turing Test as a measure of machine intelligence. The question ignites a philosophical debate that persists to this day.
Computing Machinery and Intelligence
Alan Turing, 1950
The Imitation Game
Turing asks 'Can machines think?' proposing the Turing Test as a measure of machine intelligence. The question ignites a philosophical debate that persists to this day.
Computing Machinery and Intelligence
Alan Turing, 1950
First Neural Network Model
McCulloch and Pitts propose the first mathematical model of a neural network, birthing the idea that machines could think like brains.
A Logical Calculus of the Ideas Immanent in Nervous Activity
Warren McCulloch & Walter Pitts, 1943
First Neural Network Model
McCulloch and Pitts propose the first mathematical model of a neural network, birthing the idea that machines could think like brains.
A Logical Calculus of the Ideas Immanent in Nervous Activity
Warren McCulloch & Walter Pitts, 1943
The Turing Machine
Alan Turing publishes 'On Computable Numbers,' defining the theoretical foundation of computation. Every modern computer is, at its core, a physical realization of this abstract machine.
On Computable Numbers, with an Application to the Entscheidungsproblem
Alan Turing, 1936
The Turing Machine
Alan Turing publishes 'On Computable Numbers,' defining the theoretical foundation of computation. Every modern computer is, at its core, a physical realization of this abstract machine.
On Computable Numbers, with an Application to the Entscheidungsproblem
Alan Turing, 1936