"alignment tax" meaning in English

{
  "etymology_text": "First attested in a 2019 speech by computer scientist Paul Christiano (see quote), who attributed the idea to AI researcher and writer Eliezer Yudkowsky.",
  "forms": [
    {
      "form": "alignment taxes",
      "tags": [
        "plural"
      ]
    }
  ],
  "head_templates": [
    {
      "args": {},
      "expansion": "alignment tax (plural alignment taxes)",
      "name": "en-noun"
    }
  ],
  "lang": "English",
  "lang_code": "en",
  "pos": "noun",
  "senses": [
    {
      "categories": [
        {
          "kind": "other",
          "name": "English entries with incorrect language header",
          "parents": [
            "Entries with incorrect language header",
            "Entry maintenance"
          ],
          "source": "w"
        },
        {
          "kind": "topical",
          "langcode": "en",
          "name": "Artificial intelligence",
          "orig": "en:Artificial intelligence",
          "parents": [
            "Computer science",
            "Cybernetics",
            "Computing",
            "Sciences",
            "Applied mathematics",
            "Systems theory",
            "Technology",
            "All topics",
            "Mathematics",
            "Systems",
            "Fundamental",
            "Formal sciences",
            "Interdisciplinary fields",
            "Society"
          ],
          "source": "w"
        }
      ],
      "examples": [
        {
          "ref": "2019 August 29, Paul Christiano, Current work in AI alignment, EA Global San Francisco 2019",
          "text": "I like this notion of an \"alignment tax\" […] the reason I might compromise is if there's some tension, between having the AI that's robustly trying to do what I want, and having the AI that is competent or intelligent, and the alignment tax is intended to capture that gap—that cost that I incur if I insist on alignment.",
          "type": "quotation"
        },
        {
          "ref": "2021 December 1, Askell, A. et. al., “A General Language Assistant as a Laboratory for Alignment”, in arXiv, →DOI",
          "text": "The fact that larger models are less subject to forgetting may be related to the fact that larger models do not incur significant alignment taxes.",
          "type": "quotation"
        },
        {
          "ref": "2022 March 4, Ouyang, L. et. al., “Training language models to follow instructions with human feedback”, in arXiv, →DOI",
          "text": "We want an alignment procedure that avoids an alignment tax, because it incentivizes the use of models that are unaligned but more capable on these tasks.",
          "type": "quotation"
        },
        {
          "ref": "2023 February 27, Kornai, A. et. al., “Safety without alignment”, in arXiv, →DOI",
          "text": "We note that instead of an alignment tax our proposal entails a safety dividend – the more rational the system the more capable and the safer it will be.",
          "type": "quotation"
        }
      ],
      "glosses": [
        "A cost to the capabilities of an artificial intelligence resulting from the effects of aligning it with human ethics and morality."
      ],
      "id": "en-alignment_tax-en-noun-7WUfk1ms",
      "links": [
        [
          "artificial intelligence",
          "artificial intelligence"
        ],
        [
          "cost",
          "cost#Noun"
        ],
        [
          "capabilities",
          "capability#Noun"
        ],
        [
          "artificial intelligence",
          "artificial intelligence#Noun"
        ],
        [
          "aligning",
          "align#Noun"
        ],
        [
          "human",
          "human#Adjective"
        ],
        [
          "ethics",
          "ethics#Noun"
        ],
        [
          "morality",
          "morality#Noun"
        ]
      ],
      "qualifier": "artificial intelligence",
      "raw_glosses": [
        "(artificial intelligence) A cost to the capabilities of an artificial intelligence resulting from the effects of aligning it with human ethics and morality."
      ],
      "wikipedia": [
        "Eliezer Yudkowsky"
      ]
    }
  ],
  "word": "alignment tax"
}

{
  "etymology_text": "First attested in a 2019 speech by computer scientist Paul Christiano (see quote), who attributed the idea to AI researcher and writer Eliezer Yudkowsky.",
  "forms": [
    {
      "form": "alignment taxes",
      "tags": [
        "plural"
      ]
    }
  ],
  "head_templates": [
    {
      "args": {},
      "expansion": "alignment tax (plural alignment taxes)",
      "name": "en-noun"
    }
  ],
  "lang": "English",
  "lang_code": "en",
  "pos": "noun",
  "senses": [
    {
      "categories": [
        "English countable nouns",
        "English entries with incorrect language header",
        "English lemmas",
        "English multiword terms",
        "English nouns",
        "English terms with quotations",
        "en:Artificial intelligence"
      ],
      "examples": [
        {
          "ref": "2019 August 29, Paul Christiano, Current work in AI alignment, EA Global San Francisco 2019",
          "text": "I like this notion of an \"alignment tax\" […] the reason I might compromise is if there's some tension, between having the AI that's robustly trying to do what I want, and having the AI that is competent or intelligent, and the alignment tax is intended to capture that gap—that cost that I incur if I insist on alignment.",
          "type": "quotation"
        },
        {
          "ref": "2021 December 1, Askell, A. et. al., “A General Language Assistant as a Laboratory for Alignment”, in arXiv, →DOI",
          "text": "The fact that larger models are less subject to forgetting may be related to the fact that larger models do not incur significant alignment taxes.",
          "type": "quotation"
        },
        {
          "ref": "2022 March 4, Ouyang, L. et. al., “Training language models to follow instructions with human feedback”, in arXiv, →DOI",
          "text": "We want an alignment procedure that avoids an alignment tax, because it incentivizes the use of models that are unaligned but more capable on these tasks.",
          "type": "quotation"
        },
        {
          "ref": "2023 February 27, Kornai, A. et. al., “Safety without alignment”, in arXiv, →DOI",
          "text": "We note that instead of an alignment tax our proposal entails a safety dividend – the more rational the system the more capable and the safer it will be.",
          "type": "quotation"
        }
      ],
      "glosses": [
        "A cost to the capabilities of an artificial intelligence resulting from the effects of aligning it with human ethics and morality."
      ],
      "links": [
        [
          "artificial intelligence",
          "artificial intelligence"
        ],
        [
          "cost",
          "cost#Noun"
        ],
        [
          "capabilities",
          "capability#Noun"
        ],
        [
          "artificial intelligence",
          "artificial intelligence#Noun"
        ],
        [
          "aligning",
          "align#Noun"
        ],
        [
          "human",
          "human#Adjective"
        ],
        [
          "ethics",
          "ethics#Noun"
        ],
        [
          "morality",
          "morality#Noun"
        ]
      ],
      "qualifier": "artificial intelligence",
      "raw_glosses": [
        "(artificial intelligence) A cost to the capabilities of an artificial intelligence resulting from the effects of aligning it with human ethics and morality."
      ],
      "wikipedia": [
        "Eliezer Yudkowsky"
      ]
    }
  ],
  "word": "alignment tax"
}