Earlier than enterprises can deploy AI brokers that really work, they want one thing most organizations don’t have: a single, authoritative supply of fact. Joe DosSantos, VP of Enterprise Knowledge and Analytics at Workday, joins Robb and Josh for a wide-ranging dialog about canonical information, the semantic layer, and why knowledge governance, an idea from the Nineteen Nineties, has immediately develop into important for AI deployment.
The core problem? Giant language fashions are predictive engines that “anticipate what you in all probability would imply,” as DosSantos explains. They’re “fairly good at phrases, however they suck at math.” For B2C purposes the place a number of interpretations are acceptable, this works fantastic. However in enterprise contexts, the place income was precisely $1.625651 billion final yr, organizations want deterministic fact, not probabilistic guesses.
The answer requires three layers: establishing canonical information (the laborious human work of defining what knowledge means in your group), constructing a semantic layer (the translation mechanism between human definitions and machine-readable codecs like YAML), and utilizing the LLM as an interface to deterministic back-end methods slightly than treating AI as the system itself.
DosSantos presents a compelling metaphor: attempting to deploy AI brokers with out this basis is like wanting granite counter tops with out constructing the basis of the home first. You possibly can’t skip it, it’s non-negotiable infrastructure.
The dialog additionally tackles AI anxiousness, with DosSantos referencing Kate Darling’s framework of excited about AI as animals slightly than human replacements, and Robb Wilson proposing we view AI as merely “smarter machines” — ability saws that know the distinction between wooden and fingers, stoves that stop fires, and washing machines that don’t shrink garments.
For leaders evaluating AI investments, this episode clarifies what truly wants to be constructed before brokers can ship worth: not flashy use instances, however the unglamorous, important work of information governance and semantic translation.
Disclaimer: This article is sourced from external platforms. OverBeta has not independently verified the information. Readers are advised to verify details before relying on them.
