본 게시물은 AI를 활용하여 논문 “Agents Are Not Enough”에 대한 주요 내용을 요약하고 분석한 결과입니다. 심층적인 정보는 원문 PDF를 직접 참고해 주시기 바랍니다.
📄 Original PDF: Download / View Fullscreen
영문 요약 (English Summary)
Agents Are Not Enough Chirag Shah Ryen W. White chirags@uw.edu ryenw@microsoft.com University of Washington, Seattle, WA, USA Microsoft Research, Redmond, WA USA ABSTRACT control does not generate sufficient user value. The agents may In the midst of the growing integration of Artificial Intelligence (AI) also make mistakes, require intervention or supervision, and may into various aspects of our lives, agents are experiencing a resur- be limited to performing only simple tasks. These shortcomings gence. These autonomous programs that act on behalf of humans are evident from the agentic research and development efforts over are neither new nor exclusive to the mainstream AI movement. By the past decade or so. Agents, often referred to as operators, skills, exploring past incarnations of agents, we can understand what has apps, extensions, and plugins, have been widely available through been done previously, what worked, and more importantly, what integrations into computers, smartphones, speakers, wearables,2024 did not pan out and why. This understanding lets us to examine and automoself-help agents that could tackle complex help of the agents. tasks on a user’s behalf, while ensuring privacy and trustworthiness. CCS CONCEPTS • Computing methodologies →Artificial intelligence. 2 HISTORICAL ATTEMPTS AND FAILURES[cs.AI] There are five distinct eras of agents development we can identify, KEYWORDS each differentiated by the core architecture or technology being Agens; Agentic AI used and the challenges for widespread success. Early examples, such as the General Problem Solver (GPS), aimed to In the midst of the growing integration of Artificial Intelligence (AI) also make mistakes, require intervention or supervision, and may into various aspects of our lives, agents are experiencing a resur- be limited to performing only simple tasks. These shortcomings gence. These autonomous programs that act on behalf of humans are evident from the agentic research and development efforts over are neither new nor exclusive to the mainstream AI movement. By the past decade or so. Agents, often referred to as operators, skills, apps, extensions, and plugins, have been widely available through computers, smartphones, speakers, wearables,2024 did not pan out and why. This understanding lets us to examine and automoself-help agents that could tackle complex help of the agents. tasks on a user’s behalf, while ensuring privacy and trustworthiness. CCS CONCEPTS • Computing methodologies →Artificial intelligence. 2 HISTORICAL ATTEMPTS AND FAILURES[cs.AI] There are five distinct eras of agents development we can identify, KEYWORDS each differentiated by the core architecture or technology being Agens; Agentic AI used and the challenges for widespread success. Early examples, such as the General Problem Solver (GPS), aimed to In the midst of the growing integration of Artificial Intelligence (AI) also make mistakes, require intervention or supervision, and may into various aspects of our lives, agents are experiencing a resur- be limited to performing only simple tasks. These shortcomings gence. These autonomous programs that act on behalf of humans are evident from the agentic research and development efforts over are neither new nor exclusive to the mainstream AI movement. By the past decade or so. Agents, often referred to as operators, skills, apps, extensions, and plugins, have been widely available through computers, smartphones, speakers, wearables,2024 did not pan out and why. This understanding lets us to examine and automoself-help agents that could tackle complex help of the agents. tasks on a user’s behalf, while ensuring privacy and trustworthiness. CCS CONCEPTS • Computing methodologies →Artificial intelligence. 2 HISTORICAL ATTEMPTS AND FAILURES[cs.AI] There are five distinct eras of agents development we can identify, KEYWORDS each differentiated by the core architecture or technology being Agens; Agentic AI used and the challenges for widespread success. Early examples, such as the General Problem Solver (GPS), aimed to In the midst of the growing integration of Artificial Intelligence (AI) also make mistakes, require intervention or supervision, and may into various aspects of our lives, agents are experiencing a resur- be limited to performing only simple tasks. These shortcomings gence. These autonomous programs that act on behalf of humans are evident from the agentic research and development efforts over are neither new nor exclusive to the mainstream AI movement. By the past decade or so. Agents, often referred to as operators, skills, apps, extensions, and plugins, have been widely available through computers, smartphones, speakers, wearables,2024 did not pan out and why. This understanding lets us to examine and automoself-help agents that could tackle complex help of the agents. tasks on a user’s behalf, while ensuring privacy and trustworthiness. CCS CONCEPTS • Computing methodologies →Artificial intelligence. 