Human Tech Tree

Information

AI & Robotics

39 points from the earliest roots to the research frontier: 27 researched, 6 current research, 4 unsolved and 2 that become possible once they are solved. State of knowledge: October 2026.

Open the interactive tree →

Antiquity3000 BC – 500 AD

  • Algorithm (Euclid)~300 BCResearchedEuclid’s procedure for the greatest common divisor is one of the oldest known algorithms: a rule instead of a single case.

Machine Age1900 – 1945

  • McCulloch-Pitts neuron1943ResearchedWarren McCulloch and Walter Pitts showed in 1943 that networks of simple on/off neurons can compute logic: the first neural network model.

Atomic & Space Age1945 – 1990

  • Cybernetics & Feedback Control1948ResearchedNorbert Wiener's 'Cybernetics' (1948) treats feedback, information and control in machines and animals alike.
  • Turing Test & Birth of AI1950ResearchedTuring asks in 1950 whether machines can think; in 1956 the Dartmouth conference names the field “artificial intelligence”.
  • Perceptron & Artificial Neurons1958ResearchedRosenblatt’s perceptron learns from examples instead of being programmed: the ancestor of all neural networks.
  • Kalman Filter1960ResearchedRudolf Kalman's filter (1960) combines noisy measurements with a model to track the best estimate of a moving state.
  • Industrial Robot (Unimate)1961ResearchedThe first industrial robot lifts hot die-cast parts at General Motors: robots replace people in dangerous repetitive work.
  • Shakey the Robot1966ResearchedSRI's Shakey (1966-72) is the first mobile robot that reasons about its own actions, planning moves from camera input.
  • Expert Systems1980ResearchedPrograms with thousands of hand-written if-then rules diagnose and configure: the first commercial AI, which then stalls in the winter.
  • Hopfield Net & Boltzmann Machine1982ResearchedHopfield (1982) stores memories as energy valleys of a neural network, and Hinton's Boltzmann machine (1985) learns with statistical physics.
  • Backpropagation1986ResearchedA method by which multilayer networks learn from their mistakes: the mathematical basis of today’s AI training.
  • Convolutional networks (LeNet)1989ResearchedYann LeCun's convolutional networks, trained by backpropagation (1989), read handwritten digits and became the blueprint for image AI.

Digital Age1990 – 2015

  • Learning by reward: TD-Gammon1992ResearchedTesauro's TD-Gammon (1992) learned backgammon by playing itself, proving that reward-driven learning with neural nets works.
  • Deep Blue Beats Kasparov1997ResearchedA chess computer with special chips defeats the world champion through raw search power, not understanding.
  • LSTM memory networks1997ResearchedHochreiter and Schmidhuber's LSTM (1997) let neural networks remember across long sequences and drove speech and translation until transformers.
  • DARPA Grand Challenge2005ResearchedIn 2005 five driverless vehicles finish a 212 km desert course; Stanford's Stanley wins and its engineers go on to build robotaxis.
  • ImageNet dataset2009ResearchedFei-Fei Li's team built ImageNet (2009): millions of labelled web photos that gave deep networks the data they needed.
  • Deep Learning2012ResearchedDeep neural networks trained on graphics cards with millions of images beat all classical methods in image recognition in 2012.
  • Word Embeddings (word2vec)2013ResearchedMikolov's word2vec (2013) turns words into vectors in which meaning shows up as direction, like king minus man plus woman.
  • Attention Mechanism2014ResearchedBahdanau, Cho and Bengio (2014) let a translation network look back at the most relevant input words while writing each output word.
  • Generative Adversarial Networks2014ResearchedGoodfellow's GAN (2014) trains a generator and a discriminator against each other and produces the first photorealistic fake images.

Present2015 – Oct 2026

  • AlphaGo2016ResearchedGoogle’s AlphaGo beats world-class Go player Lee Sedol 4-1, in a game long thought out of reach for decades.
  • Learning from Human Feedback2017ResearchedChristiano et al. (2017) teach models from human preference comparisons; OpenAI's InstructGPT (2022) uses it to make language models follow…
  • Transformer Architecture2017ResearchedGoogle’s “Attention Is All You Need” (2017) replaces sequential reading with attention: trainable in parallel and scalable almost without limit.
  • Pretrained Language Models2018ResearchedGPT (June 2018) and BERT (October 2018) are first trained on huge unlabelled text and then adapted to tasks.
  • Diffusion Models2020ResearchedHo et al. (2020) show that a network trained to remove noise step by step can generate images, building on ideas from non-equilibrium physics.
  • Neural Scaling Laws2020ResearchedKaplan et al. (2020) show that model quality improves smoothly and predictably with more data, parameters and compute.
  • Generative Image & Video AI2022Current researchDiffusion models turned text prompts into images in 2022 and video by 2024-25; open models run on a home GPU, and Veo 3 added synchronised sound in…
  • Large Language Models2022Current researchChatbots like ChatGPT (30 Nov 2022) write, translate and program in natural language: now an everyday tool.
  • Open-Weight Language Models2023-2026Current researchLanguage models whose trained weights anyone can download, from Meta's Llama 2 (2023) to DeepSeek-R1 (2025), can be run and fine-tuned by outsiders.
  • Reasoning Models2024Current researchModels that think at length before answering and are trained with reward learning crack olympiad math and hard programming tasks.
  • AI Agents2025Current researchAI systems that use tools on their own, write code and work through hours-long tasks: in 2026 the biggest lever of AI.
  • Humanoid Robots2026Current researchTwo-legged robots with AI control enter first volume production, but so far only work narrowly defined factory jobs.

Research Frontier · Todayunsolved as of Oct 2026

  • AI Alignment & InterpretabilityopenUnsolvedMaking sure powerful AI does what we truly want, and understanding what goes on inside it.
  • Artificial General IntelligenceopenUnsolvedAn AI system that learns and performs practically any mental task as well as a human: whether and when is disputed.
  • Everyday Robot DexterityopenUnsolvedRobots that reliably grasp, fold and tidy in a kitchen, home or building site, where nothing is standardized.
  • Reliable, Honest AIopenUnsolvedLanguage models still invent facts and sources while sounding convincing: real reliability is missing.

If Solvedbecomes possible

  • Autonomous AI Research2030s?If solvedAI systems that independently form hypotheses, plan experiments and check results: science in continuous operation.
  • Household & Care Robots2030s?If solvedAffordable robots that reliably help out in daily life (laundry, kitchen, care) without constant supervision.