The trillion-dollar AI story is about autonomy. The real value — in fraud detection, security triage, and risk scoring — is hiding in the AI almost nobody is talking about.
Most machine learning models built to be deployed never get deployed. Meanwhile, budgets pour into generative AI pilots that dazzle in the demo and stall in production. Data scientist and author Eric Siegel estimates the real value of today's AI may be closer to 5% of what current valuations imply- still enormous, but a lot of disillusionment waiting to happen.
If you lead security, GRC, compliance, or data teams — or you've been told to "do AI" without a clear business case — this conversation is your reality check.
📖 Eric's book: The AI Playbook: Mastering the Rare Art of Machine Learning Deployment.
Chapters
00:00 Predictive and generative AI as complementary tools
02:33 What predictive AI does that generative AI does not
06:32 How hype around intelligence and agents creates scope creep
09:16 Risk scores and human review for more reliable automation
11:22 Why model accuracy is not the same as business value
13:30 BizML’s six-step path from project idea to deployment
16:08 The second wow: proving operational impact
19:20 AGI claims, uncertainty, and the limits of prediction
23:14 AI investment, business sustainability, and the bubble
27:08 Why predictive and generative AI work better together
31:35 Ethics, protected classes, and machine bias
37:20 Why hiring models cannot repair workplace culture
39:08 How to interpret probabilities without mistaking them for certainty
SOURCES & REFERENCES
• The AI Playbook by Eric Siegel → https://mitpress.mit.edu/author/eric-siegel-39891/
• Gooder AI (Eric's startup) → https://gooder.ai
• Eric's op-eds on discriminatory models → https://civilrightsdata.com
• Cory Doctorow, coiner of "enshittification" → https://pluralistic.net
• ProPublica, "Machine Bias" → https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing
• EEOC guidance on AI in hiring → https://www.eeoc.gov/sites/default/files/2024-04/20240429_Employment%20Discrimination%20and%20AI%20for%20Workers.pdf
• Nate Silver's 2016 election forecast → https://www.youtube.com/watch?v=nVNEz0W-Ud4
THE STORY
▶ The first "wow" is easy. The second is rare.
AI pilots knock everyone's socks off — and then die before deployment. Siegel argues that teams obsess over technical benchmarks and lose sight of business outcomes: "It's like being more excited about the rocket science than the launch of the rocket."
▶ AGI and the autonomy sales pitch
Using a chatbot with a human in the loop is valuable. But that's not what's being sold. "AGI is, by definition, the ultimate autonomous story," Siegel says — and that's where the hype outruns the technology.
▶ The mirror problem
Strip out a protected-class column and a model can still discriminate through proxies. Siegel revisits ProPublica's recidivism investigation, which found higher false-positive rates for Black defendants. Justin shares how a hiring model exposed a company quietly weeding curiosity out of its own culture.
🔐 SecureTalk is brought to you by Strike Graph — compliance and security that scales → https://strikegraph.com
ABOUT
Guest: Eric Siegel is a 35-year machine learning veteran, a former Columbia University professor, founder of Machine Learning Week, co-founder of Gooder AI, and author of The AI Playbook and Predictive Analytics.
Host: Justin Beals is the founder and CEO of Strike Graph and a career builder of machine learning products in education, hiring, and security.
🔔 Subscribe for weekly conversations on cybersecurity, compliance, and the tech reshaping risk.
#AIhype #PredictiveAI #MachineLearning #Cybersecurity #AIgovernance