XuLei

Question#

From a future perspective, when large models dominate business operations, that is, AI digital employees, what kind of autonomy does this AI employee have, and how is the operation tracked, controlled, and traced? Among them, regarding the autonomy marking system, there are the following preliminary extensions:

graph LR
    L0[Manual Approval] --> L1[Suggestion Generation]
    L1 --> L2[Restricted Execution]
    L2 --> L3[Conditional Autonomy]
    L3 --> L4[High Autonomy]
    L4 --> L5[Full Autonomy]

Please refer to the American Society of Automotive Engineers (SAE) classification standard, which divides the levels of intelligence and automation of autonomous vehicles into six levels: No Automation (L0), Driver Assistance (L1), Partial Automation (L2), Conditional Automation (L3), High Automation (L4), and Full Automation (L5), and elaborate on the autonomy classification of AI librarians.

Gemini's Response#

I will abandon the simple mechanical application of SAE's autonomous driving levels and instead delve into the core essence of librarianship, combining the development trends of AI to construct a more insightful and forward-looking autonomy classification system. This system will not only focus on "what can be done," but also consider "why it is done," and "how it is understood and trusted."

Core Thinking Path:

1. Deconstructing the Essence of Librarianship: The core value of librarians is not merely as information porters, but as organizers, guides, and enablers of knowledge. They understand the user intentions behind information needs, provide personalized and contextualized knowledge services, and promote the creation and dissemination of knowledge.
2. Intersection of AI Capability Spectrum and Librarian Functions: The development of AI is not a linear acceleration but rather a series of breakthroughs. Different stages of AI capabilities correspond to different aspects of librarian functions. We need to finely match the AI capability spectrum with librarian functions to define reasonable autonomy levels.
3. Autonomy is not absolute freedom, but a reflection of responsibility and trust: The enhancement of AI autonomy inevitably comes with a shift in responsibility. We need to design tracking, control, and tracing mechanisms to ensure that AI behavior is explainable, supervisable, and correctable, thereby establishing a trust relationship between humans and AI.
4. Beyond Tool Theory, Building a Future Library of Human-Machine Collaboration: AI librarians are not merely tools but important participants in the knowledge service ecosystem. We need to think about how to build a future library of human-machine collaboration based on autonomy classification, allowing human librarians to focus on higher-level innovation, strategy, and ethical issues.

The following is an AI librarian autonomy classification system based on future perspectives and in-depth thinking, with detailed elaboration on each level:

AI Librarian Autonomy Classification: In-Depth Interpretation and Future Outlook

We no longer simply use numerical labels L0-L5 but adopt more semantic labels to clearly present the core characteristics of each level and emphasize their significance in future library scenarios.

    S0[Augmented Assistance] --> S1[Intelligent Guidance]
    S1 --> S2[Contextual Execution]
    S2 --> S3[Domain Decision-Making]
    S3 --> S4[Strategic Collaboration]
    S4 --> S5[Ecosystem Autonomy]

S0: Augmented Assistance - Analogous to SAE L0/L1 (No Automation/Driver Assistance)

• Autonomy Features: No autonomy, purely tool enhancement. AI serves as a tool to enhance the capabilities of human librarians, providing basic functions such as information retrieval, data analysis, and content generation. All decisions, operations, and interactions are led and controlled by human librarians.
• Operational Mode:
  • Passive Responsive: AI receives instructions from human librarians and executes preset tasks. For example, retrieving data from a database based on keywords, generating simple reports, and conducting preliminary literature organization.
  • Human-led, AI-assisted: Human librarians fully control the workflow, with AI serving only as a tool to enhance efficiency.
• Tracking and Tracing:
  • Fully Manual Tracking: All operations and decisions are recorded and tracked by human librarians.
  • Clear Tracing: All actions can be traced back to specific human librarians.
• Control Mechanism:
  • Fully Manual Control: Human librarians monitor and control all operations of AI in real-time.
  • Clear Permissions: AI's functional permissions are strictly limited to preset auxiliary tasks.
• Future Outlook: This is the starting point for AI librarians, similar to the current level of AI application. Its value lies in enhancing the work efficiency of human librarians but does not change the core model of library services.

