A Baconian approach to the mostly Aristotelian corporate AI. And what that means for your business

In recent months, a recurring theme has emerged in discussions about enterprise AI - the essential difference between Aristotelian and Baconian approaches. This contrast is particularly relevant in the realm of large language models (LLMs), which have demonstrated exceptional capabilities in generating answers. However, these models were not designed with the specific needs of organizations in mind.

LLMs excel at generating coherent responses based on vast amounts of pre-existing knowledge, much like Aristotle's method of reasoning from premises to conclusions. They can reason, compare, summarize, infer, explain, and combine ideas across domains with impressive sophistication. Yet, they lack the ability to observe outcomes, test hypotheses, or revise their operating structures based on real-world feedback.

Bacon, on the other hand, emphasized a different approach. He proposed a learning structure where ideas are formulated, tested against reality, observed the results, revised hypotheses, and repeated the cycle. This iterative process, which Bacon termed the scientific method, allowed for knowledge to accumulate, errors to be exposed, and discoveries to be validated across multiple generations.

The key distinction lies in the existence of loops - a mechanism for feedback and continuous improvement. Today's frontier models of AI, akin to Aristotle's models, are powerful individual engines capable of remarkable inference. However, the next step is not merely to create larger, more capable Arittotles. Instead, the focus should be on building a Baconian structure around these models.

For businesses, this translates to moving beyond the current open-loop system of AI, where a request is made, an answer is provided, and the interaction ends. Instead, companies need to cultivate a Baconian enterprise AI system that treats every action as an experiment, connected to outcomes. This would involve defining clear objectives, measuring the impact of AI recommendations on business metrics such as conversion rates, churn, margins, and customer trust, and adjusting the AI's behavior based on these real-world results.

In essence, the transition from Aristotelian to Baconian AI represents a shift from isolated, answer-generating systems to integrated, outcome-driven enterprises. This evolution is not merely technical, but epistemological, requiring a fundamental rethinking of how AI is built, managed, and utilized within corporate settings.