2 HISTORICAL ATTEMPTS AND FAILURES[cs.AI] There are five distinct eras of agents development we can identify, KEYWORDS each differentiated by the core architecture or technology being Agens; Agentic AI used and the challenges for widespread success. Early examples, such as the General Problem Solver (GPS), aimed to In the midst of the growing integration of Artificial Intelligence (AI) also make mistakes, require intervention or supervision, and may into various aspects of our lives, agents are experiencing a resur- be limited to performing only simple tasks. These shortcomings gence. These autonomous programs that act on behalf of humans are evident from the agentic research and development efforts over are neither new nor exclusive to the mainstream AI movement. By the past decade or so. Agents, often referred to as operators, skills, apps, extensions, and plugins, have been widely available through computers, smartphones, speakers, wearables,2024 did not pan out and why. This understanding lets us to examine and automoself-help agents that could tackle complex help of the agents. tasks on a user’s behalf, while ensuring privacy and trustworthiness. CCS CONCEPTS • Computing methodologies →Artificial intelligence. 2 HISTORICAL ATTEMPTS AND FAILURES[cs.AI] There are five distinct eras of agents development we can identify, KEYWORDS each differentiated by the core architecture or technology being Agens; Agentic AI used and the challenges for widespread success. Early examples, such as the General Problem Solver (GPS), aimed to In the midst of the growing integration of Artificial Intelligence (AI) also make mistakes, require intervention or supervision, and may into various aspects of our lives, agents are experiencing a resur- be limited to performing only simple tasks. These shortcomings gence. These autonomous programs that act on behalf of humans are evident from the agentic research and development efforts over are neither new nor exclusive to the mainstream AI movement. By the past decade or so. Agents, often referred to as operators, skills, apps, extensions, and plugins, have been widely available through computers, smartphones, speakers, wearables,2024 did not pan out and why. This understanding lets us to examine and automoself-help agents that could tackle complex help of the agents. tasks on a user’s behalf, while ensuring privacy and trustworthiness. CCS CONCEPTS • Computing methodologies →Artificial intelligence. 2 HISTORICAL ATTEMPTS AND FAILURES[cs.AI] There are five distinct eras of agents development we can identify, KEYWORDS each differentiated by the core architecture or technology being Agens; Agentic AI used and the challenges for widespread success. Early examples, such as the General Problem Solver (GPS), aimed to In the midst of the growing integration of Artificial Intelligence (AI) also make mistakes, require intervention or supervision, and may into various aspects of our lives, agents are experiencing a resur- be limited to performing only simple tasks. These shortcomings gence. These autonomous programs that act on behalf of humans are evident from the agentic research and development efforts over are neither new nor exclusive to the mainstream AI movement. By the past decade or so. Agents, often referred to as operators, skills, apps, extensions, and plugins, have been widely available through computers, smartphones, speakers, wearables,2024 did not pan out and why. This understanding lets us to examine and automoself-help agents that could tackle complex help of the agents. tasks on a user’s behalf, while ensuring privacy and trustworthiness. CCS CONCEPTS • Computing methodologies →Artificial intelligence. 2 HISTORICAL ATTEMPTS AND FAILURES[cs.AI] There are five distinct eras of agents development we can identify, KEYWORDS each differentiated by the core architecture or technology being Agens; Agentic AI used and the challenges for widespread success. Early examples, such as the General Problem Solver (GPS), aimed to In the midst of the growing integration of Artificial Intelligence (AI) also make mistakes, require intervention or supervision, and may into various aspects of our lives, agents are experiencing a resur- be
한글 요약 (Korean Summary)
한글 요약 생성에 실패했습니다.
주요 기술 용어 설명 (Key Technical Terms)
이 논문의 핵심 개념을 이해하는 데 도움이 될 수 있는 주요 기술 용어와 그 설명을 제공합니다. 각 용어 옆의 링크를 통해 관련 외부 자료를 검색해 보실 수 있습니다.
