Generative AI is a Race to the Middle

Generative AI, while undeniably powerful, is rapidly becoming ubiquitous. The initial advantage gained from understanding techniques such as prompt engineering is destined to be short-lived as the technology becomes increasingly commoditized.

At its core, generative AI functions as a sophisticated prediction mechanism, drawing upon a vast sampling of available knowledge. It excels at producing statistically probable outcomes—perfectly adequate but rarely exceptional*. Consider a job application scenario: AI can generate a respectable cover letter, but it's unlikely to produce anything truly remarkable. And if everyone else is doing the same, you hardly benefit.

This phenomenon is reminiscent of a study conducted at UMD years ago to determine the average color of web images. The result was beige – a color that is inoffensive yet utterly unremarkable. Similarly, generative AI often produces content that is acceptable but lacks the distinctive qualities that set truly outstanding work apart.

To effectively leverage generative AI, there are two main considerations:

1. Proprietary Edge: To confer genuine advantage, it's crucial to integrate proprietary or specialized data into the AI's knowledge base. This approach can help differentiate outputs from the statistical average. However, it's important to note that even with this strategy, completely escaping the averaging effect on factors such as style remains challenging.
2. Strategic Application: Identify areas where the "race to the middle" actually improves overall capability. Generative AI can be particularly effective in elevating performance from subpar to average, but falls short in domains where excellence and originality are paramount.

The implications extend beyond content creation. In fields such as product design, customer service, and data analysis, generative AI has the potential to standardize and improve baseline performance. However, this standardization also presents a risk of homogenization, and a deadlock between competitors that does nothing more than necessitate the bring-up of more power stations. Consider, for example, the current war between AI-powered applicant tracking systems and the AI job application bots.

As generative AI becomes more prevalent, the competitive edge will shift from mere access to these tools to the strategic integration of human expertise and creativity. Those who can effectively combine AI-generated output with domain-specific knowledge and innovative thinking will be best positioned to thrive.

In Cal Newport's recent New Yorker article he looks at how students are using ChatGPT, with this fascinating insight: "The reality is something different and new; less a method to speed up the task of writing and more an approach to reducing its cognitive burden."

For anyone who has used an AI as a brainstorming partner this makes sense, but how organizations measure or implement this benefit is way less clear. The current generation of GenAI yields advantages more akin to the benefits derived in calculation from the pocket calculator, extremely valuable but falling way short of the breathless hype.

The optimistic future for generative AI is not about who possesses the most advanced tools, but rather who can leverage them most effectively while maintaining insight and originality. In the meantime, we are racing hard to the statistical middle, and a landscape where mediocrity is the norm.

--

* Of course, when it isn't hallucinating.

--

This article is part of a series in which I'm sharing hard-won insights from my tech career. Subscribe to Wilder Thoughts to read more, and if you like this post, please share it to your network!