ChatGPT is not intelligent. In the following, I embark on a critical examination of ChatGPT's capabilities, comparing them against the multifaceted criteria that define human intelligence. Through a detailed analysis, I aim to explore whether ChatGPT, a prominent artifact of modern artificial intelligence, can truly exhibit intelligent behavior akin to that of humans. By delving into nine attributes such as learning, memory, language comprehension, and critical thinking, I intend to provide a nuanced perspective on the nature of intelligence in artificial systems. Ultimately, I conclude that despite ChatGPT's sophisticated functionalities and its ability to mimic certain aspects of human intelligence, it is not truly intelligent, because it misses a 10th attribute : subjectivity.
This article is a summary. Read the full paper with all the experiments here.
This short article is also available as an interactive version here.
0. (Artificial) Intelligence
Intelligence is a complex and multifaceted concept that encompasses the ability to learn from experience, adapt to new situations, understand and handle abstract concepts, and use knowledge to manipulate one's environment. It involves various cognitive processes, including perception, learning, memory, problem-solving, and reasoning.
Artificial Intelligence (AI) refers to the simulation of human intelligence in machines that are programmed to think like humans and mimic their actions. The term can also apply to any machine that exhibits traits associated with a human mind, such as learning and problem-solving. AI technology employs various disciplines, including computer science, cognitive science, linguistics, psychology, neuroscience, and philosophy. It has numerous applications across different fields, such as robotics, natural language processing, image recognition, healthcare, and autonomous vehicles, significantly impacting society and industries by enhancing efficiency and enabling new capabilities.
Can an artifact truly be intelligent from a human standpoint, considering we view ourselves as intelligent beings? To explore this, I will compare ChatGPT, a significant modern artifact, against nine criteria that are frequently associated with human intelligence. I will examine ChatGPT closely, using sample prompts and clear explanations to set the stage for this comparison. However, it's up to you, dear reader, to thoroughly analyze and critically assess these points. My detailed evaluation uncovers some intricacies of this endavour, highlighting the possibility of intelligent behavior in artifacts like ChatGPT.
1. Action
Action, when considered as an important ability of human or artificial agency, refers to the capacity to initiate and execute tasks or behaviors towards achieving specific goals or responding to environmental stimuli. In the context of agency, which denotes the capability of an entity to act independently and make choices, action embodies the tangible expression of that autonomy. For humans, this encompasses a wide range of voluntary and involuntary activities, guided by cognitive processes such as decision-making, reasoning, and planning, and is deeply influenced by individual intentions, desires, and motivations. In the realm of artificial agency, such as in intelligent systems or robots, action is similarly fundamental but is enacted through programmed algorithms and decision-making processes that allow these entities to perform tasks, solve problems, or interact with their environment.
Can ChatGPT act?
Sure. Tell it anything. And you will get a reaction. For this, I use a prompt which basically is an instruction or set of instructions given to a computer system to initiate a specific action or response.
In my view, the response from ChatGPT fundamentally represents an interaction. I communicate with it using language, which highlights the importance of understanding language as a crucial component of intelligence. Language serves as the medium through which I express my commands and through which ChatGPT articulates its responses, making it central to our exchange and, by extension, to the demonstration of intelligent behavior.
This example clearly shows that ChatGPT possesses the ability to perform actions to some extent. Through its responses, it demonstrates an ability to interact and execute tasks based on the prompts I provide, indicating a level of agency in its operations.
2. Language
Language is a complex system of communication that is a fundamental aspect of human culture and cognition. It is characterized by the use of symbols, sounds, and gestures to express ideas, emotions, and information. Language can be spoken, written, or signed, and it is not limited to human speech alone. Various forms of communication in the animal kingdom exhibit language-like features, but human language is unique in its complexity and capacity for abstract and creative expression. Language enables humans to convey not only concrete information but also abstract concepts, emotions, and complex ideas. It is a central aspect of human identity and social organization, shaping thought, culture, and social relationships.
