{
  "docId": "019dd923-5e88-73ef-bd5c-f6fdd8f57895",
  "docSlug": "3a892393f72c1ff5",
  "documentTitle": "2019 Air Street Capital The State of AI Report 2019",
  "authorId": "AirStreetCapital",
  "authorName": "Nathan Benaich and Ian Hogarth",
  "documentKindSlug": "consulting-deck",
  "documentKindLabel": "Consulting deck",
  "sourceTypeSlug": "vc_research",
  "sourceTypeLabel": "VC research",
  "presentationDate": null,
  "orientation": "landscape",
  "aspectRatio": 1.777,
  "pageNumber": 15,
  "pageCount": 136,
  "prevPage": 14,
  "nextPage": 16,
  "slideType": "case_study",
  "function": "illustrate_case",
  "density": "dense",
  "nDataPoints": 2,
  "notes": "The slide highlights a specific AI research methodology (Play-LMP) from the 2019 State of AI report.",
  "elementsJson": [
    "headline_text",
    "bullet_list",
    "paragraph",
    "photo"
  ],
  "metadataConfidence": 0.95,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5c-f6fdd8f57895/15",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5c-f6fdd8f57895",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5c-f6fdd8f57895.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5c-f6fdd8f57895#slide-15",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "Despite not being trained on task-specific data, this system is capable of generalizing to 18 complex user-specified manipulation tasks with average success of 85.5%, outperforming individual models trained on expert demonstrations (success of 70.3%).",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-41ab-72d9-bbc2-23b6f51b1ec8",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.45,
        "w": 0.4,
        "x": 0.58,
        "y": 0.28
      },
      "kind": "image",
      "text": "Visual comparison of Goal vs Single Play-LMP policy",
      "attrs": null,
      "subkind": "illustration",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "3c63bfb5-9bf7-41e4-bb83-0d00e2d7696a",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.65,
        "w": 0.55,
        "x": 0.02,
        "y": 0.28
      },
      "kind": "list",
      "text": "As children, we acquire complex skills and behaviors by learning and practicing diverse strategies and behaviors in a low-risk fashion, i.e. play time. Researchers used the concept of supervised play to endow robots with control skills that are more robust to perturbations compared to training using expert skill-supervised demonstrations.\nHere, a human remotely teleoperates the robot in a playground environment, interacting with all the objects available in as many ways that they can think of. A human operator provides the necessary properties of curiosity, boredom, and affordance priors to guide rich object play.\nDespite not being trained on task-specific data, this system is capable of generalizing to 18 complex user-specified manipulation tasks with average success of 85.5%, outperforming individual models trained on expert demonstrations (success of 70.3%).",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "f6dddd6f-9baf-4abb-af37-c26081badb47",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "metric",
      "text": "Success rate: 85.5%",
      "attrs": null,
      "subkind": "primary",
      "toolName": "Quantification",
      "toolSlug": "quantification",
      "confidence": null,
      "componentId": "019dd952-41ab-72d9-bbc2-25263d44e97d",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.03,
        "w": 0.95,
        "x": 0.02,
        "y": 0.22
      },
      "kind": "title",
      "text": "Training a single robot using play to perform many complex tasks without having to relearn each from scratch.",
      "attrs": null,
      "subkind": "action-title",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "2fa8cae2-fbed-4d39-9f1b-0002658d34a2",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.04,
        "w": 0.45,
        "x": 0.02,
        "y": 0.14
      },
      "kind": "title",
      "text": "What's next in RL: Play-driven learning for robots",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "96ae418e-293d-43e2-8718-0a10c0bf05af",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [],
  "frameworks": [],
  "arcBeats": [
    {
      "to": 92,
      "from": 5,
      "beatId": "019dd95a-0682-776c-8e34-ce57e5bbe574",
      "arcName": "The Triple Take",
      "arcSlug": "triple-take",
      "beatName": "The Facts (What)",
      "beatSlug": "triple-take-the-facts-what",
      "evidence": "Definitions through Research/Talent/Industry — fact-dense case studies and benchmarks.",
      "position": 1,
      "confidence": 70,
      "parentBeatName": "Setup",
      "parentBeatSlug": "setup"
    },
    {
      "to": 92,
      "from": 9,
      "beatId": "019dd95a-0682-776c-8e34-ddbccb391af1",
      "arcName": "Voyage and Return",
      "arcSlug": "voyage-return",
      "beatName": "The Unknown",
      "beatSlug": "voyage-return-the-unknown",
      "evidence": "Research/Talent/Industry frontiers explored.",
      "position": 2,
      "confidence": 45,
      "parentBeatName": "Complication",
      "parentBeatSlug": "complication"
    }
  ],
  "loops": [
    {
      "to": 15,
      "from": 10,
      "name": "Pattern Hunter",
      "slug": "02-pattern-hunter",
      "bestFor": "Time-pressed audiences, building consensus, when data is strong",
      "matchId": "019dd95a-07fe-70ce-8d3c-9ac033d2887b",
      "evidence": "Six case studies (Montezuma, StarCraft II, CTF, Dota 99.4%, robot dexterity) all titled 'RL conquers new territory'.",
      "position": 2,
      "objective": "Show RL is conquering new domains via stacked breakthrough examples",
      "structure": "Evidence A -> Evidence B -> Evidence C -> Pattern/Conclusion",
      "confidence": 80,
      "description": "Group multiple pieces of evidence that together point to a pattern or conclusion"
    }
  ],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}