Skip to content
Last Dictionary
Plane
Reinforcement LearningGenerative MediaPhilosophy & MindRobotics & Embodied AI

world model

nounAlso: world models, world simulator, internal model, dynamics model

In plain wordsAn AI's internal simulation of how its world works, which lets it predict what will happen next — especially in response to its own actions — and so plan, imagine or practice without acting for real.

  1. 1

    Reinforcement learning

    A learned model of an environmentenvironment, n.: Everything an AI agent interacts with but does not control — a game, a simulator, a website, the physical world. It answers each of the agent's actions with a new situation and, in reinforcement learning, a score.'s dynamics that predicts future observations or latent statesstate, n.: A snapshot of the situation an AI agent is in — for a chess program, the arrangement of pieces and whose turn it is — containing what it needs to decide what to do next., and usually rewardsreward, n.: The score an AI system receives after an action, telling it how good the outcome was. Reinforcement learning trains the system to collect as much reward as possible over time., given the current state and a proposed actionaction, n.: A choice an AI agent makes at one moment — move left, place a stone, type a command — which changes its situation and may earn it a reward., so that an agent can plan or train in imagination. Typical designs encode observations into a compact latent space, model transitions there with a recurrent or transformertransformer, n.: The design behind today's chatbots and much of modern AI. It handles text by letting every word look at every other word and decide which of them matter for understanding it. network, and decode rewards and sometimes images. Model-based reinforcement learningmodel-based reinforcement learning, n.: Learning in which an AI agent builds, or is given, an inner simulation of how its world responds, and uses it to plan or practice in imagination before acting. uses world models to choose actions by search or model predictive controlmodel predictive control, n.: A control method that repeatedly plans the best moves a few seconds ahead using a model of the machine, carries out only the first move, then plans again with fresh measurements., to train policies on imagined rolloutsrollout, n.: Letting an AI play out a task with its current strategy, from some starting point to the end, to see how things turn out and gather data for learning or planning. — as in Dreamer, whose third version (2023) collected diamonds in Minecraft from scratch without human data — or to predict only planning-relevant quantities, as in MuZeroMuZero, prop. n.: A DeepMind game-playing program from 2019 that mastered Go, chess, shogi and Atari games without being told the rules, by learning its own simplified model of each game to plan with.. The main weakness is compounding error: small prediction mistakes grow over long imagined horizons and can be exploited by the policy.

    We trained the policy entirely inside the world model and then transferred it to the real car.

  2. 2

    Generative media and roboticsemerging

    A large generative modelgenerative model, n.: A model that learns what a kind of data looks like — faces, sentences, songs — well enough to produce new examples that resemble the originals without copying them., usually trained on video, that generates an explorable environment frame by frame in response to user or agent actions, aiming to simulate physics, objects and their persistence. Google DeepMind's Genie (2024) learned controllable worlds from unlabeled internet video, and Genie 3 (August 2025) produced 720p worlds navigable in real time at 24 frames per second for a few minutes. 'World foundation models' are also offered for training robots and vehicles in simulation. Some labs describe video generatorstext-to-video model, n.: An AI that creates a short video clip — increasingly with matching sound — from a written description, such as 'a paper boat sailing down a rainy street'. such as SoraSora, prop. n.: OpenAI's video-generating AI, unveiled in 2024 with startlingly realistic clips; a 2025 version added sound and a social app, but OpenAI shut the service down in 2026. as world simulators, a claim critics dispute because plausible pixels need not obey consistent physics.

  3. 3

    Cognitive science and language modelscontested

    An internal representation of the structure of the world — objects, causes, other agents — that a person, animal or AI system uses to predict and reason, as opposed to relying on surface statistics. Whether large language modelslarge language model, n.: A computer program trained on enormous amounts of text to predict the words that come next. At vast scale, that simple skill lets it chat, answer questions, translate, summarize and write code. form world models in this sense is debated: probing studies such as Othello-GPTOthello-GPT, prop. n.: A small language model trained only on lists of moves from the board game Othello, which researchers found had built an internal picture of the board — evidence that such models can form world models. found board-state representations in a model trained only on move sequences, while critics argue such representations are partial and brittle. Yann LeCun, among others, argues that future AI needs world models learned from sensory data rather than text.

Origin

Kenneth Craik (1943) proposed that minds carry 'small-scale models' of external reality. In symbolic AI the STRIPS planner (Richard Fikes and Nils Nilsson, 1971) called its logical description of the robot's surroundings a world model. Jürgen Schmidhuber built learned recurrent world models for RL from 1990, and David Ha and Schmidhuber's 'World Models' (2018) popularized the term in deep learning; from 2024 it spread to interactive video generators.

Usage note

The term is used loosely: a world model may be a small dynamics network, a video generator or a hypothesized cognitive structure. Useful questions are what it predicts, conditioned on what, and how it is used; a video model that is not conditioned on actions cannot by itself support planning.

