A sense of disappointment and deflation, but I will no longer deny it. Following Openai's release of GPT-5, the internet was quickly flooded with tweets and posts from industry insiders, reluctantly acknowledging the work of Silicon Valley's Gadfly Gary Marcus. Since the late 2010s, cognitive scientists have been warning about the limitations of large-scale language models (LLMs). This challenges the frustration of deep learning enthusiasts and figures like Openai CEO Sam Altman, who publicly defended more difficult stories.
Marcus's critique rests on what he sees as an inherent vulnerability of deep learning: a data- and energy-hungry, brute-force approach that “understands” and produces natural languages that have proven to be dazzling but fundamentally vulnerable. These fear quotes are guaranteed. LLMS doesn't understand anything. It is not usually the way to mean this term. Instead, these vast symbol manipulation machines use vast computational resources to predict the most statistically likely next word or token based on patterns extracted from the collective human corpus. The results are so impressive that even general artificial intelligence (AGI) and conscious machines have re-entered mainstream discourse. However, such speculation rests on the sandy foundations of personified projections and philosophical naiveness, disrupting depth and imitation and flow of the surface to mimic the mind and imitate.
Marcus's story resembles the tales of previous figures in the history of AI realists, and his work revealing the inherent limitations of his technology has also made him a pariah. In the early 1980s, the so-called expert systems (symbolic logic engines also known as the good old days of AI (Gofai)) surged optimistically around symbolic logic engines designed to mimic human reasoning, so that Philosofa's Huber Treyfus declared himself undeclared. He had already spent more than a decade trying basic assumptions in AI research. Following his 1965 Land Report, his 1972 book What the computer can't doargued that authentic intelligence cannot be captured in a rule-based system or computational representation alone, because it is embodied, positioned and context-dependent.
Using Heidegger and Merlot Ponty, Dreyfus argued that expertise and meaning arise not from the follow-up of rules, but from embodied know-how and world existence. These claims were hostile by many people in this field, especially in places like MIT, where symbolic AI was dominated. If his critics considered it an issue of abstract symbol manipulation, Dreyfus argued that such manipulation never approximates an intuitive grasp with the pre-reflection of the meaning that characterizes human. There is.
Just as Marcus opposes the flawed assumptions underlying modern LLM, Dreyfus warned that, no matter how powerful, he cannot achieve authentic human-like insights and expertise as long as he is identifiable, embodied, and blinded the meaningful whole in which human cognition is constantly operating. Despite these restrictions, Marcus sees hope in revisiting Gofai. He argues that its fundamentally different approach could provide much-needed resilience and improved inference ability when deep learning exceeds the points that reduce returns. He may be right. Nevertheless, there is reason to suspect that Dreyfus's half-century-old evaluation is just as relevant as ever, especially with regard to the higher aspirations of modern AI followers.
The hemispherical theory of philosopher and neuroscientist Ian McGilchrist provides additional weight to Dreyfus' criticism by shedding new light on how we understand the human mind. Unlike previous theories on the differences in the brain hemisphere, focusing on what They claim that McGilchrist claims that there is an important difference in how They do that. Based on his experience as a psychiatrist and neuroscientist, and a vast collection of research derived from patients suffering from brain injury, he offers an intriguing account of the fundamentally different world in which each hemisphere of the brain resides.
With the right hemisphere, things are unique, context-dependent, constantly changing, never fully grasped, and not completely separate from their involvement. It reveals a world of depth, ambiguity, beauty, and moral importance. It's the world we participate in. It's richer and more true, but it's also difficult to pin and explicitly fix in language. Still, what really matters: relationships, meanings, deep and intuitive understanding – is essential.
The left hemisphere, on the other hand, provides a fragmented, descriptive and described vision of the world aimed at controlling and manipulating. It recognizes reality as a collection of static, isolated parts: abstract, unembodied aspects that are stripped of nuance, ambiguity, or emotional resonance. In this mode, anxiety becomes a problem of bottom-up structures aimed at generating conclusions that seem unpredictable because it excludes all that resists classification.
