Reading objectives and introduction
This is an English translation of the published reading. We will compare intelligence and agency, examine different views on how humans act, form judgments about AI agency, and consider responsibility within a system composed of people, institutions, and technology.
"Intelligent" Often involves abilities related to knowledge, learning, analysis, and understanding concepts and their relationships; Agency More focused on taking action: can set goals, consider options, plan, and execute, and produce impact in uncertain environments. A system being able to act does not automatically imply it possesses human-like understanding, consciousness, or moral responsibility.
I. The intellectual lineage of human agency
Human agency refers to the capacity for independent action, choice, and shaping of one's life, rather than being entirely determined by biological factors, social structures, or fate. It is linked to autonomy, responsibility, and freedom, and acknowledges that humans are active participants in changing the world.
Classical discussions link moral agents to the ability to discern right from wrong, rational reflection, and intentional action; modern discussions further emphasize autonomy: acting according to principles that one can rationally justify, rather than merely following desires or instincts.
Modern perspectives also consider how power, institutions, and cultural norms influence choices. Structure is not a fixed background existing independently of people; human actions also shape structure, while structure in turn influences actions. Thus, agency is both an individual capacity and an interaction of freedom, reason, and social conditions.
Two: Why does AI make this issue more complicated?
After enhanced capabilities in response, decision-making, and creation, generative systems are increasingly referred to as Agents in commercial contexts. Different materials define "agency" in various ways, with some emphasizing proactive initiation and others highlighting sustained action toward a goal.
There is no entirely consistent usage of "intelligence" or "consciousness." The source material contrasts one view that regards AI as intelligent but distinct from consciousness, with another that critically examines the terms "artificial" and "intelligent" from social and political perspectives. When reading, it is essential to clarify the concepts involved, as differences in wording do not necessarily imply opposing factual judgments.
Another perspective argues that humans have consistently delegated cognitive tasks to tools. AI can enhance cognitive capabilities, but it functions more like human interaction and agency, challenging the notion of human centrality in action. Understanding its impact requires studying how action is distributed among humans, tools, and the environment.
Technology not only complements existing activities but may also alter social relationships and our ways of understanding reality. Therefore, research should not merely ask "Is AI like a human?" but should also inquire how it reorganizes human action.
Original reading activity specifies the article "AI as Agency without Intelligence," discussing artificial agency and its various realizations. This page translates the introductory remarks and existing text version, not treating the literature summary as the full paper.
Optional reflection: Identify an AI ethics dilemma on your own, distinguish which goals, choices, and consequences originate from humans and which are generated by the technical system, and examine how these two elements interact.
Video script one: How to differentiate intelligence from agency
Intelligence involves knowing what is known, forming concepts, and learning, analyzing, and understanding relationships. Agency involves what one can do, how one uses knowledge to pursue goals, and how one collaborates with others to form groups and act within environments.
First, become familiar with several research directions concerning human intelligence, which helps clarify which attributes are called "artificial intelligence" and which functions are called "machine action." The following is a comparative overview from a theoretical perspective, not indicating that any one theory fully explains all aspects of intelligence.
Psychological measurement: structure and ability differences
Psychological measurement focuses on assessing intelligence, with IQ tests being one important tool in this area. This research tradition assumes the existence of a general ability that influences various intellectual activities. As theory has evolved, it has further distinguished broader abilities such as learning and visual perception, from more specialized abilities like those required in playing the flute. As a result, measurement now provides not just a single overall score, but also attempts to describe a hierarchical structure of different ability levels.
The original explanation mentioned general factors and hierarchical ability research, and recommended further reading of the work by Charles E. Spearman and John B. Carroll, particularly the 1993 book *Human Cognitive Abilities*. It also noted that cultural factors pose challenges to measurement: how test results are influenced by language, experience, and environment requires separate discussion. The original explanation recommended Patricia M. Greenfield's research as an entry point for understanding how cultural factors challenge psychological measurement perspectives.
Cognitive Psychology: How do processes occur?
This line of research focuses on how mental representations are formed, retrieved, and connected, and also investigates processes such as memory, perception, and learning speed. Cognitive neuroscience also observes these activities from a physiological perspective.
Early explanations linked intelligence with processing speed and used analogies such as "the relationship between a bird and the sky corresponds to the relationship between an object and a tree." However, viewing the process as entirely sequential might be overly simplistic; real-world activities may involve parallel processing and mutual influences.
Research is still ongoing, aiming to understand how various processes collectively support intelligence. Computer models are also a research tool. To delve deeper, one can begin with the work of Allen Newell and Herbert A. Simon, learning how researchers use computational methods to simulate human problem-solving. Building models can help test hypotheses about cognitive processes, but does not equate to proving that computers and human minds are identical.