Large language models (LLMs) are advanced artificial intelligence systems designed to understand, generate, and work with human language at a large scale. These models are a subset of machine learning models trained on vast amounts of text data. They learn the statistical patterns, structures, and nuances of language from this data, enabling them to perform a wide range of language-based tasks. Key characteristics and capabilities of large language models include:
Understanding Context: LLMs can comprehend context and meaning in text, allowing them to respond to queries, summarize information, and even engage in conversation in a way that often seems intuitive or human-like.
Text Generation: They can generate coherent and contextually relevant text based on the input they receive. This capability is used in applications such as content creation, conversation agents (chatbots), and more.
Language Translation: LLMs can translate text from one language to another, learning from the vast corpus of multilingual text available in their training data.
Information Retrieval and Summarization: They are adept at sifting through large volumes of text to find relevant information and can summarize long documents, highlighting key points and information.
Sentiment Analysis and More: LLMs can assess the sentiment of text (positive, negative, neutral) and perform other nuanced language tasks, such as detecting sarcasm, intent, or emotion.
Customization and Fine-tuning: While the base model is trained on a diverse and extensive dataset, LLMs can be fine-tuned on specific datasets or for particular tasks, enhancing their performance in niche applications or industries.
Understanding the capabilities of Large Language Models (LLMs), I'm now equipped to craft more effective prompts than those we encountered in the previous chapter. Indeed, I'm not confined to a single line; I have ample space to elaborate. To leverage ChatGPT's LLM capabilities more fully, I could, for instance, provide detailed context or specify the exact nature of the information or response I'm seeking. The more information I supply, the more tailored and focused the response will be. I can also structure my prompts in any way I see fit, such as setting the scene before giving an instruction or clearly stating the application of the requested information.
This example highlights that ChatGPT can effectively use symbols to convey ideas and information. By interpreting my prompts and generating coherent responses, it showcases its capability to engage in communication.
3. Memory
Memory is a fundamental cognitive process that enables individuals to encode, store, and retrieve information over time. It plays a crucial role in learning, decision-making, and navigation through daily life, allowing past experiences to influence current and future behavior. Memory processes are influenced by a variety of factors, including attention, emotion, repetition, and the use of mnemonic devices or strategies. Understanding memory and its mechanisms is a central topic in the fields of psychology, neuroscience, and cognitive science, with significant implications for education, therapy, and the treatment of memory-related conditions.
In my interactions with ChatGPT, it's evident that it has a form of memory. During our dialogues, it retains information from our previous exchanges and incorporates this into our ongoing conversation. This ability to store and reference past interactions enhances the relevance and coherence of its responses, making our dialogue more meaningful and connected.
Clearly, ChatGPT has the ability to remember the context from an initial interaction, grasp the situation presented in a subsequent exchange, and follow the instructions and evaluation criteria laid out in a third interaction. It effectively memorizes previous interactions, which allows for a more cohesive and informed dialogue as we progress. This capability underscores the sophisticated nature of its memory function in maintaining the continuity of our conversation.
Seeing such results, I find it puzzling that many individuals don't utilize ChatGPT's capability for longer-term dialogues. Often, people ask a few questions and then shift to other topics or start new conversations. To me, maintaining a long-term dialogue, which ChatGPT can store and recall, seems like a gold mine for more detailed and precise interactions. This continuity can significantly enrich the quality and depth of the conversation, making it a valuable tool for more complex and nuanced exchanges.
4. Learning
Learning is a fundamental cognitive process that involves acquiring new, or modifying existing, knowledge, behaviors, skills, attitudes, or preferences. It is a complex phenomenon that results from the interaction between an individual's experiences and their neurobiological processes. Learning enables individuals to adapt to their environment, solve problems, make decisions, and process new information effectively.
To an extent that I'll leave to your judgment, ChatGPT demonstrates at least some capabilities of learning. Through our interactions, it seems to adapt and respond in ways that suggest a form of information retention and application, indicative of learning processes at play.
5. Autonomy
Autonomy refers to the capacity and right of individuals or entities to make their own choices and govern themselves independently. This concept is multifaceted and can be applied in various contexts, including personal, ethical, political, and organizational spheres. Autonomy is widely valued as a fundamental aspect of human dignity and rights, underpinning democratic principles, ethical frameworks, and legal systems that aim to respect and protect individual and collective freedoms. The pursuit and implementation of autonomy should here consider ethical principles like justice, equity, and respect for the autonomy of others, balancing individual freedoms with social responsibilities and collective well-being.
