Select Page

Create narratives with o1. OpenAI has just launched its new artificial intelligence model: Will its alleged new “reasoning abilities” enable creators to produce genuinely original stories and narratives?

As reported by The Atlantic, the new model marks a significant advancement from the widely known ChatGPT-4. While ChatGPT-4 relies heavily on its predictive abilities, o1 is designed to emulate human-like reasoning, making it a more sophisticated tool for various applications.

Until now, writers using AI to craft stories, plots, or to create narratives have faced potential accusations of plagiarism. This concern arises because AI models like ChatGPT-4 generate content based on vast amounts of pre-existing material created by others.

Generative AI, the technology behind models like ChatGPT, learns by scanning enormous volumes of text. It identifies statistical relationships between words and phrases and uses these patterns to predict the next word in a sequence. Theoretically, this power of prediction allows the AI to generate normally coherent and contextually relevant content.

However, predicting the next word is not the same as understanding the text. These models mimic linguistic rules without grasping the underlying phenomena and ideas. They are “stochastic parrots,” as Emily M. Bender, a computational linguist at the University of Washington and prominent critic of the AI industry, argues. They are capable of generating language based on statistical patterns but lacking true comprehension. No chatbot can fully understand the person on the other end of the conversation.

According to investigations from The Information, Bloomberg, TechCrunch, and Reuters, major AI companies are beginning to see the limitations of this predictive approach. Simply ingesting more data does not equate to increased intelligence. The predictive model’s ability to improve diminishes as more data is added.

OpenAI’s o1 represents a shift in strategy. Unlike its predecessors, o1 searches through numerous potential emulated reasoning chains. This model explores different hypotheses much like a human would, akin to a mouse navigating a maze.

In this analogy, the rodent learns the correct path to the cheese by testing various possibilities, most of which are wrong. This trial-and-error method is considered a sign of intelligence, as it demonstrates the ability to learn from mistakes and refine strategies.

The mouse, however, is unaware of its role as an experimental subject. It does not understand that it is part of an experiment conducted by a more intelligent species. This lack of awareness highlights a fundamental limitation: the mouse, much like an AI model, operates without true comprehension of its context.

While o1 can apply grammatical rules through reasoning, it still does not address the function of the words. The meaning behind the words is crucial, and no machine yet understands why they say what they say. It is important not to confuse language with its meaning, as no chatbot can develop a precise understanding of the person they interact with.

The ultimate goal of AI is to become functional, taking action based on its reasoning. This requires an objective, something that AI models are still striving to define and pursue effectively.

If we are to entrust machines with the creation of our narratives, we must rethink the purpose of narration and the motivations of the narrating agent. (Why tell this particular story? What advantage(s) does its narrating agent derive from this narration?) Without this reevaluation, we risk producing repetitive and empty content, devoid of the depth and understanding that true narratives require.

Join the Discussion

About this website   »