Enabling Real-Time Data for AI Systems

AUTONOMY IN PRACTICE

More importantly, agents are directly responsible for a shift in the very nature of what a “task” is, from one that is predefined and static to one that naturally evolves based on mutable business circumstances, if not user requirements. Such sophistication hints at the very potential of real-time AI—when underpinned by low-latency data reflecting the latest state to an objective. “With an agent, a task is an emergent concept,” Livingstone explained. “You go to ChatGPT; you start typing in the box. If the task starts off with ‘Research what the best pair of jeans are,’ then the task evolves from ‘I found them,’ to a purchase decision. It’s not a concrete thing that’s preregistered. It’s sort of emergent properties of interaction patterns.”

The facility at which agents can take the next step in a workflow based on real-time developments strikes at the epicenter of the autonomy for which many laud them. It also  raises pertinent questions about governance, traceability, and the limits to such autonomy—as well as how to impose them. Conventional methods are predicated on auditing, which requires analyzing events after they take place to see if they adhered to established protocols. Contemporary solutions provide such functionality in addition to eventbased controls that can catalyze actions with negligible, if any, latency.

They’re based in large part on “collecting telemetry of what the agent is doing in real time,” Livingston divulged. “So, as that agent becomes unaligned and starts doing things someone didn’t ask it to do, you can immediately detect that and kick off an autonomous response.” Responses and remediation measures can include revoking an agent’s access privileges, modifying governance policies, and instituting human approvals for agent actions.

DATA DICTATES

With streaming data platforms, time-series databases, other operational databases, and modern governance measures, organizations can avail themselves of a reality in which, in real time, their agents are “invoked by a change in the data itself,” Falconer said.

One can readily extrapolate those capabilities to include everything AI systems can do via real-time data. Enabling real-time data for AI systems is an ongoing effort showing real benefits to organizations.