LIVE INTELLIGENCE CARTOGRAPHY / 1950-NEXT / 28 NODES
Map the distance to general intelligence.
Move through 26 documented breakthroughs, track today's frontier signals, then stress-test what comes next.
Alan Turing reframes machine intelligence as an observable conversation rather than a metaphysical definition.
NODE 02 / FOUNDATIONS
A field declares its ambition.
The Dartmouth workshop gives artificial intelligence a name and a research agenda.
NODE 03 / EARLY LEARNING
A machine learns its own boundary.
Frank Rosenblatt demonstrates a trainable neural classifier implemented on dedicated hardware.
NODE 04 / LANGUAGE
Conversation creates an illusion of understanding.
Joseph Weizenbaum's ELIZA shows how quickly people project intelligence onto patterned dialogue.
NODE 05 / ROBOTICS
Reasoning enters the physical world.
SRI's Shakey links perception, planning, and action in one mobile robotic system.
NODE 06 / KNOWLEDGE
Expertise becomes deployable software.
Rule-based systems such as XCON move AI from laboratories into high-value commercial decisions.
NODE 07 / LEARNING
Credit flows backward through the network.
Backpropagation makes multilayer neural networks practical to train on internal representations.
NODE 08 / SEARCH
Machine search defeats a world champion.
IBM Deep Blue defeats Garry Kasparov in a regulation chess match.
NODE 09 / VISION
Neural vision reads the real world.
Convolutional networks learn to recognize handwritten digits in operational systems.
NODE 10 / DEEP LEARNING
Depth becomes trainable again.
Layer-wise pretraining renews interest in deep neural networks and learned hierarchical features.
NODE 11 / DATA
The field gets a shared visual world.
ImageNet assembles a large labeled image corpus that turns visual progress into a measurable race.
NODE 12 / DEEP LEARNING
Perception crosses a threshold.
Deep convolutional networks reset ImageNet and shift the field toward learned representations.
NODE 13 / LANGUAGE
Meaning gains a geometry.
Distributed word vectors reveal that semantic relationships can emerge as directions in learned space.
NODE 14 / GENERATION
Networks learn by competing.
Generative adversarial networks turn synthesis into a game between a generator and a discriminator.
NODE 15 / REINFORCEMENT
Intuition and search converge.
AlphaGo defeats Lee Sedol by combining deep networks, reinforcement learning, and tree search.
NODE 16 / ATTENTION
Sequence becomes parallel.
The Transformer replaces recurrence with attention and unlocks a new scaling path for general-purpose models.
NODE 17 / PRETRAINING
One model transfers across language tasks.
Bidirectional pretraining produces a reusable language representation that can be fine-tuned broadly.
NODE 18 / SCALE
Capability begins to emerge in context.
A 175-billion-parameter language model performs new tasks from instructions and examples without weight updates.
NODE 19 / SCIENCE
AI cracks a grand scientific challenge.
AlphaFold 2 reaches near-experimental accuracy on many protein-structure predictions.
NODE 20 / MULTIMODAL
Images and language share a map.
Contrastive pretraining aligns visual concepts with natural-language descriptions at web scale.
NODE 21 / GENERATION
Text becomes a visual instrument.
Latent diffusion makes high-quality text-to-image generation efficient enough for broad creative use.
NODE 22 / INTERFACE
Language becomes the interface.
A conversationally aligned foundation model moves advanced generative capability into everyday workflows.
NODE 23 / MULTIMODAL
A general model sees and reasons.
GPT-4 combines text and image input with stronger performance across professional and academic evaluations.
NODE 24 / SCIENCE
Biological interaction becomes predictable.
AlphaFold 3 models joint structures involving proteins, DNA, RNA, ligands, and other biomolecules.
NODE 25 / REASONING
Inference becomes a compute dimension.
OpenAI o1 demonstrates that models can improve difficult answers by spending more computation reasoning before responding.
NODE 26 / AGENTS
Models gain an operating layer.
New agent platforms combine reasoning models with tools, search, computer use, orchestration, and traces.
LIVE FRONTIER / CONTINUOUS SWEEP
The frontier is moving in real time.
VECTOR-01 continuously scans research and lab feeds for capability, autonomy, science, and safety signals.
SCENARIO / ASSUMPTION ENGINE
The next phase is a range, not a date.
Explore how capability velocity, compute efficiency, and autonomous reliability move the scenario window.
Cross-domain generalist systems
Waiting for the first persistent monitoring sweep.
A transparent scenario heuristic. It is designed to compare assumptions, not to certify an AGI arrival date.