Oracle's Deep Trust AI Architecture Runs App Definitions Through APEX

Oracle’s Deep Trust AI Architecture routes AI-generated application definitions through its APEX runtime, placing software, data, data access and application development within a four-layer framework.

Oracle unveiled the architecture during Oracle AI World in Las Vegas on Oct. 26. Its four areas are the software platform, data, data access and application development.

In the application-development layer, the APEX AI Application Generator turns natural-language descriptions of applications into APEXlang definitions. Oracle APEX runs those definitions in a centralized runtime. The generator produces an application definition rather than executable application code, so the definition is what APEX runs.

APEXlang is an open, human-readable language for defining APEX applications. It gives the generator a format to produce from a natural-language request, while Oracle APEX provides the runtime that executes the resulting definition.

The architecture also includes database controls. Deep Data Security applies end-user permissions when Oracle AI Database processes SQL requests. Oracle SQL Assertions enforce business rules, placing data access and business requirements among the controls in the framework.

“AI trust is the biggest challenge we face today,” said Juan Loaiza, Oracle’s executive vice president of AI Database Technologies.

APEXlang predates the architecture announcement. Oracle introduced the language in May as an open, human-readable way to define APEX applications. Mike Hichwa, Oracle senior vice president of software engineering, compared its output to a condensed version of an application: “It’s like a CliffsNotes. It’s not the full book.”

Oracle announced general availability for APEX 26.2 on-premises deployment on Oct. 6. The release included file-level imports, allowing developers to import individual page or shared-component .apx files.

Former Google CEO Eric Schmidt has also discussed the need for developers to direct and review AI-generated work. His comments on directing and reviewing AI-generated work touched on how coding knowledge can help people assess what AI produces.