Note: This is a fictional conversation between a LLM and an Intentist who is trying to represent Chomsky’s views that language is primarily for internal use.
If language primarily evolved for mental computation and only secondarily for communication, how would this affect our view of LLM outputs that seem to be primarily for communication? (This conversation was before the advent of the ‘Deepthink’ setting that has some of the hallmarks of an internal thought process.)
Yes, if Noam Chomsky’s idea is taken as correct — that language primarily evolved for mental computation and only secondarily for communication — it would suggest a fundamental difference between human language and AI’s use of language. AI systems are, at their core, built for communication rather than intrinsic “thinking” or computation in the human sense. Here’s how this distinction plays out:
1. Human Language: Thinking First, Communication Second
• Chomsky’s theory emphasizes that the primary function of language is internal thought (i.e., structuring, reasoning, and conceptualizing ideas) rather than interpersonal communication.
• According to this view, aspects of linguistic structure (e.g., recursion, ambiguity, complex syntax) are not optimized for efficient communication but instead reflect the needs of the human mind for abstract computation and creativity.
• The complexity and redundancy in human speech could hinder clear communication but serve cognitive or expressive purposes.
2. AI Language: Communication First, Thinking Absent
• AI language models (e.g., GPT) are designed primarily for efficient and coherent communication. They generate responses based on patterns in data, with no intrinsic mental computation or internal “thought.”
• AI doesn’t “think” about what it says; it processes data to optimize outputs for understanding by humans or other systems. It lacks the internal generative grammar Chomsky attributes to human cognition.
• AI outputs are engineered for clarity, conciseness, and usefulness, avoiding the ambiguities or inefficiencies Chomsky suggests might hinder communication in natural language.
3. Implications of the Difference
• AI as a tool for communication: AI sidesteps the deeper cognitive architecture of human language and focuses entirely on the practical utility of generating comprehensible text.
• Human uniqueness in language: If language evolved for internal computation, it underscores the idea that human language has a purpose beyond communication — such as supporting consciousness or problem-solving — which AI does not replicate.
• Structure hindering communication: While AI seeks to simplify and streamline language, human language’s intricate structures (ambiguities, redundancies) may reflect the complexity of the human mind rather than the practicalities of exchanging information.
Conclusion
Yes, if Chomsky’s view is correct, it highlights a philosophical divide: human language serves thought first, communication second, while AI is purely communication-driven. This divergence underscores how AI, while powerful in generating human-like language, remains fundamentally different from human cognition. AI lacks the internal generative process that makes human language a tool for abstract mental computation.
What are the arguments for and against Chomsky’s position that language is primarily for internal computation?
The argument that human language is fundamentally different from AI language because humans use language for thought first while AI uses it for communication raises intriguing questions but has several problems and counterarguments. Here are the main issues, followed by potential answers:
Problems with the Argument
1. Lack of Evidence for “Thought-First” Language
• Problem: Chomsky’s claim that language is primarily for mental computation is not universally accepted. Some linguists (e.g., Michael Tomasello) argue that language evolved primarily for communication, not internal thought. If language’s primary function is communicative, then AI and human language might share more similarities than the argument suggests.
• Response: Supporters of Chomsky might point to the unique features of human grammar (e.g., recursion) as evidence of language being tailored for abstract thought. However, this remains theoretical.
2. AI Does Perform “Internal Computation”
• Problem: While AI doesn’t “think” like humans, it does process information internally before generating responses. Large language models use latent space representations to simulate abstract reasoning and meaning-making. This internal “computation” could be likened to human mental processes, complicating the distinction.
• Response: Critics might argue that AI’s internal computations lack intentionality, creativity, and subjective experience, making its “thinking” fundamentally different from human cognition.
3. Human Language Is Also Communicative
• Problem: Even if language aids thought, humans also use language extensively for communication. Features of language, like shared context, idioms, and redundancy, seem optimized for interpersonal understanding. If humans balance both thought and communication, the strict dichotomy between humans and AI weakens.
• Response: Proponents of the argument might say that while humans communicate, the origin of language in humans lies in thought, which is not the case for AI.
4. Defining “Language” Is Tricky
• Problem: The argument assumes a clear definition of language, but the concept is nebulous. If language is broadly defined as a system for encoding and transmitting information, then AI and human language might not be fundamentally different.
• Response: To distinguish human and AI language, one could argue that intention and consciousness are essential components of human language, which AI lacks.
5. AI’s Communicative Optimization Mirrors Human Pragmatics
• Problem: AI’s focus on clarity and goal-oriented communication might resemble aspects of human language use (e.g., Grice’s conversational maxims). The distinction between thought-first and communication-first becomes blurry when considering that humans adapt their language for specific goals, much like AI does.
• Response: Chomskyans might argue that such pragmatics are secondary adaptations in humans, whereas for AI, they are primary.
Good Answers to These Problems
1. Human-AI Language Comparison as a Continuum
• Language in humans and AI could be seen as points on a spectrum. While humans use language for both thought and communication, AI might represent a more extreme, communication-only case. This allows for differences without dismissing the analogy entirely.
2. Focus on Intentionality and Subjectivity
• A strong counterargument to AI being similar to human language use is that humans imbue language with intentionality and subjective meaning. AI lacks these qualities, as it doesn’t have desires, goals, or an inner life. This distinction preserves the argument’s core.
3. Acknowledging Different Evolutionary Trajectories
• Human language evolved over millions of years with pressures for both thought and communication. AI language models, by contrast, are engineered within a few decades with explicit goals for optimization. The differing evolutionary contexts support the argument for fundamental differences.
4. Revising the Role of Communication in Humans
• Even if language originated as a tool for thought in humans, its extensive adaptation for communication could indicate that both functions coexist in a unique balance. AI, however, has no such duality, maintaining its focus solely on external communication.
5. Limitations of AI Language Models
• AI’s communicative abilities are constrained by their training data and lack of true understanding. They cannot generate meaning beyond patterns in data, underscoring their fundamental difference from human language use.
Conclusion
The argument that human and AI language differ due to their origins (thought-first vs. communication-first) raises valid points but is complicated by overlapping features like pragmatics and internal computation. A nuanced answer might recognize the functional similarities between AI and human language while emphasizing the qualitative differences in intentionality, subjective experience, and evolutionary context.
The argument that human language is fundamentally different from AI language because humans use language for thought first while AI uses it for communication raises intriguing questions but has several problems and counterarguments. Here are the main issues, followed by potential answers: