Blog
There's No Fairy Dust in AI
By Teasha Cable
What the inventor of transfer learning needs you to know...
I don't introduce guests lightly, but for this one, I went all in on the show, because, the future of decision intelligence runs straight through her.
My guest was Dr. Lorien Pratt, my mentor, my OpenDI.org co-founder, and, by any honest accounting, the original architect of the field I've built my career on.
Our conversation on the Dynamic Decisions Podcast gave me a rare chance to sit with the person who built the foundations of modern AI and ask her why she thinks the industry keeps missing the point.
That framing isn't hype.
Back in the early 1990s, as a tenure-track computer science professor at the Colorado School of Mines, Lorien invented transfer learning, the technique that lets a neural network reuse what it already learned to master a new task faster.
It's the idea sitting quietly underneath today's large language models and computer vision systems. She holds three computer science degrees, including a master's in AI from 1988 and a PhD in AI from 1992...back when "AI" barely showed up on anyone's radar.
She won an NSF Career Award and sat on the NeurIPS program committee. By any measure, she was already one of the most accomplished scientists of her generation, and then she walked away from pure machine learning to chase something bigger: decision intelligence, the subject of her Decision Intelligence Handbook and years of writing ever since.
Lorien is prolific.
The Gap Nobody Wants to Talk About
Lorien's big pivot came after she left academia to become a technology analyst. She interviewed hundreds of senior leaders every year about how they "actually" used tech.
What she found stunned her: many couldn't spell AI, didn't want spreadsheets in the room, and made multi-billion dollar calls in five minutes flat.
The most sophisticated machine learning in the world was hitting a wall long before it ever touched a real decision. She calls it the last mile problem, and honestly, it's the same gap I see in almost every organization I work with today.
One story from our conversation makes the point vividly. A trillion-dollar resources company spent ten years and tens of millions of dollars building a digital twin of its operations. By the time Lorien's team got involved, nobody was using it.
The tech was there. The link connecting it to an actual choice was missing, and one missing link breaks the whole chain. Kindly read that twice.
In 2008, Lorien wrote the first paper on decision intelligence and, as she puts it, "labored alone in the wilderness for ten years until she found me."
Today she's chief scientist and co-founder of Quantellia LLC, and together with Dr. David Roberts, we co-founded OpenDI. Our mission? Build an open, community-driven standards organization, the same way the telephone industry standardized on interoperability, or the web standardized on HTML.
Why a GPS is Smarter than a Dashboard
One of the memorable moments in our conversation was Lorien's explanation of why dashboards and KPIs fall short.
A KPI is like a GPS that only tells you there's traffic ahead, instead of turn left in 300 feet. It's informative, but it doesn't tell you what to do next. Decision intelligence uses causal decision diagrams to map how today's choices ripple into tomorrow's outcomes, turning a decision from a gut call into something you can actually engineer.
That mapping also exposes problems leaders often can't see from inside their own department. In one example Lorien shared, a government agency trying to decide which buildings to shut down during the pandemic discovered that another team's decisions were quietly undermining its own, an authority and responsibility mismatch she calls a recipe for organizational insanity.
Keep the Human in the Loop
With agentic AI dominating the headlines, Lorien draws a clear line, and I agree with her completely.
Automation works well for repeatable, high-volume, inside-the-digital-world choices, like fraud detection or content recommendations. It works far worse for one-off decisions with real-world ripple effects, like hiring, contracts, or opening a new location, the kind large language models are, in her words, terrible at because of their pattern-matching nature.
Belief, judgment, and accountability still belong to people.
Lorien was refreshingly candid about her own decision-making style too, scoring herself just 35 percent feeler, 25 percent data driver, and only 20 percent each on planner and challenger when I ran her through a persona quiz live on the show, proof that even the field's inventor is still refining her own instincts.
Want the full conversation? Catch the episode on the Dynamic Decisions Podcast, and find Lorien's books, free resources, and OpenDI's standards at learndi.com.
Learn more about CModel and OpenDI.org