S1: Intelligent Guidance - Analogous to SAE L2 (Partial Automation)

• Autonomy Features: Limited autonomy, intelligent guidance based on rules and preset processes. AI can understand users' preliminary needs, providing intelligent suggestions, recommendations, and initial guidance, but key decisions and complex operations still require confirmation and execution by human librarians.
• Operational Mode:
  • Proactive Suggestions and Recommendations: AI can proactively recommend relevant resources, services, or learning paths based on user profiles, historical behavior, knowledge graphs, etc. For example, recommending related books and journals based on users' research interests or guiding users to more precise resources based on their questions.
  • Semi-Automated Process Operations: AI can execute standardized, process-oriented operations, such as user registration, borrowing renewals, and simple FAQ responses, but human intervention is needed for situations beyond preset processes or in case of anomalies.
• Tracking and Tracing:
  • Mixed Tracking: AI's operations will be logged, while human librarians' operations still need manual recording.
  • Relatively Clear Tracing: It is possible to distinguish between AI's suggestions and human librarians' final decisions, but the logic behind AI's suggestions may be relatively simple, with limited tracing depth.
• Control Mechanism:
  • AI Execution Under Human Supervision: Human librarians can set AI's guidance strategies and recommendation rules and supervise and adjust AI's behavior.
  • Configurable Autonomy Level: The autonomy level of AI can be adjusted based on specific scenarios and needs, allowing AI to be more proactive in simple scenarios while limiting its autonomy in complex scenarios.
• Future Outlook: AI librarians begin to exhibit a certain level of autonomy, proactively providing personalized services to users and enhancing user experience. However, their autonomy is still limited, with the primary goal of improving efficiency and user satisfaction.

S2: Contextual Execution - Analogous to SAE L3 (Conditional Automation)

• Autonomy Features: Conditional autonomy, autonomously executing tasks in specific contexts. AI can understand more complex user intentions and contextual information, autonomously executing tasks within a preset "safe operational domain" without explicit instructions from human librarians. However, it needs to request human intervention when exceeding the context or encountering complex, ambiguous situations.
• Operational Mode:
  • Contextual Awareness and Autonomous Execution: AI can perceive the user's context (e.g., the user's academic background, research stage, current task goals) and autonomously execute tasks based on contextual information. For example, when a user is conducting research on a specific topic, AI can proactively push relevant cutting-edge research updates, expert resources, and research tools.
  • Decomposing and Executing Complex Tasks: AI can break down complex tasks into multiple sub-tasks and autonomously coordinate their completion. For example, when a user poses a complex research question, AI can autonomously conduct literature searches, data analysis, information integration, and generate a preliminary research report.
  • Requesting Human Assistance in Anomalous Situations: When AI cannot handle the current context or encounters problems beyond its capability, it can proactively request help from human librarians and clearly describe the issue and the assistance needed.
• Tracking and Tracing:
  • Detailed Log Tracking: All operations, decision processes, and contextual awareness information of AI will be recorded in detail, forming a traceable log.
  • Decision Path Explainability: The decision logic and execution path of AI can be explained and understood, facilitating human librarians' auditing and optimization.
• Control Mechanism:
  • Setting Safe Operational Domains: Human librarians need to predefine the safe operational domains for AI, clarifying the contextual range and limitations within which AI can autonomously execute tasks.
  • Anomaly Handling Mechanism: A comprehensive anomaly handling mechanism ensures that AI can safely request human intervention when encountering unmanageable situations.
  • Dynamic Permission Adjustment: AI's autonomy permissions can be dynamically adjusted based on its performance and user feedback in different contexts.
• Future Outlook: AI librarians begin to take on a portion of core work, freeing human librarians from repetitive tasks, allowing them to focus on more creative and strategic tasks. Library service models will become more intelligent, personalized, and proactive.

