Rethinking AI: Beyond Word Prediction

In the discourse surrounding artificial intelligence, a common critique surfaces: “Large Language Models (LLMs) are merely advanced algorithms that predict the next word. They lack true intelligence.” While this skepticism is understandable, it invites a deeper examination of AI’s potential, particularly in a distant future dominated by Artificial General Intelligence (AGI).

Imagine a future—whether 200, 400, or even 1000 years ahead—where AGI exists. If such a form of intelligence is indeed artificial and digital, it will need a method of communication with the world around it. The question arises: What form of interaction could encompass the full breadth of human language and requests for action beyond just word prediction?

It may seem reasonable for an AGI to possess a singular, absolute action it wishes to execute. However, this vision may overlook the complexity of real-world interactions, which often occur as a continuous sequence of thoughts and actions. Rather than merely outputting a single, definitive word or action, an intelligent system might analyze a vast array of possibilities, weighing and selecting from a diverse set of outcomes to communicate effectively.

From my own experience in the field of machine learning, I recognize that the underpinning of any AI, including language models, is indeed mathematics and algorithms. The application of these algorithms, including neural networks—which I have hands-on experience with—indicates their potential to achieve meaningful outputs through statistical inference rather than simplistic prediction.

For those skeptical about the capabilities of current AI models, I pose an important question: What output methods would you deem worthy of an AI? How should these systems interact with humans to differentiate themselves from mere auto-complete features, however sophisticated?

Ultimately, any advanced model requires a means of delivering its results, and predicting the next token or word is as valid a method as any. As we continue to evolve our understanding of AI, it’s essential to consider the vast array of outputs it could offer—not as autonomous beings, but as complex systems navigating the nuances of human communication and interaction. In doing so, we might just uncover the true potential of artificial intelligence.