hypothesis От ассистента к контурной автономии
✎ редактироватьH2
статус: weakened
маркеры
- multi-step agentic workflows внутри LMS
- AI ведёт цикл diagnose → intervene → assess
- feedback loops без человеческого вмешательства
- системы с persistent learner state и долговременной памятью
история статуса
- wave-3 · proposed — сформулирована как фазовый переход: AI берёт цикл диагностика → план → выполнение → оценка
- meta-audit · weakened — Gartner: 40% agentic-проектов отменятся к 2027; широкое распространение 'agent washing'; локально подтверждается (IgniteAI Agent, Scientific Reports RCT 2025), но как общеполевой переход — нет
База эмпирики
contradicts (4)
- Australian National University · отключение Turnitin AI detector · Australian National University · AU D a1 (high)
- Deakin University · Deakin GEM (controversy) · Deakin University · AU B a2
- Gartner · прогноз отмен agentic AI проектов (контр-сигнал) · Gartner · INT D a0 (high)
- Stanford HAI · AI-detector bias against non-native English · Stanford HAI · US D a1 (high)
supports (29)
- AI-for-Science Multi-Agent Research Pipelines (обобщённый кейс) · AI-for-Science Consortium · INT B a4
- AITEE · agentic tutor for electrical engineering · AITEE research group · INT B a3
- Accenture · AI Learning Platform · Accenture · INT B a3
- Anthology · Blackboard AI Conversation (persona+reflection) · Anthology · US C a2 (low)
- Anthropic · Claude for Education (Learning mode) · Anthropic · US C a2
- CodeEdu · multi-agent coding education platform · CodeEdu research group · INT B a3
- CogEvo-Edu · hierarchical MAS with cognitive evolution · CogEvo-Edu research group · INT B a4 (high)
- Coursera · Coach (Socratic interactive instruction) · Coursera · INT C a2
- DARPA · AI-Driven Training Programs · DARPA · US F a4
- DeepMind · AI-for-Science training interfaces · Google DeepMind · UK B a4 (high)
- Duolingo · Duolingo Max (AI roleplay + explanations) · Duolingo · INT C a2 (low)
- EducationQ · multi-agent teaching evaluation framework · EducationQ research group · INT D a3
- FACET · teacher-centred MAS for personalized worksheets · FACET research group · INT B a2 (low)
- GenMentor · multi-agent goal-oriented learning framework · GenMentor research group · INT B a3 (high)
- GraphMASAL · graph-based multi-agent adaptive learning · GraphMASAL research group · INT B a3 (high)
- IBM · SkillsBuild + Watsonx · IBM · INT B a3
- Instructure · IgniteAI Agent (workflow agent) · Instructure · US B a3
- Instructure · IgniteAI suite (Canvas) · Instructure · US B a3
- IntelliCode · StateGraph orchestration for coding tutor · IntelliCode research group · INT B a3 (high)
- Khan Academy · Khanmigo (AI tutor) · Khan Academy · US C a2 (low)
- MIT · Generative Design Studios · MIT · US B a2 (low)
- Multi-Agent XR Authoring Framework · classroom authoring MAS · XR Authoring research group · INT B a3 (low)
- Old Dominion University · ITAS (Intelligent Teaching Assistant System) · Old Dominion University · US B a3 (high)
- Role-Based Feedback Agents · formative assessment MAS · Role-Based Feedback research group · INT D a3
- Scientific Reports (2025) · AI tutoring outperforms in-class active learning · Scientific Reports / academic research · INT C a3 (high)
- Tutor CoPilot · RCT-based human-AI tutor augmentation · Tutor CoPilot research group · US C a2 (high)
- UCL · AI-Governed Assessment Experiments · University College London · UK C a2 (low)
- UT Austin · UT Sage (Socratic tutorbot builder) · The University of Texas at Austin · US C a2
- University of Rochester · AI Conversations (Blackboard) · University of Rochester · US C a2 (low)
💬
Поговорить с ботом
— выбери позицию ниже · разговор поверх
hypothesis/H2