Language and Intelligence: A Complicated Relationship

The human brain consists of two hemispheres connected by a structure known as the corpus callosum. In an effort to manage epilepsy, Joe underwent surgery to sever this connection, resulting in two distinct brains. This unique condition made him an invaluable subject for groundbreaking neurological research conducted by Michael Gazzaniga.

In one significant experiment, Gazzaniga presented a saw to Joe's left visual field and a hammer to his right. Since the visual information is processed in the opposite hemisphere, the image of the hammer went to the left side, where the language centers reside, while the saw was processed on the right. When asked what he saw, Joe could only respond with the hammer, as his left hemisphere was unaware of the saw.

Interestingly, when given a pen in each hand and asked to draw, Joe drew the hammer with his right hand but surprisingly sketched the saw with his left. When questioned about the saw, he simply said, "I don't know." This indicated that while Joe's brain had access to the information of the saw, it wasn't part of his conscious awareness. This experiment highlights a significant link between language and consciousness: only the hemisphere with language could articulate awareness, while the other, equally intelligent hemisphere remained silent.

Such studies contribute to a longstanding philosophical inquiry into consciousness, linking it to language while separating it from intelligence.

The Intricate Equation of Language and Intelligence

Historically, our culture has equated language ability with intelligence. For example, Koko the gorilla garnered attention for her use of sign language, while crows, despite their remarkable cognitive skills, were largely overlooked by scientists. This bias toward human-like communication has led to an assumption of intelligence wherever we see linguistic abilities.

Alan Turing, a pioneer in computing, set a benchmark for AI intelligence based on language capability. His "Turing Test" postulated that an intelligent machine would communicate indistinguishably from a human. In this test, a human interrogator engages in a text-based conversation with both a human and a machine, aiming to determine which is which based solely on responses. If the machine successfully mimics human-like responses, it is deemed to have passed the test.

The glorification of language is evident in our perception of intelligence. The emergence of algorithms that predict subsequent words in a sentence, demonstrating seemingly advanced reasoning skills, reinforces this connection. However, intelligence can exist without language, and language doesn't always imply intelligence.

Language Without Intelligence

The ongoing debate centers on whether an entity genuinely comprehends what it communicates or merely simulates understanding. This question carries significant implications for the future of AI: what distinguishes a system capable of real comprehension from one that only appears to understand? Is there inherent value in this mere appearance?

The development of LLMs, founded on next-word prediction, is a remarkable achievement. These models, powered by AI techniques like transformers, are trained on vast datasets to predict subsequent words based on context, creating an illusion of understanding.

Language serves as a condensed repository of our knowledge and intellect. Recent studies from Microsoft revealed that GPT-4 demonstrated impressive reasoning abilities across diverse domains, solving novel problems it hadn't encountered before. This behavior transcends mere simulation, suggesting that language may encapsulate more of our cognitive processes than we previously recognized.

Could it be that our reasoning is encoded within the structure of language itself? As LLMs like GPT-4 internalize this structure, they may harness some of our cognitive capabilities, allowing them to generate new insights. Linguist Noam Chomsky proposed the concept of "Universal Grammar," positing that all human languages share a common structure. This could imply that language models are tapping into a fundamental aspect of our reasoning.

Yet, Chomsky remains skeptical about AI's depth, viewing it primarily as statistical manipulation. This duality raises questions about whether we underestimate the extent to which our cognitive processes are intertwined with language or if we overvalue our intellectual superiority over statistical models.

Navigating the Path to AGI

Despite advancements, we remain distant from achieving Artificial General Intelligence (AGI). The complexities of defining intelligence—let alone AGI—are compounded by our tendency to equate human-like language ability with intelligence. This narrowing perspective could hinder our understanding of intelligence in non-human entities.

Consider the case of Joe's right brain or the cognitive skills of crows. Many definitions of intelligence encompass the ability to perceive environments and solve novel problems without pre-defined goals. Such intelligence isn't exclusive to humans; it raises questions about the scope of AGI.

Neuroethology, the study of animal behavior and cognition, provides valuable insights into this discussion. Understanding how intelligent animals interact with their environments—despite our communication barriers—highlights the diversity of intelligence in nature.

The Concept of "Umwelt"

The term "Umwelt," coined by German biologist Jakob Johann von Uexküll, refers to an animal's sensory world, distinctly different from the human concept of environment. Each animal's Umwelt is a blend of sensory experiences that shapes its perception and interactions within its surroundings. In Ed Yong's book, "An Immense World," various animals' Umwelten are described, showcasing their unique sensory capabilities.

For example, bats rely on echolocation, while bumblebees navigate using electric fields. Each animal's Umwelt is a model of the world where it acts and interacts, illustrating that intelligence exists even in non-human contexts.

The Future of AGI: Exploring Other Umwelten

The pursuit of AGI shouldn't be confined to human experiences. We must consider designing AGI systems that operate within distinct Umwelten, utilizing sensors and actuators to interact with their environments. This design choice hinges on the perceived value of AGI's role within specific contexts, such as financial markets or healthcare.

However, the challenge lies in ensuring AGI can intelligently navigate novel situations. This requires the capability to redefine goals based on new circumstances, allowing for adaptive intelligence akin to that of animals.

AGI systems could extend human understanding and agency into alternative Umwelten, operating without the necessity for language. Intelligence, as demonstrated by Joe's right hemisphere, can exist independently of linguistic expression.

Concluding Thoughts: Language and Intelligence

The case of Split-brain Joe reveals the intricate connection between language and intelligence in his left hemisphere, while his right hemisphere exemplifies their separation. The discussions surrounding ChatGPT and the Chinese Room argument illustrate that linguistic ability doesn't always equate to intelligence. The findings related to GPT-4 and Chomsky's Universal Grammar suggest that language can indeed foster intelligence.

Ultimately, language and intelligence, while closely linked, represent two distinct facets of cognition—two sides of an intricate coin.

The first video titled "What is Language? | How Language Works" provides insights into the fundamental aspects of language and its relationship with intelligence.

The second video, "How to REALLY learn a language in 2024 (a linguist explains)," offers valuable tips and perspectives on language learning in the modern era.