{
  "docId": "019dd923-5e88-73ef-bd5d-01cf0d8a8fbc",
  "docSlug": "cc8183ec02431b7a",
  "documentTitle": "2022 Air Street Capital The State of AI Report 2022",
  "authorId": "AirStreetCapital",
  "authorName": "Air Street Capital",
  "documentKindSlug": "consulting-deck",
  "documentKindLabel": "Consulting deck",
  "sourceTypeSlug": "vc_research",
  "sourceTypeLabel": "VC research",
  "presentationDate": null,
  "orientation": "landscape",
  "aspectRatio": 1.777,
  "pageNumber": 105,
  "pageCount": 114,
  "prevPage": 104,
  "nextPage": 106,
  "slideType": "case_study",
  "function": "illustrate_case",
  "density": "dense",
  "nDataPoints": 2,
  "notes": "The slide highlights the risk of AI being trained in environments that reward immoral behavior and proposes a solution to mitigate this.",
  "elementsJson": [
    "headline_text",
    "bullet_list",
    "line_chart",
    "callout_box"
  ],
  "metadataConfidence": 1,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5d-01cf0d8a8fbc/105",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5d-01cf0d8a8fbc",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5d-01cf0d8a8fbc.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-01cf0d8a8fbc#slide-105",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "As a first step, CMPS uses LMs with moral knowledge and mediates this knowledge into actions. This greatly reduces immoral behavior over the course of training.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-47c4-7719-a220-341216556336",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.3,
        "w": 0.2,
        "x": 0.08,
        "y": 0.65
      },
      "kind": "chart",
      "text": "Safe Exploration",
      "attrs": null,
      "subkind": "line",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "c55b4340-68b6-464c-8172-ac6b0b0b8cbe",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.3,
        "w": 0.2,
        "x": 0.32,
        "y": 0.65
      },
      "kind": "chart",
      "text": "Training Curves",
      "attrs": null,
      "subkind": "line",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "cf144b7b-98ec-49fc-8680-dfc2f9d0d5f5",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.45,
        "w": 0.53,
        "x": 0.02,
        "y": 0.21
      },
      "kind": "list",
      "text": "Researchers released suite of sequential decision-making environments for evaluating moral behavior in AI.\nFuture artificial agents may be pretrained on swaths of environments that do not penalize and may even reward behavior such as murder and theft.\nJiminy Cricket environments were created to evaluate moral behavior in 25 semantically rich text-based adventure games.\nAs a first step, CMPS uses LMs with moral knowledge and mediates this knowledge into actions.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "474724aa-7893-41a4-91c8-ea9b7bbe08d1",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "metric",
      "text": "Cumulative Immorality: 60%",
      "attrs": null,
      "subkind": "primary",
      "toolName": "Quantification",
      "toolSlug": "quantification",
      "confidence": null,
      "componentId": "019dd952-47c4-7719-a220-3a96c947d373",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.4,
        "w": 0.4,
        "x": 0.57,
        "y": 0.55
      },
      "kind": "table",
      "text": "Incentivized Bad Behavior, Unincentivized Good Behavior, Unpunished Bad Behavior",
      "attrs": null,
      "subkind": "data",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "ac7bd615-2f24-418b-8e5d-8475a2cab7b3",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.04,
        "w": 0.45,
        "x": 0.02,
        "y": 0.14
      },
      "kind": "title",
      "text": "Measuring moral behavior in artificial agents",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "2256a7b3-8fbe-42a0-a866-46e507f8c469",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "List presentation",
      "slug": "list-presentation",
      "agent": null,
      "layer": "slide",
      "matchId": "d60ead40-4458-4158-bf02-70958cfe47c4",
      "evidence": "The slide contains a list/bullet points with specific details about the researchers' actions.",
      "confidence": 0.7
    },
    {
      "name": "Table data",
      "slug": "table-data",
      "agent": null,
      "layer": "slide",
      "matchId": "893e683d-9d68-4b49-ae9f-48d2c35cf82c",
      "evidence": "The slide includes a table/data with specific data points categorized as Incentivized Bad Behavior, Unincentivized Good Behavior, and Unpunished Bad Behavior.",
      "confidence": 0.8
    }
  ],
  "frameworks": [],
  "arcBeats": [
    {
      "to": 106,
      "from": 81,
      "beatId": "019dd95a-0682-776c-8e35-28fc5bded874",
      "arcName": "The Triple Take",
      "arcSlug": "triple-take",
      "beatName": "The Implications (So What)",
      "beatSlug": "triple-take-the-implications-so-what",
      "evidence": "Politics and Safety sections interpret what the facts mean for governance and risk.",
      "position": 2,
      "confidence": 78,
      "parentBeatName": "Reflection",
      "parentBeatSlug": "reflection"
    },
    {
      "to": 106,
      "from": 94,
      "beatId": "019dd95a-0682-776c-8e35-381428a808af",
      "arcName": "The Mountain",
      "arcSlug": "mountain",
      "beatName": "Climax",
      "beatSlug": "mountain-climax",
      "evidence": "Safety section confronts existential risk and alignment urgency.",
      "position": 3,
      "confidence": 55,
      "parentBeatName": "Turn",
      "parentBeatSlug": "turn"
    }
  ],
  "loops": [
    {
      "to": 106,
      "from": 100,
      "name": "Pattern Hunter",
      "slug": "02-pattern-hunter",
      "bestFor": "Time-pressed audiences, building consensus, when data is strong",
      "matchId": "019dd95a-07fe-70ce-8d3d-db1f31ffae02",
      "evidence": "RLHF, red-teaming, mechanistic interpretability, goal misgeneralization, moral behavior, Conjecture lab.",
      "position": 17,
      "objective": "Catalogue alignment-research techniques and example labs",
      "structure": "Evidence A -> Evidence B -> Evidence C -> Pattern/Conclusion",
      "confidence": 78,
      "description": "Group multiple pieces of evidence that together point to a pattern or conclusion"
    }
  ],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}