Related

See also
  • model-based reinforcement learningmodel-based reinforcement learning, n.: Learning in which an AI agent builds, or is given, an inner simulation of how its world responds, and uses it to plan or practice in imagination before acting.
  • MuZeroMuZero, prop. n.: A DeepMind game-playing program from 2019 that mastered Go, chess, shogi and Atari games without being told the rules, by learning its own simplified model of each game to plan with.
  • DynaDyna, prop. n.: An early design in which a learning agent builds a model of its world from real experience, then also practices on imagined experience from that model, squeezing more learning out of each real step.
  • generative modelgenerative model, n.: A model that learns what a kind of data looks like — faces, sentences, songs — well enough to produce new examples that resemble the originals without copying them.
  • text-to-video modeltext-to-video model, n.: An AI that creates a short video clip — increasingly with matching sound — from a written description, such as 'a paper boat sailing down a rainy street'.
  • physics simulatorphysics simulator, n.: Software that imitates how real objects move, collide, bend and fall, so robots and their AI can be designed, trained and tested on a computer before touching the real world.
  • Othello-GPTOthello-GPT, prop. n.: A small language model trained only on lists of moves from the board game Othello, which researchers found had built an internal picture of the board — evidence that such models can form world models.
  • predictive processingpredictive processing, n.: The idea that the brain is a prediction machine: it constantly guesses what its senses will report next and learns mainly from its mistakes, so what we perceive is largely its best guess.
Contrast with
  • model-free reinforcement learningmodel-free reinforcement learning, n.: Learning to act well purely from trial and error, without ever building an internal picture of how the world works — like learning a game by habit rather than by thinking moves ahead.

Sources

  1. Ha, D. & Schmidhuber, J. (2018). World Models. arXiv:1803.10122. arxiv.org/abs/1803.10122
  2. Bruce, J. et al. (2024). Genie: Generative Interactive Environments. arXiv:2402.15391. arxiv.org/abs/2402.15391
  3. Hafner, D., Pasukonis, J., Ba, J. & Lillicrap, T. (2023). Mastering Diverse Domains through World Models. arXiv:2301.04104. arxiv.org/abs/2301.04104

Mentioned in

Google DeepMindGoogle DeepMind, prop. n.: Google's main artificial intelligence laboratory, formed in 2023 by joining DeepMind and Google Brain. It builds the Gemini models and scientific AI systems such as AlphaFold., modelmodel, n.: A simplified working version of something. In AI it usually means the trained program, a big set of learned numbers, that takes an input and produces an answer, prediction or piece of content., stochastic parrotstochastic parrot, n.: A critical label for AI language systems, likening them to parrots that echo patterns in the text they were trained on, fluently but with no grasp of what the words mean., text-to-video modeltext-to-video model, n.: An AI that creates a short video clip — increasingly with matching sound — from a written description, such as 'a paper boat sailing down a rainy street'., Yann LeCunYann LeCun, prop. n.: A French-American computer scientist who pioneered the neural networks that recognize images, led Meta's AI research for over a decade, and argues that chatbots alone will not reach human-level AI., commonsense reasoningcommonsense reasoning, n.: The everyday background knowledge and reasoning people take for granted — dropped glasses break, hungry people eat, nobody is in two places at once — which machines long struggled to match., deep reinforcement learningdeep reinforcement learning, n.: Reinforcement learning in which a large neural network does the learning, letting an AI learn tasks such as video games or robot control directly from raw inputs like screen pixels or camera images., embodied AIembodied AI, n.: AI that has a body, usually a robot and sometimes a simulated character, and learns or acts by sensing and moving in a world rather than only processing text or images., fancy autocompletefancy autocomplete, n.: A dismissive description of AI chatbots as nothing more than a souped-up version of the feature on a phone keyboard that suggests your next word., Fei-Fei LiFei-Fei Li, prop. n.: A Chinese-American computer scientist whose ImageNet project, millions of hand-labeled photos, gave deep learning the proving ground that launched its boom; she now champions 'human-centered' AI., machine understandingmachine understanding, n.: Whether, and in what sense, an AI system grasps what words, situations or problems mean — rather than just producing answers that look as though it does., model predictive controlmodel predictive control, n.: A control method that repeatedly plans the best moves a few seconds ahead using a model of the machine, carries out only the first move, then plans again with fresh measurements., model-based reinforcement learningmodel-based reinforcement learning, n.: Learning in which an AI agent builds, or is given, an inner simulation of how its world responds, and uses it to plan or practice in imagination before acting., perceptionperception, n.: How an AI system or robot turns raw sensor signals, such as camera pixels, microphone sound or laser scans, into an understanding of what is around it: objects, people, words and distances., philosophy of artificial intelligencephilosophy of artificial intelligence, n.: The branch of philosophy that asks what AI is and could be: whether machines can think, understand or be conscious, what intelligence is, and what we owe to and should expect from intelligent machines., physics simulatorphysics simulator, n.: Software that imitates how real objects move, collide, bend and fall, so robots and their AI can be designed, trained and tested on a computer before touching the real world., predictive processingpredictive processing, n.: The idea that the brain is a prediction machine: it constantly guesses what its senses will report next and learns mainly from its mistakes, so what we perceive is largely its best guess., reward modelreward model, n.: A separate AI model trained to score responses the way people would, so it can stand in for human judges while another model is trained on millions of examples., scaling hypothesisscaling hypothesis, n.: The bet that today's AI methods, simply made bigger with more computing power and data, will keep getting smarter — possibly all the way to human-level intelligence and beyond — without needing fundamentally new ideas., SoraSora, prop. n.: OpenAI's video-generating AI, unveiled in 2024 with startlingly realistic clips; a 2025 version added sound and a social app, but OpenAI shut the service down in 2026., active inferenceactive inference, n.: A theory that living things, and machines built on the same idea, act in order to make the world match their predictions, choosing actions expected to keep them in preferred states and reduce their uncertainty., blurry JPEG of the webblurry JPEG of the web, phr.: Writer Ted Chiang's 2023 comparison of ChatGPT to a low-quality compressed photo of the internet: it keeps the gist of what it learned but blurs the details and fills gaps with plausible guesses., DynaDyna, prop. n.: An early design in which a learning agent builds a model of its world from real experience, then also practices on imagined experience from that model, squeezing more learning out of each real step., guaranteed safe AIguaranteed safe AI, n.: A proposed approach to AI safety in which systems come with mathematical proofs or checkable guarantees that they will not cause certain harms, instead of relying on testing alone.