## The Wittgensteinian Turn in Artificial Intelligence
When looking beyond mental offloading and human-AI collaboration, a deeper philosophical puzzle emerges: *How can large language models (LLMs) write so fluently without actually understanding what they are saying?* To untangle this, technology theorists increasingly turn to the Austrian-British philosopher Ludwig Wittgenstein. In his seminal work *Philosophical Investigations*, Wittgenstein argued against the idea that words get their meaning by acting as labels for mental objects or fixed dictionary definitions. Instead, he proposed that meaning is derived from use within social contexts, which he called **language games**—rule-bound activities where words function like moves in a game (such as giving orders, joking, or telling stories).
This philosophy mirrors how modern generative AI operates. An LLM does not possess an internal, human-like concept of a "bank" or "freedom"; rather, it maps statistical relationships based on how words appear relative to one another across vast datasets. In a sense, transformer-based AI systems are master players of Wittgensteinian language games, predicting the next likely "move" in a conversation without needing to look up a hidden definition.
## Fluency Without a "Form of Life"
Yet applying Wittgenstein to AI also exposes a profound limitation in how machines affect human thought. Wittgenstein asserted that language games are rooted in what he called a **form of life**—the shared biological realities, cultural habits, and emotional experiences of being human.
Consider asking an AI to draft a condolence letter. The resulting text may be tactful, comforting, and stylistically flawless, even though the system has never lost a loved one, experienced grief, or stood by a graveside. This creates a disorienting illusion for human thinkers:
> If you read it blind, you may not find anything you can point to. This is not an exceptional case. Models pass bar exams and medical tests, run customer-service conversations without being detected, argue, phrase, joke.
> — The Philosophers of AI, *Can You Master Language Perfectly Without Understanding a Single Word?*
This fluency tricks our cognitive architecture. Because humans naturally link language proficiency with inner consciousness, we assume the machine is reasoning alongside us. In reality, it is executing structural patterns divorced from any lived reality.
| Dimension | Human Language Use | LLM Language Generation |
| :--- | :--- | :--- |
| **Grounding** | Anchored in a shared biological and social "form of life" | Anchored in statistical probability and token prediction |
| **Comprehension** | Tied to personal intentions, feelings, and consequences | Output generated without internal awareness or stakes |
| **Rule-Following** | Flexible adaptation based on unwritten social norms | Execution of formal parameters and prompt constraints |
## Redefining the AI Interaction
Recognizing LLMs through the lens of language games changes how we ought to approach them. Rather than treating AI as an autonomous intellect or a seamless partner, a Wittgensteinian framework suggests we view interactions with AI as strict, bounded rule-games. Prompt design—specifying a role, goal, and format—is essentially defining the rules of a specific language game.
The primary risk to human thinking, therefore, is not just lazy mental outsourcing, but a subtle colonization of our own language games. If we begin to adopt the statistical, surface-level patterns of AI as our standard for deep communication, we risk hollowing out the experiential depth that gives human thought its true meaning.