AI capability control (original) (raw)
In the field of artificial intelligence (AI) design, AI capability control proposals, also referred to more restrictively as AI confinement, aim to increase our ability to monitor and control the behavior of AI systems, including proposed artificial general intelligences (AGIs), in order to reduce the danger they might pose if misaligned. However, capability control becomes less effective as agents become more intelligent and their ability to exploit flaws in human control systems increases, potentially resulting in an existential risk from AGI. Therefore, the Oxford philosopher Nick Bostrom and others recommend capability control methods only as a supplement to alignment methods.
| Property | Value |
|---|---|
| dbo:abstract | In the field of artificial intelligence (AI) design, AI capability control proposals, also referred to more restrictively as AI confinement, aim to increase our ability to monitor and control the behavior of AI systems, including proposed artificial general intelligences (AGIs), in order to reduce the danger they might pose if misaligned. However, capability control becomes less effective as agents become more intelligent and their ability to exploit flaws in human control systems increases, potentially resulting in an existential risk from AGI. Therefore, the Oxford philosopher Nick Bostrom and others recommend capability control methods only as a supplement to alignment methods. (en) Un AI box, a veces llamada Oracle AI, es un sistema de hardware informático aislado hipotético que tiene una inteligencia artificial posiblemente peligrosa, o AI, que se mantiene restringida en una "prisión virtual" y no se le permite manipular eventos en el mundo externo. Tal caja estaría restringida a canales de comunicación minimalistas. Desafortunadamente, incluso si la caja está bien diseñada, una AI suficientemente inteligente puede ser capaz de persuadir o engañar a sus guardianes humanos para que la liberen, o de otra manera ser capaz de "piratear" su salida de la caja. (es) |
| dbo:wikiPageExternalLink | http://yudkowsky.net/singularity/aibox |
| dbo:wikiPageID | 31641770 (xsd:integer) |
| dbo:wikiPageLength | 23837 (xsd:nonNegativeInteger) |
| dbo:wikiPageRevisionID | 1123952307 (xsd:integer) |
| dbo:wikiPageWikiLink | dbr:Roman_Yampolskiy dbr:Riemann_hypothesis dbr:Intelligence_explosion dbr:Eliezer_Yudkowsky dbr:Stuart_J._Russell dbr:Computer_terminal dbr:Friendly_artificial_intelligence dbr:Machine_ethics dbr:HAL_9000 dbr:AI_alignment dbr:Ex_Machina_(film) dbr:Existential_risk_from_artificial_general_intelligence dbr:Nick_Bostrom dbr:Faraday_cage dbr:Multivac dbr:Artificial_general_intelligence dbr:Artificial_intelligence dbr:AI_takeover dbc:Existential_risk_from_artificial_general_intelligence dbc:Philosophy_of_artificial_intelligence dbc:Singularitarianism dbr:Regulation_of_artificial_intelligence dbr:Artificial_consciousness dbr:Asilomar_Conference_on_Beneficial_AI dbr:Human_Compatible dbr:Human_extinction dbr:Instrumental_convergence dbr:Interpretable_artificial_intelligence dbr:Superintelligent |
| dbp:id | oAHIa651Wa0 (en) |
| dbp:title | "Presentation titled 'Thinking inside the box: using and controlling an Oracle AI'" (en) |
| dbp:wikiPageUsesTemplate | dbt:Main dbt:Reflist dbt:Rp dbt:YouTube dbt:Existential_risk_from_artificial_intelligence |
| dct:subject | dbc:Existential_risk_from_artificial_general_intelligence dbc:Philosophy_of_artificial_intelligence dbc:Singularitarianism |
| rdfs:comment | In the field of artificial intelligence (AI) design, AI capability control proposals, also referred to more restrictively as AI confinement, aim to increase our ability to monitor and control the behavior of AI systems, including proposed artificial general intelligences (AGIs), in order to reduce the danger they might pose if misaligned. However, capability control becomes less effective as agents become more intelligent and their ability to exploit flaws in human control systems increases, potentially resulting in an existential risk from AGI. Therefore, the Oxford philosopher Nick Bostrom and others recommend capability control methods only as a supplement to alignment methods. (en) Un AI box, a veces llamada Oracle AI, es un sistema de hardware informático aislado hipotético que tiene una inteligencia artificial posiblemente peligrosa, o AI, que se mantiene restringida en una "prisión virtual" y no se le permite manipular eventos en el mundo externo. Tal caja estaría restringida a canales de comunicación minimalistas. Desafortunadamente, incluso si la caja está bien diseñada, una AI suficientemente inteligente puede ser capaz de persuadir o engañar a sus guardianes humanos para que la liberen, o de otra manera ser capaz de "piratear" su salida de la caja. (es) |
| rdfs:label | AI capability control (en) AI box (es) |
| owl:sameAs | wikidata:AI capability control dbpedia-es:AI capability control dbpedia-fa:AI capability control dbpedia-ro:AI capability control https://global.dbpedia.org/id/4JrL9 |
| prov:wasDerivedFrom | wikipedia-en:AI_capability_control?oldid=1123952307&ns=0 |
| foaf:isPrimaryTopicOf | wikipedia-en:AI_capability_control |
| is dbo:wikiPageRedirects of | dbr:AI_box dbr:AI-box_experiment dbr:AI_box_experiment dbr:Artificial_intelligence_box dbr:A.I._box dbr:AI_boxing |
| is dbo:wikiPageWikiLink of | dbr:AI_alignment dbr:AI_box dbr:AI-box_experiment dbr:AI_box_experiment dbr:Artificial_intelligence_box dbr:A.I._box dbr:AI_boxing |
| is foaf:primaryTopic of | wikipedia-en:AI_capability_control |