Next Role for AI Agents: Recommending and Acting on Real-Time Choices

PinIt

Use of AI agents is transitioning from systems that learn from decisions to systems that learn to improve the decision environment itself.

Artificial intelligence is increasingly making autonomous decisions for businesses in terms of market actions, customer relations, and supplier relations. However, AI alone can never be making the final decisions. That means it’s time to consider the next step: with AI agents not just presenting alternatives or likely outcomes, but helping to actually make intelligent choices based on human and machine collaboration.

“In a world of hyper complexity, exponential advances in AI capabilities, and compounding uncertainty, strategic value no longer comes from human decision-making alone. It arises from underlying decision environments, concludes a recent study published by MIT Sloan Management Review and underwritten by Tata Consultancy Services.. The study’s authors call this supportive framework “intelligent choice architectures.”

This is where the true value of AI emerges. Without such an architecture, it is not scalable, relegated to siloed applications and proofs of concept. But how does one go about designing, building, and deploying such an architecture? The authors define intelligent choice architectures as “dynamic systems that combine generative and predictive AI capabilities to create and refine choices and present them to human decision makers.”

In addition, such systems are capable of actively generating novel possibilities, learning from outcomes, and seeking information. In other words, AI systems, or perhaps AI agents, that can go out and rustle up all available data and present choices to their human co-workers. And these AI systems aren’t just serving up information, they’re actively collaborating with humans to arrive at the best solutions to problems.

Such systems can provide decision makers with insights into potential outcomes for each option in real time, helping decision-makers “weigh trade-offs and risks more effectively. For example, a retail manager assessing inventory decisions might see not only the immediate costs but also the projected downstream impacts on sales, supply chain dependencies, and seasonal trends. This predictive foresight helps decision makers align their choices with longer-term strategic goals rather than just short-term gains”.

Such intelligent choice architectures are “a decisive break from conventional uses of AI to support decision frameworks,” the MIT Sloane report claims. “Combining generative and predictive AI transforms artificial intelligence from a decision aid to a collaborative choice architect that better empowers human decision-making.”

The study’s authors cite several prominent examples of an intelligent choice architecture in action:

  • Walmart’s HR team uses an intelligent choice architecture “as one aspect of identifying talent in local stores, expanding options for developing its internal management team.”
  • Liberty Mutual integrates intelligent choice architectures “into claims processing, enabling adjusters to explore scenario-based alternatives informed by historical outcomes and strategic negotiation models.”
  • Cummins explores how to use generative AI to simulate thousands of edge-case scenarios in powertrain design, demonstrating how “intelligent choice architectures can expand the design space, improve resilience, and reduce time to market.”

See also: Taming AI Agent Sprawl in Industrial Organizations

Infrastructure for AI agents

While promising, implementing intelligent choice architectures may require investments of resources and budget, the MIT Sloan co-authors caution. “These systems are not trivial to implement, given that they require sustained investment in data infrastructure, cross-functional talent, change management, and organizational design. Most legacy companies still struggle with fragmented data environments and siloed decision processes — foundational gaps that must be addressed before ICA adoption at scale is viable.” In addition, these systems need deep training on human logic and intents.

Still, such systems represent a “transition from systems that learn from decisions to systems that learn to improve the decision environment itself.” And, tellingly, one of the executives interviewed for the study indicated that his company “stopped separating IT, OT, and AI. It’s all decision infrastructure now.”

Avatar

About Joe McKendrick

Joe McKendrick is RTInsights Industry Editor and industry analyst focusing on artificial intelligence, digital, cloud and Big Data topics. His work also appears in Forbes an Harvard Business Review. Over the last three years, he served as co-chair for the AI Summit in New York, as well as on the organizing committee for IEEE's International Conferences on Edge Computing. (full bio). Follow him on Twitter @joemckendrick.

Leave a Reply

Your email address will not be published. Required fields are marked *