Author Eric Siegel Separates AI Fact from AI Fantasy

Winning at artificial intelligence requires companies to stop dreaming and start deploying, Siegel argues.

By Adam Stone

By Adam Stone January 19, 2024

It’s been just over a year since November 2022, when ChatGPT burst onto the scene and ignited a gargantuan hype cycle around artificial intelligence and machine learning in general, and around generative AI in particular. Since then, Eric Siegel — the bestselling author of Predictive Analytics: The Power to Predict Who Will Click, Buy, Lie or Die — has seen some AI projects fly and many others flounder.

A former Columbia University professor, Siegel’s latest book is The AI Playbook: Mastering the Rare Art of Machine Learning Deployment. In it, he observes that ambitious AI and machine learning projects all too often stall at the starting gate.

“In a nutshell, they don't launch. They don’t deploy,” Siegel told The Forecast in an interview. “The data scientist puts together a viable model, but then organizational stakeholders get cold feet.”


Seeing AI’s Impact on Enterprises

If they want to succeed, organizations that spent 2023 salivating over the potential of generative AI and other, future flavors of artificial intelligence must ask themselves why so few AI dreams turn into actual AI deployments. And then, Siegel said, they must follow his advice for turning machine learning fantasies into business-building realities.

What Goes Wrong

The potential use cases for generative AI are numerous and diverse. For example, AI-generated suggestions can improve productivity substantially for new employees, Siegel said. With proper governance, organizations that communicate frequently or produce a lot of content can use generative AI to accelerate their work. Individuals can use it to craft bespoke images in support of presentations, and businesses can use it to design products.

In short: There’s a lot of potential.

So, why do efforts stall? Why do key stakeholders balk when it comes time to launch a machine learning deployment?

Often, the problem is more organizational than technological, according to Siegel. In a common scenario, the machine learning model is able to generate useful outputs, “but it doesn’t necessarily align closely enough with business objectives,” he said.


IT Leaders Get AI-Ready and Go

When business-line leaders don’t know how the generative AI or AI-supported prediction is going to improve their operations, innovation falls flat. For a machine learning-supported business function to launch, “the stakeholder has to understand exactly how this thing is going to deliver value,” Siegel stressed.

Getting It Right

Organizations need a disciplined approach in order to reap the benefits of machine learning, Siegel suggested. Generative AI isn’t just a technical proposition, he said; it’s an offer to elevate a specific business function. Getting that end result requires a team effort.

If the IT team and data scientists simply hand over a solution — telling people how to use machine learning without having first engaged stakeholders in articulating the problem to be solved — that project will go nowhere fast.

Creating an algorithm that fulfills a need “is everyone’s job,” Siegel said. For stakeholders to get excited enough to want to put machine learning into production, “they’re going to have to help form the details. They have to get involved collaboratively with the data professionals from end to end, from the inception of the project to its culmination with deployment.”

In his book, Siegel describes a six-step approach to running a machine learning project, which he calls bizML:

  1. Value: Establish the deployment goal; define the business value proposition.
  2. Target: Establish the prediction goal; define exactly what the model will predict.
  3. Performance: Establish the evaluation metrics.
  4. Fuel: Prepare the data.
  5. Algorithm: Train the model.
  6. Launch: Deploy the model.

There’s a specific place in all this for IT leaders and software developers.

The people on the technical side “are supporting the development and ultimately the deployment of the model,” typically through the implementation of machine learning operationalization or MLOps, Siegel said. The term refers broadly to “all the tools and techniques to set you up for successful deployment. It’s the organizational techniques, the organizational paradigm, practice or playbook.”


9 Predictions for IT in Age of AI

In addition to focusing on those technical capabilities — “the number-crunching,” as Siegel put it — the technology team must think about the practical application of the machine learning steps.

Too often, “it’s an afterthought because the focus is on the analytical technology,” Siegel continued. “It’s like being more excited about the rocket science than the actual launch of the rocket. To the degree that we also care about launching this rocket, we need IT professionals to be thinking about how it will improve operations.”

Enter the Software Developers

Improving operations typically requires AI solutions to interface with existing systems, databases and technology-driven processes. Software developers “are going to be the ones who need to integrate a change to how that system works,” Siegel said.

With AI, “we’re going to be changing things with an injection of science, with predictions from a model. That is going to directly inform which transaction to audit, or which ad to display, or whatever the operation is that you’re improving,” Siegel continued. The developers “need to be involved from the get-go, because they need to set out feasibility and requirements and timelines.”

When it comes time to deploy an AI model, “the most consequential, impactful projects are going to require change in some existing system, and therefore software is going to have to change,” Siegel said. “Software developers need to be central to that conversation.”

Hype vs. Reality

One year into the generative AI boom, Siegel is encouraging business leaders to be realistic about their expectations.


Generative AI Propels IT Modernization

So far, generative AI has performed well when it comes to answering simple questions. With editorial oversight, it can write credible first drafts to help accelerate content creation. It’s great at making predictions based on more data than humans could ever possibly analyze by themselves.

“The hype, meanwhile, says we are on a path to AGI — artificial general intelligence, where systems will be able to autonomously complete not just tasks, but jobs. You can install it and let it riff autonomously, just the same as onboarding a human employee,” Siegel said.

If that sounds exciting, take a beat: It’s not happening, according to Siegel. Nor have we yet seen the killer app that will bring AI to life at the push of a button. Rather than wait for that killer app, or dream of an AGI future, Sigel suggested that business leaders today must get real about the future of AI.

“First, they need to ask things like: Is it even worth doing? Will it help the organization build value?” he said. “You need to have a very specific notion of how this technology is going to improve some existing process, hopefully making it more efficient and effective.”

To seize the potential, “that's where the focus needs to be,” Siegel concluded. “What’s the end goal? What’s the end game? When you get concrete like that, then you are setting yourself up to successfully measure that improved performance, to measure that value.”

In the coming year, that’s where the AI and machine learning rubber will hit the road.

Editor’s note: Learn more about the Nutanix platform for AI, including Nutanix GPT-in-a-Box, a full-stack software-defined AI-ready platform designed to simplify and jump-start your initiatives from edge to core. 

Adam Stone is a journalist with more than 20 years of experience covering technology trends in the public and private sectors.

© 2024 Nutanix, Inc. All rights reserved. For additional legal information, please go here.