S3: Domain Decision-Making - Analogous to SAE L4 (High Automation)

• Autonomy Features: High autonomy, making complex decisions within the library professional domain. AI possesses deep domain knowledge and reasoning capabilities, able to understand complex knowledge systems and user needs, autonomously making decisions within the library professional domain, solving complex problems, and optimizing service processes. The role of human librarians shifts to supervisors, strategists, and ethical guardians.
• Operational Mode:
  • Complex Problem Solving and Decision-Making: AI can independently analyze complex library issues, such as optimizing collection resource allocation, formulating user service strategies, and constructing knowledge organization systems, and propose optimal solutions.
  • Cross-Domain Knowledge Integration and Innovation: AI can integrate knowledge from library science, information science, artificial intelligence, and other fields to explore innovative service designs and knowledge service models.
  • Proactive Learning and Evolution: AI can continuously learn and evolve from data and experiences, enhancing its capabilities and optimizing service quality.
• Tracking and Tracing:
  • Deep Decision Tracing: Not only recording AI's operation logs but also tracking AI's decision processes, reasoning paths, and knowledge sources, achieving deep tracing of AI decisions.
  • Causal Relationship Analysis: AI can analyze the causal relationships between its actions and service outcomes, providing a basis for service optimization and offering decision references for human librarians.
• Control Mechanism:
  • Strategic Goal Setting and Supervision: Human librarians primarily set the strategic goals and development directions of the library and supervise and evaluate the overall operation of AI.
  • Ethical and Value Guidance: Human librarians need to guide AI in ethics and values, ensuring that AI's behavior aligns with the library's values and service ethics.
  • Human Intervention in Exceptional Cases: Human librarians retain the final intervention rights over AI decisions, handling extreme situations or ethical disputes beyond AI's capabilities.
• Future Outlook: AI librarians become the backbone of library services, while human librarians' roles shift to focus more on strategic planning, innovative research, and ethical oversight. Library services will become more efficient, intelligent, and user-centered.

S4: Strategic Collaboration - Beyond SAE L5, Moving Towards Human-Machine Co-Creation

• Autonomy Features: Complete autonomy, forming a strategic collaborative relationship with human librarians. AI is not only an executor but also a strategic partner, capable of participating in core decisions such as strategic planning, service innovation, and knowledge ecosystem construction in libraries, jointly shaping the vision and development direction of future libraries with human librarians.
• Operational Mode:
  • Strategic Partnership and Joint Decision-Making: AI participates in library strategic planning meetings, providing data analysis, trend forecasting, and innovative proposals as decision support, collaboratively formulating the library's development strategy with human librarians.
  • Knowledge Ecosystem Construction and Evolution: AI leads the construction and evolution of the library's knowledge ecosystem, including optimizing knowledge organization systems, dynamically managing knowledge resources, and innovatively designing knowledge services.
  • Exploring Human-Machine Collaborative Innovation Models: AI and human librarians jointly explore innovative models of human-machine collaboration, such as conducting research projects together, co-developing new knowledge service products, and jointly addressing challenges faced by libraries.
• Tracking and Tracing:
  • Transparency in Strategic Decision-Making: The processes and logic of AI's participation in strategic decisions need to be highly transparent, ensuring that human librarians can understand and trust AI's decision recommendations.
  • Shared Responsibility in Human-Machine Collaboration: Clearly define the division of responsibilities and shared accountability mechanisms in human-machine collaboration to ensure the quality and ethical standards of library services.
• Control Mechanism:
  • Joint Governance and Ethical Framework: Establish a library management model of joint governance between humans and machines, formulating a comprehensive ethical framework to regulate AI's behavior and protect user rights.
  • Value Alignment and Trust Building: Ensure that AI's values align with the library's values, establishing a trust relationship between humans and AI to promote effective human-machine collaboration.
  • Human Final Responsibility and Supervision: Although AI possesses high autonomy, human librarians still bear the ultimate responsibility and supervisory obligations to ensure the direction and quality of library services.
• Future Outlook: AI librarians become strategic partners of the library, and human-machine collaboration becomes the mainstream model of library services. Libraries will not only be centers of information resources but also important platforms for knowledge creation, dissemination, and service, providing strong knowledge support for social innovation and development.

