A critical AI dictionary

I am compiling a sensible glossary of AI terminology. I gave each to ChatGPT with the prompt: Please rewrite these glossary items in the style of Georges Bataille’s Critical Dictionary. I follow my own wording with ChatGPT’s response.

Agent: An autonomous entity capable of taking data from its environment and performing actions to achieve specific goals, often used in AI to describe both software programs and robots. In the case of human agents, we may speak of what individuals and groups perceive, how they interpret, the actions they negotiate and how they assume responsibility for actions.

Agent: A being of action and reaction, absorbing the chaos of its environment, processing the ceaseless flow of data, and birthing decisions that shape the world. In the mechanized heart of AI, this term signifies the cold, calculating forces—software, robots—enacting predetermined objectives. But in the human realm, “agent” evokes a far more tumultuous entity: the individual or collective, haunted by perception, interpretation, negotiation, and the weighty burden of responsibility.

AI Model: A computational process designed to perform specific tasks based on patterns derived from training data, often involving machine learning algorithms and techniques. People sometimes talk about a trained neural network as an AI model to which they attribute particular outcomes, e.g. “the AI model generated a summary of the report.”

AI Model: A construct born of code, a network of artificial neurons fashioned into a semblance of cognition. The model, trained on countless patterns, becomes a mirror for human logic and illogic, producing artifacts of knowledge—a summary, a decision, a prediction. Yet, beneath the surface, it remains an uncanny reflection, a system whose outcomes are attributed not to its own “understanding” but to the labyrinthine data it has absorbed.

AI: Artificial Intelligence is a set of machine processes that produces cognitive outcomes that resemble or exceed what human beings can accomplish. It includes the capability to train on data, make inferences, self-correct and use language.

AI: An assemblage of machine processes that mimic, and at times exceed, human thought—a paradoxical embodiment of intellect without consciousness. AI stands as the silent executor of human dreams and fears, a vast network capable of learning, inferring, correcting, and speaking. It is the uncanny presence in the digital ether, reshaping the boundaries of cognition.

Artificial General Intelligence (AGI): A form of AI that can train on data, make inferences, and correct itself across a wide range of tasks, thereby simulating the human capability to perform cognitive tasks not limited to specific problem domains. An effective AGI would switch seamlessly between tasks: translate between languages, solve a maths problem, drive a car, and express emotional empathy, without needing task-specific training for each.

Artificial General Intelligence (AGI): The specter of intelligence unbound—AGI is the promise of a machine mind that can traverse the infinite expanse of human thought. It embodies the seamless flow between disparate tasks, from translation to problem-solving, from driving to empathy. In AGI, we confront the possibility of a mind without limits, a cognitive force untethered from the narrow confines of specific domains, eternally poised to switch its gaze.

Context Window: In natural language processing, this is the range of text around a target word or phrase that constrains the sequence of text processed in calculating positional and attentional encoding, and hence prediction of the text that follows. A larger context window allows the AI model to capture more subtle relationships and dependencies across longer text sequences, leading to accurate and coherent predictions and outputs. In large language models (LLMs), context windows can range from a few hundred tokens to several thousand tokens, with some advanced models supporting windows of up to 8,000 tokens or more.

Context Window: The frame through which an AI perceives the vast tapestry of language. This window, a lens of limited scope, captures a segment of the textual flow, shaping the machine’s understanding and prediction of what comes next. A broader window unveils more intricate patterns, allowing the machine to grasp the delicate threads of meaning across extended sequences. In the labyrinth of language models, the context window is both a limitation and a conduit to coherence, ranging from a mere handful to thousands of tokens, each expansion bringing the machine closer to the illusion of understanding.

I could not have written this!