AI's hidden learning trails raise a power question: who owns the intelligence economy, asks Gordon Ritter | Fortune

The contest over intelligence

Ritter says companies should count the traces people leave across systems, not just the jobs AI may replace. These include judgment calls, corrections and solutions that are not written in manuals.

He compares the emerging dispute over human intelligence with the early fight over radio frequencies. Broadcasters claimed frequencies before legislation caught up; the Radio Act of 1927 and Communications Act of 1934 later established rules.

Ritter warns that treating human reasoning as a resource to extract could weaken the source of future advances. If rules are not shaped soon, he says, companies with the strongest models may dictate terms to others.

Choice for AI platforms

Ritter’s proposed social contract has three layers: platform, organization and individual. At the platform level, he calls for choice and argues that no single model should become unavoidable.

He says organizations’ lasting advantage lies in the learning processes they build on top of models, rather than in any one model. Yet providers can make switching difficult and bind organizations to their first choice. Access to AI, Ritter argues, should not depend on giving up intelligence as a condition of service.

Organizations and expertise

At the organizational level, Ritter says learning loops, workflows and specialized knowledge created around AI should belong to the organization. He points to RedCompass Labs, which uses Cognitive Task Analysis to turn thousands of hours of expert work into datasets for its Payment Expert Agent.

Employees using public AI tools at work may teach an outside system how their organization operates. Ritter notes that enterprise contracts often restrict training on internal data, while safeguards for behavioral traces and process knowledge are less settled.

He distinguishes routine work, which can remain transparent, from unusual acts of judgment that change results. Examples include a salesperson departing from a script or an engineer finding a fix absent from a runbook.