Ask George Murnane the right way to separate actual synthetic intelligence from a advertising slogan, and he offers you one query: what does the mannequin predict, and what’s its loss perform?
George Peter Murnane has spent greater than three a long time operating asset-intensive aviation companies, 14 of these years as a chief working officer, a chief monetary officer, or each without delay. He’s now chief govt of Jet.AI Inc. (NASDAQ: JTAI) and a director and CFO of AI Infrastructure Acquisition Corp., the blank-check firm that closed an upsized $138 million IPO in October 2025. That résumé sits on the actual junction the place capital, operations, and AI claims collide. It additionally makes him unusually arduous to promote to.
The George Murnane filter: identify the prediction, identify the loss perform
The take a look at is intentionally unglamorous. If an organization can let you know exactly what its mannequin forecasts and what error it’s educated to reduce, the AI might be actual. If the perfect it will probably supply is that the know-how “makes the expertise smarter,” the label is doing work the software program can’t.
That distinction issues extra in aviation than in nearly some other {industry}, as a result of the associated fee base is excessive and the margins are skinny sufficient {that a} small effectivity acquire compounds into actual cash. Murnane’s filter is a manner of routing scarce capital towards the few functions that transfer these numbers, and away from the numerous that solely transfer a pitch deck.
The place AI is actual in aviation: the AOG math
Begin with predictive upkeep, the appliance Murnane considers genuinely precious. The economics aren’t refined. A single aircraft-on-ground occasion can price an operator between $10,000 and $150,000 per hour of downtime, when you add misplaced income, crew relaxation and additional time, passenger re-accommodation, and the scramble to supply a substitute half.
Predictive upkeep assaults that price instantly. By studying sensor information, flight historical past, and upkeep data, the fashions flag a deteriorating element earlier than it fails, which lets an operator transfer an unplanned restore right into a scheduled window. A 2022 Deloitte evaluation cited by Radome Applied sciences estimated that predictive upkeep, correctly carried out, can lower upkeep prices by as much as 30% and scale back AOG occasions by greater than half. Delta lower unscheduled upkeep by greater than 30% utilizing predictive engine monitoring.
The use case is restricted sufficient to outlive Murnane’s query. The mannequin predicts a element failure. Its loss perform penalizes false negatives, the missed failures that floor an plane, and false positives, the unnecessary half swaps that waste a upkeep slot. There’s a quantity on each side of the ledger. Buyers have observed the identical factor: the predictive airplane upkeep market is projected to develop from $5.35 billion in 2026 to $18.87 billion by 2034, a compound annual fee above 17%.
Predictive upkeep shouldn’t be the one utility that clears the bar. Crew scheduling optimized in opposition to duty-time limits has an outlined goal and a tough constraint set written into federal regulation. Dynamic pricing run in opposition to ahead reserving curves predicts demand and optimizes yield. Doc automation in SEC filings and merger diligence has a measurable output and a measurable error fee. Every of those might be described in a sentence that names what’s being predicted. That’s the inform.
The failure modes George Murnane watches for
The other of a loss perform is an adjective. Murnane’s interviews return repeatedly to 2 methods corporations gown up outdated or absent know-how as AI.
The primary is rebranding. A regression mannequin that has been forecasting demand or pricing danger for 30 years will get relabeled “AI” as a result of the time period raises a valuation. The maths is similar forecast it at all times was, repackaged underneath a extra precious label.
The second failure mode is extra present and costlier. An organization bolts a big language mannequin onto a workflow with out redesigning the workflow beneath it. The result’s a chatbot marginally extra eloquent than the FAQ web page it changed, bought as a metamorphosis. The mannequin is actual, however the worth shouldn’t be, as a result of nobody re-engineered the method the mannequin was supposed to enhance.
Murnane’s warning right here is partly reputational arithmetic. As he has put it, the credibility price of overclaiming compounds sooner than the advertising profit. For a public firm whose identify carries the letters “AI,” that isn’t an summary danger. Overstate what the software program does, and the primary product failure underneath stress turns into the story.
How Jet.AI makes use of AI the place the worth is measurable
Jet.AI offers Murnane a spot to use his personal take a look at in public. The corporate’s software program, constructed when it operated as a private-aviation platform, concentrated AI on capabilities with a quantity hooked up: reserving optimization, matching prospects to the correct operator, and buyer communication. Its CharterGPT app and the Ava agentic reserving mannequin used natural-language processing to compress a reserving course of that after ran on telephone calls and faxes.
These instruments deal with a intentionally slender set of jobs. Flying the airplane, vetting an operator’s security document, and resolving a mechanical failure at midnight stick with people and regulators. The AI sits the place its prediction is reasonable to measure and its errors are low-cost to appropriate, which is strictly the place Murnane argues it belongs.
From reserving software program to AI information middle infrastructure
Essentially the most telling utility of the loss-function take a look at is the one Jet.AI is now residing by way of. The corporate has moved away from operating plane and towards constructing AI information middle infrastructure, describing itself as a know-how firm centered on information middle improvement throughout North America, with tasks spanning greater than a gigawatt of deliberate capability.
The rationale behind the pivot reads like a case research in Murnane’s personal self-discipline. Jet.AI constructed real AI merchandise, together with a big language mannequin agent for personal aviation, then ran right into a constraint that no quantity of promoting may repair: unreliable uptime for the computational assets these merchandise relied on, which sometimes slowed the corporate’s skill to serve prospects. The bottleneck sat under the algorithm, within the energy, land, and compute the merchandise ran on.
So the corporate went after the bottleneck. Based mostly in Las Vegas, with entry to land, solar energy, and pure gasoline, Jet.AI signed a letter of intent for a 50-megawatt venture on a 120-acre campus with room to scale towards a full gigawatt. The framing Government Chairman Mike Winston used was nearly a rebuke of the class’s typical rhetoric: the transfer “isn’t a flashy transfer, however it’s a wise one,” as a result of information facilities are “the bedrock of the AI economic system” and create worth that’s “tangible, secure, and significant.”
That’s the loss-function take a look at pointed at infrastructure quite than software program. The prediction is simple: compute demand retains climbing, and the belongings that provide it earn in opposition to it. The error is measurable in megawatts delivered and uptime maintained. There’s a quantity on each side.
Why the loss-function take a look at travels
Murnane’s filter works as a result of it’s industry-agnostic. It ignores whether or not a know-how seems to be spectacular and asks as a substitute whether or not anybody can state what the system is optimizing and examine the consequence in opposition to actuality.
That self-discipline is the through-line of his profession, from pricing plane belongings at world carriers to evaluating a knowledge middle SPAC. The identical query that exposes a relabeled regression mannequin additionally exposes an overhyped acquisition goal. In each circumstances, the chief who can describe the target perform is working from proof. The one reaching for “smarter” and “seamless” is working from hope.
For a sector the place roughly each firm now claims an AI technique, the worth of a one-question display screen is that it’s quick and arduous to pretend. Title the prediction. Title the loss perform. If these two solutions are particular, the know-how is probably going doing actual work. In the event that they dissolve into adjectives, the one factor being optimized is the advertising.
