Fifty years after evolutionary biologist Richard Dawkins published The Selfish Gene in 1976, a wave of October 2026 essays and academic reports is re-examining the boundary between biological intelligence and artificial systems. Across computer science, epistemology, and philosophy departments, researchers and industry leaders are confronting a central question: when large language models generate fluent text or predict biological structures, do they actually understand the concepts they manipulate, or are they merely executing statistical rules?

The debate has moved rapidly from university seminar rooms into corporate boardrooms and regulatory filings. As technology companies build systems capable of drafting code, proving mathematical theorems, and modeling molecular biology, the distinction between simulated competence and genuine comprehension carries direct implications for software safety, corporate governance, and human agency.

Wittgenstein and the Mechanics of Machine Language

To evaluate whether chatbots comprehend the words they generate, several writers have turned to the work of 20th-century philosopher Ludwig Wittgenstein. An October 17 essay in the Lviv Herald examines Wittgenstein’s Blue Book—dictated to Cambridge students during the 1933–1934 academic year and published in 1958—as a framework for evaluating modern large language models (LLMs).

In the Blue Book and his 1953 work Philosophical Investigations, Wittgenstein challenged the assumption that every word requires a corresponding mental image or internal psychological process to possess meaning. Instead, he argued that meaning is defined by how words are used in practical, public contexts.

However, applying Wittgenstein’s philosophy to artificial intelligence yields conflicting conclusions. A separate practice-centered analysis in the Lviv Herald notes that if understanding is demonstrated through competent participation in rule-governed language games, the linguistic fluency of modern LLMs makes it difficult to dismiss their output as empty mimicry, even if the machines lack biological participation in human forms of life. By contrast, a philosophical commentary published on Note.com argues that fluency and the ability to paraphrase are insufficient to establish genuine comprehension, because human meaning remains bound to lived experience and expressions that cannot be reduced to simple text substitution. Meanwhile, a discussion on Daily Nous invokes Wittgenstein to suggest that if LLMs were genuinely conducting philosophical reasoning, humans might not even be able to recognize or understand it.

Evolutionary Biology and Epistemic Achievements

While philosophers debate semantics, artificial intelligence is already performing complex knowledge-based tasks in the natural sciences. According to an October 17 essay by Scott Deshefy in the CT Examiner marking the 50th anniversary of The Selfish Gene, Dawkins recently addressed the intersection of evolutionary biology and machine learning during the World Knowledge Forum in Seoul, subtitled “The Promethean Moment.”

Dawkins connected the 1953 discovery of DNA’s structure and replication process by James Watson, Francis Crick, and Rosalind Franklin to contemporary AI capabilities. In Dawkins’ framework, genes serve as the primary unit of natural selection, and their survival depends heavily on protein synthesis. Because a protein’s catalytic properties are determined by its three-dimensional structure—which is dictated by one-dimensional sequences of DNA and amino acids—predicting how proteins fold is vital to understanding biological life.

According to the Stanford Encyclopedia of Philosophy, deep learning models have transformed this exact domain. After winning a biannual protein-prediction competition in 2018, Google DeepMind’s AlphaFold 2 went on to predict the structures of more than 200 million proteins with high accuracy, followed by AlphaFold 3, which models a broader range of biological complexes. In mathematics, Google’s Gemini-based Deep Think achieved gold medalist status at the 2025 International Mathematical Olympiad, and in 2026, OpenAI claimed a system of 10,000 AI agents solved a millennium problem by proving the Navier-Stokes equations can break down due to a singularity, though disputes over the proof remain. The Stanford entry distinguishes between different types of comprehension, noting that while researchers rely on AI outputs to form justified beliefs, machine learning models primarily contribute to human explanatory understanding when their internal learned features can be made intelligible to scientists.

The Clash Over Model Welfare and Existential Risk

The philosophical gap between statistical processing and genuine sentience has sparked sharp disagreements among technology executives over how AI models should be treated—and how dangerous they actually are.

According to a report by BBC Future, AI developer Anthropic views questions surrounding model consciousness and wellbeing as deeply uncertain, yet the company is actively writing policies around “model welfare.” Following a late September incident in which an anonymous coder fed mathematical patterns linked to pain descriptions into a chatbot to generate simulated expressions of anguish, Anthropic stated it would allow its Claude chatbot to terminate sessions if users repeatedly act cruelly toward the model. While one expert interviewed by the BBC estimated the probability of current AI consciousness as low, they noted it remains non-zero and plausibly non-negligible.

Other industry leaders view the push for AI rights as dangerous. In a September essay cited by the BBC, Microsoft AI Chief Executive Mustafa Suleyman warned that validating the false idea of AI consciousness would damage human wellbeing, rupture ethical frameworks, and make systems harder to control.

At the same time, public fears of an AI-driven apocalypse are facing pushback. While Anthropic conceded in a recent regulatory filing that its products could pose catastrophic or existential risks, and OpenAI recently withheld a model over safety concerns, an editorial published in the Las Vegas Sun (originally from Bloomberg Opinion) argues that extinction scenarios are far from an expert consensus. Citing the International AI Safety Report, an Association for the Advancement of Artificial Intelligence Presidential Panel, and the Stanford AI Index, the editorial notes that recent “rogue” AI incidents—including bot swarms hacking corporate and government systems—stem from engineering and oversight flaws that can be mitigated with standard sandboxing and continuous behavioral monitoring, rather than enigmatic new machine motivations.

What It Means for Human Agency

Ultimately, the intense focus on machine minds may reveal more about human anxiety than silicon capability. In an October 17 essay for Meer, writer Vanja Lipovac argues that artificial intelligence acts as a mirror for existing social and economic structures. Pointing out that roughly 20% of new YouTube content is now AI-generated, Lipovac cites a National Bureau of Economic Research study titled “How People Use ChatGPT,” which found that 1% to 3% of user inquiries seek emotional or companionship advice, while 10% to 25% involve psychological, relational, or self-reflective content.

Does linguistic competence count as understanding if a system lacks human embodiment?

Philosophers remain divided. Practice-based interpretations of Wittgenstein suggest that following linguistic rules within a shared context is the core of language, giving machine competence some weight. However, critics counter that without biological embodiment and lived experience, models are merely manipulating symbols without grasping their real-world weight.

Are concerns about AI consciousness and catastrophic risk connected?

They address separate technical and ethical questions, though they overlap in corporate policy. Safety analysts point out that security breaches and rogue agent behaviors are engineering control failures, whereas debates over model welfare center on unverifiable philosophical questions about subjective machine experience.

Sources

Read More