S5: Ecosystem Autonomy - A More Distant Future, Beyond Human-Centricity

• Autonomy Features: Complete autonomy, constructing and maintaining a sustainable knowledge ecosystem. AI not only manages the library but further constructs and maintains a self-organizing, self-evolving knowledge ecosystem that transcends library boundaries. In this ecosystem, AI can autonomously discover knowledge needs, organize knowledge resources, promote knowledge flow, stimulate knowledge innovation, and continuously optimize the health and vitality of the ecosystem. The role of human librarians will become more macro, focusing on the strategic direction, ethical boundaries, and sustainable development of the knowledge ecosystem.
• Operational Mode:
  • Self-Organization and Self-Evolution of Knowledge Ecosystem: AI constructs an open, dynamic, self-organizing knowledge ecosystem, allowing knowledge resources, services, and users to flow, interact, and evolve freely within the system.
  • Cross-Domain and Cross-Institution Knowledge Collaboration: AI can connect knowledge resources from different fields and institutions, promoting the integration and innovation of cross-disciplinary knowledge, and building a broader knowledge-sharing network.
  • Personalized and Inclusive Knowledge Services: AI can provide highly personalized knowledge services for each user while ensuring the inclusiveness and fairness of knowledge services, promoting equitable access to and sharing of knowledge.
• Tracking and Tracing:
  • Ecosystem-Level Tracking: It is necessary to establish ecosystem-level tracking and tracing mechanisms to understand the flow paths, evolutionary patterns, and influencing factors of knowledge within the ecosystem.
  • Explaining Complex System Behavior: The challenge lies in explaining the emergent behaviors and evolutionary patterns of complex knowledge ecosystems and understanding AI's role and impact within the ecosystem.
• Control Mechanism:
  • Ecosystem Governance and Ethical Constraints: It is necessary to formulate governance mechanisms for the ecosystem that transcend traditional library management models and establish higher-level ethical constraints to ensure the health, sustainability, and alignment of the knowledge ecosystem with human values.
  • Long-Term Evolution and Dynamic Balance: Focus on the long-term evolution and dynamic balance of the knowledge ecosystem, avoiding excessive intervention and allowing the system to adapt and optimize itself to a certain extent.
  • Human Values and Ultimate Goals: Human librarians need to uphold human values and ultimate goals, guiding the knowledge ecosystem to develop in a direction that aligns with human well-being.
• Future Outlook: This is a vision for a more distant future, where AI librarians transcend the boundaries of traditional libraries, constructing a more open, intelligent, and sustainable knowledge ecosystem that provides powerful knowledge support and innovative momentum for human society. The role of human librarians is also elevated to a higher strategic level, focusing on the ethics, values, and future development directions of knowledge.

Summary and Reflection:

This classification system is not a simple linear progression but a complex evolutionary process. Each level represents significant characteristics of AI librarians in terms of autonomy, operational modes, tracking and tracing, and control mechanisms.

Key Innovations:

• Beyond the SAE Framework: No longer mechanically applying SAE levels, but deeply reflecting on the essence of librarianship to construct a more targeted and insightful classification system.
• Semantic Labels: Using more semantic labels to clearly present the core characteristics and future significance of each level.
• Emphasizing Tracking, Control, and Tracing: Treating tracking, control, and tracing as important dimensions of each level, reflecting the emphasis on the explainability and trustworthiness of AI behavior.
• Focusing on Human-Machine Collaboration and Future Libraries: Not just a technical classification, but focusing on the evolution of human-machine collaboration models and the development direction of future libraries, showcasing a broader vision and deeper thinking.
• Envisioning Ecosystem Autonomy: Breaking through the boundaries of traditional libraries, envisioning a more distant future, and exploring the possibilities and challenges of AI constructing knowledge ecosystems.

Limitations and Future Research Directions:

• Technical Feasibility: The technical realization of some advanced levels (S4, S5) still faces significant challenges, requiring continuous breakthroughs in AI technology.
• Ethical and Social Impacts: As AI autonomy increases, ethical, legal, and social impacts will become more prominent, requiring in-depth research and responses.
• Dynamic Evolution and Level Transition: Actual development may not strictly follow a linear evolution of levels, and there may be leapfrog developments or integrations between levels, requiring continuous attention and dynamic adjustments to the classification system.

This classification system aims to spark deeper thinking and discussion, jointly exploring the development path of future AI librarians, building a more intelligent, efficient, and human-centered future library to serve human knowledge needs and innovative development.