False perceptions
On the coronary heart of open-washing is a distortion of the ideas of openness, transparency, and reusability. Transparency in AI would entail publicly documenting how fashions are developed, educated, fine-tuned, and deployed. This would come with full entry to the information units, weights, architectures, and decision-making processes concerned within the fashions’ building. Most AI corporations fall in need of this degree of transparency. By selectively releasing components of their fashions—usually stripped of key particulars—they craft an phantasm of openness.
Reusability, one other pillar of openness, is far the identical. Corporations permit entry to their fashions through APIs or light-weight downloadable variations however stop significant adaptation by tying utilization to proprietary ecosystems. This partial launch affords a calculated degree of reusability that maximizes massive cloud’s worth extraction whereas minimizing the chance of opponents.
For instance, OpenAI’s GPT fashions are accessible, however their integrations are invariably tied to particular net purchasers, upkeep libraries, and functions owned by the corporate. Enterprise builders don’t obtain free rein to regulate, adapt, or redistribute these fashions, which runs afoul of licensing agreements. One developer buddy of mine put it greatest when he stated, “These things is about as open as a financial institution vault.”
