Originally published in Forbes
Recently on The Dr. Data Show, my co-host Luba Gluhova and I dug into the evolving discourse surrounding artificial general intelligence – and its stubborn incoherence. A recent publication by the venture capital firm Sequoia Capital projected the arrival of AGI by 2026, defining the concept simply as “the ability to figure things out.”
From an engineering and practical standpoint, this definition effectively only reiterates traditional, highly subjective definitions of AI. It is basically another way to say “capable of reasoning,” which has long been a common, if circular, attempt to define AI.
AGI was meant to differentiate from AI. It was originally supposed to signify the “whole enchilada” – a virtual human capable of doing anything that a person can. This would make such a system fully autonomous, capable of matching human performance across a wide range of tasks – effectively operating as a virtual employee.
However, as the profound technical challenges of achieving supreme autonomy become apparent, definitions of AGI shift toward more subjective criteria, blurring the distinction between AGI and AI, which itself has always faced the existential problem of being undefinable. This shift appears to be an exercise in waving hands in order to dodge criticism.
AI hype often emphasizes supreme autonomy – and most of the hype at least implies it. The notion of AGI is a natural extension of that aspect of the hype: If a system can do everything a person can, it needs no human in this loop.
While genAI possesses remarkable capabilities and offers substantial commercial value, it currently faces a critical reliability challenge. In automated enterprise workflows, such as customer service interactions or healthcare claims processing, a minor error rate – even as low as 5% or less – can render a fully autonomous system non-viable due to the operational risks of factual inaccuracies, ethical missteps or mishandled transactions.
So, how do we sober up and pursue feasible deployments that realize the potential value of these technologies? The answer is hybrid AI.
Hybrid AI offers a practical alternative to pursuing the ever-elusive ideal AGI. GenAI hallucinates and exhibits other unacceptable behaviors that preclude its deployment – especially for its more ambitious intended uses, such as performing the role of customer service agent, analyst, educator or all-capable virtual assistant. Rather than supreme autonomy, a feasible route to leveraging genAI and pursuing its more ambitious uses is to hybridize it with predictive AI.
Here’s how hybrid AI works: Machine learning models serve as a vital reliability layer. By assigning probability-based risk scores to the outputs of generative models, predictive AI can systematically identify the specific cases with the highest likelihood of failure or behavioral error. These high-risk cases are then routed to human operators for review. This judicious inclusion of a “human-in-the-loop” mitigates the operational risks associated with large language models while successfully automating a significant portion of the workload.
Hybrid AI is already moving from theory to practice. A diverse array of industry leaders – including Netflix, Amazon, JPMorgan and Microsoft – are actively deploying these hybrid systems (they’ve lined up to speak on the topic at the conference I chair, HYBRID AI 2026). This empowers businesses to navigate the limitations of genAI and deploy reliable, valuable semi-autonomy.
The move to hybrid AI represents a sobering up, an evolution from unrealistic elation about supreme autonomy. The intoxication hinges on a misbelief, a hope and a prayer, that LLMs will somehow evolve into “virtual humans.”
Instead, we need to take the pressure off these models – dispense with the unrealistic performance expectations – and judiciously leverage what they feasibly can do by pairing them with the predictive safeguards they desperately need in order to achieve launch-worthiness.
About the author
Eric Siegel, Ph.D., is a former Columbia University professor who helps companies deploy machine learning. He is the cofounder and CEO of Gooder AI, the founder of the long-running Machine Learning Week conference series, the instructor of the acclaimed online course “Machine Learning Leadership and Practice – End-to-End Mastery,” executive editor of The Machine Learning Times, and a frequent keynote speaker. He wrote the bestselling Predictive Analytics: The Power to Predict Who Will Click, Buy, Lie, or Die, which has been used in courses at hundreds of universities, as well as The AI Playbook: Mastering the Rare Art of Machine Learning Deployment. Eric’s interdisciplinary work bridges the stubborn technology/business gap. At Columbia, he won the Distinguished Faculty award when teaching the graduate computer science courses in ML and AI. Later, he served as a business school professor at UVA Darden. A Forbes contributor, Eric publishes op-eds on analytics and social justice.
Eric has appeared on Bloomberg TV and Radio, BNN (Canada), Israel National Radio, National Geographic Breakthrough, NPR Marketplace, Radio National (Australia), and TheStreet. A Forbes contributor, Eric and his books have been featured in BBC, Big Think, Businessweek, CBS MoneyWatch, Contagious Magazine, The European Business Review, Fast Company, The Financial Times, Fortune, GQ, Harvard Business Review, The Huffington Post, The Los Angeles Times, Luckbox Magazine, MIT Sloan Management Review, The New York Review of Books, The New York Times, Newsweek, Quartz, Salon, The San Francisco Chronicle, Scientific American, The Seattle Post-Intelligencer, Trailblazers with Walter Isaacson, The Wall Street Journal, The Washington Post, and WSJ MarketWatch.