Unlike the right hemisphere, the left hemisphere recognizes an inanimate universe, where the emphasis is on usefulness and efficiency for understanding the truth. There is clarity, but it sacrifices depth. If they are registered at all, beauty, morality, and empathy are reduced to calculations or results. Its ratings are confident and even arrogant, but in the end it's shallow. And this approach, which is necessary to reduce the infinite complexity of reality and allow you to map, navigate and respond with decisive actions, is a poor guide to meaning, relationships, and the dimensions of life: love, humor, beauty, and holiness.
Both perceptual modes are essential, but they are neither symmetric nor interchangeable. The left hemisphere is dependent on the right at both the beginning and end of the cognitive process. Initially, it relies on the right to disclose the world as a living and dynamic whole. The left then constructs a simplified outline (MAP) that is often useful and essential but essentially limited. Maps can be misleading. They are partial and inevitably reduce the multidimensional richness of what is represented as two-dimensionally manageable. Finally, at the far end of the process, what is needed to interpret these expressions in whole light and to revive the map with meaning is once again the correct hemisphere.
The right hemisphere understands the role of the left, while the left hemisphere operates within a stiffer, separated frame, but understands the role of the right. A well-documented trend in the neurological literature is to dismiss its inability to grasp. It is to reduce confidence, denial, or value that cannot be reduced to its own terms. The results are often tragic, not only for individuals with damage to the right hemisphere, but also for societies and systems that are overly dependent on the mode of attention in the left hemisphere due to similarity.
This is all very beneficial when it comes to AI. In many ways, the history of AI reflects the history of efforts to model the mind. And it reflects something deeper. Attempts to model intelligence are always shaped by hemispheres that express, abstract, and clarify things. We model the mind using the only tool available to us: tools based on left hemisphere cognition. Even our most sophisticated models inevitably reflect their limitations. We create maps of thinking, language, learning, and now intelligence itself more refined than ever before, but we tend to forget that we cannot pull the topography from the map.
For this reason, we may consider AI in almost the same way that McGilchrist characterizes the left hemisphere. It is possible to have an incredibly powerful tool, accuracy, speed, and abstraction, but not something that is recognizable, and must remain part of a broader cognitive process than distort. Like an isolated left hemisphere, AI, left to its own devices, sets its way towards fantasy, delusions, and confusion. Without the right hemisphere contributions that cannot be modeled and continue to be unique human events, our machines will always lack the important ones. This is because the machine relies entirely on pre-digested inputs, symbolic proxy, and statistical correlations that represent reality without revealing it. They can manipulate languages, but it doesn't make sense. They can guess the pattern, but it's not important. And while they may be eye-opening with the flow, they inevitably flatten what they touch. The more you understand this, the higher the risk of recreating the world with that image. I was stripped of my fragmented world, depth, context, and human values.
From this perspective, the recent wave of disillusionment generated by the GPT-5 is less technical than a philosophical wake-up call. The hype of AGI and speculative fantasies of machine consciousness rest on a fundamental misunderstanding of personified projections and the nature of the mind. Dreyfus understood decades ago. Whether complex or powerful, we cannot ignore the world in our context and think like we are ignorant yet human. McGilchrist shows why. The approach to modeling intelligence itself is the product of the hemisphere that is expressed in abstraction, but we cannot see what we cannot understand.
The left hemisphere can build great maps, like AI, but it essentially blinds the topography. When integrating these advanced technologies into our lives and institutions, their contributions must be part of a broader process that reflects humanity. There is Overall. They must be located within the broader horizon of meaning, shaped by the right hemisphere mode of attention that reality is not merely processed or analyzed, but also encountered in all its ambiguous richness. Without proper hemisphere attention, our most powerful machines are at risk of augmenting the vision of the left hemisphere of the already major world, becoming increasingly blind to what we cannot reduce to expression. It makes us human.