Reflecting on the rapid advancement and integration of AI, particularly with the addition of plug-ins and GPTs from the store, I'm struck by how ChatGPT can now operate beyond its initial confines. A prime example is the Aurora extension, which enables ChatGPT to interact with light systems, adjusting brightness, changing colors, and setting the ambiance. While this might seem amusing initially, the implications leave me astonished. The thought that if ChatGPT can control lights, it might, in theory, have the capacity to initiate far more consequential actions, such as launching a nuclear missile, is deeply concerning. It appears we may have overlooked a critical aspect by granting such levels of autonomy to AI, underestimating the potential ramifications on safety and control.
ChatGPT can act outside of ChatGPT.
6. Critical thinking
Critical thinking is a process of actively and skillfully conceptualizing, applying, analyzing, synthesizing, and evaluating information gathered from observation, experience, reflection, reasoning, or communication, as a guide to belief and action. It involves the ability to engage in reflective and independent thinking. Critical thinking is widely recognized as a foundational skill, valuable not only in academic and professional settings but also in everyday decision-making. It empowers individuals to make reasoned judgments that are logical and well-thought-out. It is about being an active learner rather than a passive recipient of information, and it involves a skeptical and questioning approach to statements, assertions, and information, regardless of the source. Developing critical thinking skills is a lifelong process that is central to education and personal growth.
To an extent that again I'll leave for you to assess, ChatGPT appears to exhibit capabilities of critical thinking. Beyond that, it seems able to critically evaluate its own reports, ones that it previously generated from case studies created in separate, independent dialogues. This level of self-reflection and critique in its responses suggests a sophisticated processing capability that mirrors aspects of critical thinking, adding an intriguing dimension to its interactions.
7. Logic
Logic is a branch of philosophy that studies the principles and criteria of valid inference, reasoning, and argumentation. It is concerned with distinguishing good reasoning from bad, correct from incorrect. Logic provides a systematic framework for constructing, analyzing, and evaluating arguments by assessing the structure and content of statements and their interrelations. Logic is fundamental not only in philosophical inquiry but also in mathematics, computer science, and linguistics, as well as in everyday reasoning and problem-solving. It helps individuals to think critically and analytically, providing tools to construct coherent arguments and to dissect and evaluate the arguments of others.
To demonstrate ChatGPT's logic capabilities within the context of the digitization project at (fictional) Berne City Hospital, I asked for an analysis of the potential challenges and solutions related to the integration of the new digital system with existing healthcare technologies.
In a specific context and situation, ChatGPT showcased its logical capabilities by proposing solutions and then translating the rationale behind its reasoning into a Prolog program. I subsequently executed this program successfully in a third-party environment, confirming the effectiveness and logical soundness of its suggestions.
8. Empathy
Empathy is the ability to understand and share the feelings of another person. It involves putting oneself in another's shoes, to comprehend their emotions and perspectives, and often to express and share in those feelings. Empathy plays a crucial role in social interactions, relationships, and emotional intelligence. It fosters emotional connections, facilitates effective communication, and underpins compassionate and cooperative behaviors. Empathy is vital in various professional fields, especially those involving interpersonal interactions, such as healthcare, education, and social work. Developing empathy involves a combination of nature and nurture, influenced by genetic factors, individual experiences, upbringing, and education. It is considered a crucial element in reducing conflicts, enhancing relationships, and promoting emotional well-being and mutual understanding among individuals and communities.
ChatGPT proved its ability to tailor a complex message for various personal profiles, considering their unique perspectives to devise effective communication alternatives. This adaptability in addressing diverse viewpoints highlights its sophisticated approach to personalized interaction.
9. Originality
Originality refers to the quality of being novel, unique, and creative, displaying a departure from the norm or the conventional. It is often associated with the ability to produce ideas, expressions, or solutions that are fresh, unconventional, or unprecedented. Originality is a highly valued trait in various fields such as art, literature, science, technology, and business, where innovative and distinct contributions can lead to significant advancements and enrich cultural and intellectual landscapes. Originality is not just about being different for the sake of difference; it involves meaningful innovation and the creation of value through new combinations, interpretations, or expressions. While originality is inherently challenging due to the vast array of existing ideas and creations, it remains a driving force behind cultural, scientific, and technological progress.