주요 기술 용어가 확인되지 않았거나 LLM에서 추출되지 않았습니다.
원문 발췌 및 번역 보기 (Excerpt & Translation)
원문 발췌 (English Original)
Agents Are Not Enough Chirag Shah Ryen W. White chirags@uw.edu ryenw@microsoft.com University of Washington, Seattle, WA, USA Microsoft Research, Redmond, WA USA ABSTRACT control does not generate sufficient user value. The agents may In the midst of the growing integration of Artificial Intelligence (AI) also make mistakes, require intervention or supervision, and may into various aspects of our lives, agents are experiencing a resur- be limited to performing only simple tasks. These shortcomings gence. These autonomous programs that act on behalf of humans are evident from the agentic research and development efforts over are neither new nor exclusive to the mainstream AI movement. By the past decade or so. Agents, often referred to as operators, skills, exploring past incarnations of agents, we can understand what has apps, extensions, and plugins, have been widely available through been done previously, what worked, and more importantly, what integrations into computers, smartphones, speakers, wearables,2024 did not pan out and why. This understanding lets us to examine and automobiles. However, their utility has been severely limited what distinguishes the current focus on agents. While generative [1, 6]. In addition to the limited applications, there are continuing AI is appealing, this technology alone is insufficient to make new shortcomings that these agents exhibit that are not addressed by generations of agents more successful. To make the current wave simply creating more capable systems. Here, we briefly reviewDec of agents effective and sustainable, we envision an ecosystem that why this is the case and what we can do about it. Specifically, we that while making agents more capable will address some includes not only agents but also Sims, which represent user prefer- argue19 of the issues, it will not be enough. We need to build a whole new ences and behaviors, as well as Assistants, which directly…
발췌문 번역 (Korean Translation)
에이전트는 충분하지 않아 Chirag Shah Ryen W. White chirags@uw.edu ryenw@microsoft.com 워싱턴 주 시애틀, WA, WA, WA, REDMOND, REDMOND, USA USA Abstrans 컨트롤은 충분한 사용자 가치를 생성하지 않습니다. 에이전트는 인공 지능의 통합이 증가하는 중 (AI)가 실수를 저지르고, 개입 또는 감독이 필요하며, 우리 삶의 다양한 측면으로 가질 수 있으며, 요원은 단순한 작업 만 수행하는 데 제한되어 있습니다. 이 단점은 펜스입니다. 인간을 대신하여 행동하는 이러한 자율 프로그램은 에이전트 연구 및 개발 노력으로부터 명백하다. 지난 10 년 정도. 종종 운영자, 기술, 과거의 에이전트 화신을 탐구하는 에이전트는 앱, 확장 및 플러그인이 무엇인지, 이전에 수행 한 내용을 통해 광범위하게 사용할 수 있었으며, 더 중요한 것은 컴퓨터, 스마트 폰, 스피커, 웨어러블, 2024 년에 어떤 통합이 있었는지 이해할 수 있습니다. 이러한 이해를 통해 우리는 자동차와 자동차를 조사 할 수 있습니다. 그러나 그들의 유용성은 현재 에이전트에 대한 초점을 구별하는 것을 심각하게 제한했습니다. 생성하는 동안 [1, 6]. 제한된 응용 프로그램 외에도 AI가 계속 매력적이며,이 기술 만으로도이 에이전트가 세대의 에이전트가 다루지 않은 새로운 단점을 더 성공적으로 만들기에는 불충분합니다. 현재 웨이브가 단순히 유능한 시스템을 생성하게합니다. 여기서, 우리는 효과적이고 지속 가능한 요원들의 간단한 검토, 우리는 이것이 왜 그런지에 대해 무엇을 할 수 있는지에 대한 생태계를 구상합니다. 구체적으로, 우리는 에이전트를 더 유능하게 만들 수 있지만 일부는 에이전트뿐만 아니라 SIM도 포함되며,이 문제의 사용자 선호 논쟁을 나타내는 SIM도 충분하지 않을 것입니다. 우리는 완전히 새로운 ENCE와 행동과 조수를 직접 구축해야합니다.
출처(Source): arXiv.org (또는 해당 논문의 원 출처)
답글 남기기