Biological perspective and enactive cognition
Neural foundations, brain regions, and their interconnections are observed in biological research. Some discussions are oversimplified into notions such as "left-brain type" or "right-brain type," which cannot be used to categorically classify individuals. We typically use both hemispheres of the brain simultaneously; some individuals have survived extensive surgical removal of large portions of one hemisphere, demonstrating that neither hemisphere can be considered entirely insignificant. Our understanding of the brain continues to evolve, with increasing research emphasizing integrative models rather than assigning each complex function to a single, isolated region. The division of brain functions must be considered alongside overall coordination and integration.
Embodied cognition further focuses on how the body participates in cognition. Traditional intuition often views the brain as a unidirectional control center; another research perspective emphasizes the interaction between the body, action, sensation, and environment. To delve deeper into embodied cognition, consider reading Eleanor Rosch's work. The original explanation also recommends George Lakoff and Mark Johnson's 1980 book *Metaphors We Live By*, to understand how metaphors and experience contribute to thinking.
Culture and Context
Situationism examines how environments and cultures interact with cognitive processes. What is considered "smart" in different activities may relate to experience, task demands, and social context.
The original explanation places discussions on multiple intelligences within this context, while also reminding us not to turn it into a fixed label suggesting that "each person can only learn in one preferred style." People may have preferences, but preferences do not mean they can only learn through a single type of media.
The original explanation also notes that this tradition recognizes that culture may play a role in defining intelligence, while at the same time there are ongoing efforts to seek a universal explanation of intelligence. These two tendencies are worth comparing. It recommends Robert Sternberg's 1985 book *Beyond IQ*, which examines intelligence from three interrelated perspectives: internal to the individual, the external world, and the individual's experience of interacting with the world. The emphasis is on recognizing interdependence, rather than reducing intelligence to a single, context-free number.
Video Script Two: Four Forms of Initiative and Collective Action
The original explanation, based on the read paper, divides agency into several levels. Here, the term "agency" is used more broadly than the everyday concept of "autonomous will," thus each definition must be examined carefully.
Natural spontaneity
Non-living systems interacting with their environment can also have significant impacts. For example, rivers flowing through landscapes gradually form canyons. They follow physical laws, possess no known intent or purpose, yet still alter the world.
biological motility
Organisms interact with their environment and other organisms, acting in response to needs such as food, shelter, and reproduction. Unlike rivers, they exhibit tendencies and regulatory behaviors related to survival.
Animals can also form cooperative activities, such as ants building nests or bees constructing hives. This can serve as an example of animal group agency: multiple individuals collectively producing an organized outcome.
The agency of man-made things
Non-living systems are manufactured by living organisms to achieve specific objectives. AI is a key example discussed here: the objective is set by humans, and the system carries out the task within a corresponding structure and environment.
The original connects this purpose with human design, but one cannot conclude that the system possesses the human's intent, nor can one regard all animal tool-use issues as resolved. The key is to distinguish the effects of artifacts from the human origin of purpose.
Human agency
Humans still have biological needs such as food, shelter, and companionship, while also possessing complex cognition and abstract thinking. The original explanation particularly emphasizes abstract ability: humans can consider distant futures, multi-generational impacts, cause-and-effect chains, and compound relationships that are not directly visible.
Human purposes thus extend beyond the immediate environment and can be imagined, regulated, and organized through long-term planning.
Human social agency
Groups jointly take action within institutions, producing culture and structure. These structures are built through human actions and in turn influence human choices. Understanding this mutual relationship helps avoid attributing all consequences of AI to individual users or individual models.
Bring the problem back to the specific system
First, specify the definition of agency you are adopting, then assess which capabilities AI has and which it lacks. Investigate who sets the goals, who provides the tools, who determines the permissions, and who can correct errors.
"and" Business Agent planning, memory, and tools Reading together can distinguish goal-oriented execution in engineering from autonomy and responsibility in philosophy.
Related Reading
When applied to specific systems, it can be compared Business Agents engineering capabilities, and use Trustworthy AI Supervision and accountability of inspectors.
Reference and Extended Reading
- AI as Agency without Intelligence : English, entry point to query the paper discussed in the original reading. The formally published version is available for reference. Publisher's page . Currently, the bibliography information and public preview are available; a fully translated, publicly accessible version has not been obtained. This page still provides only the original introductory text and accompanying written materials.
- UNESCO AI Ethics Recommendations : English, editorial addition, used to connect technical capabilities with human rights and responsibilities.