Together, ChatGPT and Midjourney showcased their ability to be creative and generate meaningful, visual alternatives, demonstrating a remarkable blend of textual understanding and visual interpretation.
10. Subjectivity
Objectively speaking, ChatGPT can exhibit intelligent behavior. It demonstrates interesting capabilities aligned with nine criteria commonly associated with intelligence, showcasing both knowledge and know-how.
Knowledge encompasses the information, insights, and skills gained through experience, education, observation, or discovery, including factual, descriptive, and principled understanding acquired via study or practice. It's foundational in human culture and societal development, facilitating informed decision-making, problem-solving, and expertise contribution.
Know-how, in contrast, pertains to practical expertise or proficiency in efficiently and effectively accomplishing tasks. It embodies the necessary skills, information, and experience for task execution or problem resolution. Primarily tacit, know-how is derived from direct experience, comprising techniques, practices, shortcuts, and hands-on understanding within a specific domain.
Thus, knowledge represents the theoretical or practical grasp of subjects, focusing on facts, information, and principles, while know-how emphasizes the practical application of skills in real-world contexts, involving procedural knowledge and hands-on experience. Both elements are crucial and interlinked, with deep subject knowledge enhancing practical know-how, and practical experiences enriching overall knowledge.
Hence, despite these nine objective performances, there remains one final barrier for me to consider ChatGPT as intelligent: it is still an object, even if it exhibits intelligent behavior. As long as we do not perceive it as a subject, meaning as long as we maintain control and remain the subjects who operate it as an object, it cannot be deemed intelligent.
This is closely linked to a third "know-question": When addressing the "Who?" question in the context of knowledge and skills, the term "know-who" often emerges. Know-who pertains to understanding who possesses specific knowledge or expertise, crucial for solving a particular problem or executing a task. It involves:
An awareness of the social network and the ability to navigate it effectively, recognizing who holds the requisite knowledge, skills, or competencies and understanding how to access that expertise.
The significance of "know-who" in collaborative settings and sectors where leveraging a group or community's collective expertise is vital for success, playing a key role in project management, research, development, and fields demanding innovation and intricate problem-solving.
The capacity to forge and sustain relationships, discern strengths and weaknesses within a network, and effectively use social capital to foster knowledge exchange and accomplish objectives.
Know-who complements other forms of knowledge, such as "know-what" (knowledge) and "know-how" (skills), by focusing on identifying who holds necessary knowledge or skills and how to engage with them. It emphasizes the social aspects of utilizing knowledge and skills to achieve goals or resolve issues.
Here, we are faced with a choice that is as stark as black or white, zero or one:
In French, a crucial distinction exists between the questions
"Comment faire pour bien faire faire?"
and
"Comment faire faire pour bien faire?"
The former implies delegating to another entity, while the latter suggests retaining the role of the actor. This distinction is profoundly significant. To me, subjectivity is the tenth attribute of intelligence. Despite observing somewhat intelligent behaviors across the first nine criteria, I conclude that ChatGPT lacks intelligence. Artifacts are mere objects, and while they can exhibit intelligent actions, they cannot possess intelligence themselves. This quality is exclusive to us, human beings.
Conclusion
Labeling objects as intelligent, and by extension, the term "artificial intelligence," is fundamentally flawed in a strict sense. Current discussions around AI, especially in the educational sphere, largely revolve around the former question. However, to approach the ethical and responsible use of ChatGPT and AI at large, we must confront the latter question. My stance on this matter is unequivocal:
I will never attribute subjectivity to an artifact.
Disclaimer
This essay represents a personal point of view. It must not be assimilated with any of my institutional affiliations.
I made extensive use of ChatGPT, v.4 team version, when writing this essay to
• explore and understand,
• structure,
• develop,
• deepen and
• widen my arguments, as well as to
• write, translate, and correct the text.
Objectively speaking, there is no part of the text in which ChatGPT has not been used.
Subjectively considered, however, all sentences reflect my opinion and are based on my initiative.