{
  "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": 10,
  "pageCount": 136,
  "prevPage": 9,
  "nextPage": 11,
  "slideType": "case_study",
  "function": "illustrate_case",
  "density": "dense",
  "nDataPoints": 15,
  "notes": "The chart shows the progression of various AI models' scores over time, highlighting the jump in performance with RND.",
  "elementsJson": [
    "headline_text",
    "paragraph",
    "line_chart"
  ],
  "metadataConfidence": 1,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5c-f6fdd8f57895/10",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5c-f6fdd8f57895",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5c-f6fdd8f57895.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5c-f6fdd8f57895#slide-10",
  "components": [
    {
      "bbox": {
        "h": 0.04,
        "w": 0.8,
        "x": 0.03,
        "y": 0.22
      },
      "kind": "callout",
      "text": "Rewarding 'curiosity' enables OpenAI to achieve superhuman performance at Montezuma's Revenge.",
      "attrs": null,
      "subkind": "primary",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "1d80b9a1-4c42-4ea2-b0eb-29a5aceb5ab6",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "callout",
      "text": "Rewarding ‘curiosity’ enables OpenAI to achieve superhuman performance at Montezuma’s Revenge.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-41ab-72d9-bbc1-9e5590951295",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.55,
        "w": 0.45,
        "x": 0.5,
        "y": 0.3
      },
      "kind": "chart",
      "text": "Progress in Montezuma's Revenge",
      "attrs": null,
      "subkind": "line",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "da99d33f-0f44-42ab-9f47-3a157d715159",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "metric",
      "text": "Game Score",
      "attrs": null,
      "subkind": "primary",
      "toolName": "Quantification",
      "toolSlug": "quantification",
      "confidence": null,
      "componentId": "019dd952-41ab-72d9-bbc1-a15d9275bb3e",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.6,
        "w": 0.45,
        "x": 0.03,
        "y": 0.28
      },
      "kind": "paragraph",
      "text": "In 2015, DeepMind's DQN system successfully achieved superhuman performance on a large number of Atari 2600 games. A major hold out was Montezuma's Revenge. In October 2018, OpenAI achieved superhuman performance at Montezuma's with a technique called Random Network Distillation (RND), which incentivised the RL agent to explore unpredictable states. This simple but powerful modification can be particularly effective in environments where broader exploration is valuable. The graph on the right shows total game score achieved by different AI systems on Montezuma's Revenge.",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "19f77ded-716f-407b-86a2-cf580a6b3a15",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.05,
        "w": 0.6,
        "x": 0.03,
        "y": 0.14
      },
      "kind": "title",
      "text": "Reinforcement learning (RL) conquers new territory: Montezuma's Revenge",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "1dd2a1b6-2746-4669-ba58-7ee00614f98b",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Inductive Reasoning",
      "slug": "inductive-reasoning",
      "agent": "Architect",
      "layer": "loop",
      "matchId": "019dd95a-0fd5-7148-8ecc-9cdd75445738",
      "evidence": "First of multiple RL wins building inductively to 'RL conquers' pattern.",
      "confidence": 80
    },
    {
      "name": "Action Titles",
      "slug": "action-titles",
      "agent": "Architect",
      "layer": "slide",
      "matchId": "019dd95a-0fd5-7148-8ecc-a21d6d27bd8b",
      "evidence": "Title states the insight: 'RL conquers new territory: Montezuma's Revenge'.",
      "confidence": 85
    },
    {
      "name": "Concrete Language",
      "slug": "concrete-language",
      "agent": "Storyteller",
      "layer": "slide",
      "matchId": "019dd95a-0fd5-7148-8ecc-a401748f336a",
      "evidence": "Names specific game and 'curiosity' reward.",
      "confidence": 70
    }
  ],
  "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
}