Vibe reporting at MIT
We deserve better than a vague, self-serving, oversimplifying, unserious, overhyped “report” about the 95% failure rate of AI pilots in organizations (@kervin Werbach has a great post about this marketing document masquerading as a report, link in comments). Ill-defined (or undefined) concepts, opaque quantification and sweeping generalizations notwithstanding, the document's central message, that enterprises are struggling to realize value from AI, resonates with many practitioners. It may be 65% or 95%, there may be varying degrees of “failure”, but we know what happens to pilots that can’t scale for the myriad reasons that experimental projects die in organizations.
That is why we deserve better: understanding failure modes and their context is more important than an arbitrary binary boundary on a continuum amplified by an arbitrary percentage. Is there anything specific to AI pilots that may (or may not) make them more failure-prone? More than data pilots? More than [insert your favorite experimental nightmare]? If so, what is it? What can we do about it that does not necessarily involve another layer of AI magic? Agentic AI, as much as I can see its potential, is certainly not going to help increase the probability of success of pilots: if anything, it makes everything more complicated and harder to scale.
Amplified without critical thinking by Fortune Magazine: link
Self-Fulfilling Tragedy of Algorithms
The Self-Fulfilling Algorithm
Switching costs, investment/pain -> religion, polarization
A speciation event in the US
Asymmetry of consent
Echo chambers
Doxing, flaming, shaming (Gamergate), dog piling
Harassment, conspiracy theories, mis/disinformation, bandwagon effect, fans
Super useful framework from NfX