Practice · AI & Automation
Enterprise AI is being sold today on claims that were tested, at scale, on real hardware, decades ago. IQ4hire's AI practice is led by someone who ran that test and watched which approach survived contact with the people who had to use it.
The Receipt · Ford SBDS
In the late 1980s David LaRue was business planning lead for Ford's Service Bay Diagnostic System, a program chartered in 1985 and laid out publicly by Len Tedesco in SAE Technical Paper 861030. He wrote the product plans, raised the funding division by division, and ran the testing that decided what shipped. It grew to roughly 800 people — larger than a Ford division.
SBDS put the two competing theories of machine intelligence against each other on the same problem. One was a rule-based expert system written in LISP, built with the Carnegie Group: an expert's knowledge hand-encoded and queried by the technician through a text dialogue. The other was pattern recognition that learned from data, paired with a strategy engine that chose each next test from the result of the last.
The learning approach shipped. The hand-authored rule tree did not scale. Every new case needed another hand-written branch, and the count exploded past any hope of building or maintaining it. That is not a story about a buggy system — it is a paradigm hitting its ceiling, observed in production rather than argued in a seminar.
Ray Kurzweil's The Age of Intelligent Machines (1990) carried an account of SBDS by the Carnegie Group's Jeff Pepper, describing a joint Ford, Carnegie Group and Hewlett-Packard program targeting an installed base of over 5,000 North American dealerships — and suggesting it "may become the largest application of artificial intelligence in the world."
The pedigree of the expert system is worth stating, because it rules out the easy dismissal that Ford simply built it badly. The Carnegie Group was founded in 1983 by four Carnegie Mellon researchers and was one of the four firms generally credited with creating the commercial AI industry. Ford did not merely hire them — it took an equity stake.
Carnegie Mellon is where Herbert Simon and Allen Newell founded artificial intelligence as a discipline. Simon won the Turing Award in 1975 and the Nobel Prize in Economics in 1978. So the rule-based system that lost this bake-off came from the direct intellectual lineage of the people who invented the field, built by the best commercial team available, funded by a company willing to put equity behind it.
The rule-based system still did not scale. That is the point. It failed on its own merits, not for want of talent or money — which is exactly why the result still holds when a vendor tells you their hand-built logic will cover your edge cases.
The program it sat inside did the opposite. Built on David's product plan and the funding he raised division by division, SBDS shipped into roughly 3,500 dealerships against a target of 5,000, in a network of about 5,600 — a share of which were too small to be required to carry a machine at all. Each unit ran around $40,000, taken either as an outright purchase or on a seven-year lease with a $1 buyout. That lease term is what set the replacement cycle: at the end of it the fleet turned over to the successor system, FDS, by which point both David and Len Tedesco had long since moved on.
The learning approach was not a research result. It was a capital-equipment product that dealers paid for, deployed across most of a national network, and ran for the full length of its financing.
The Patents
The diagnostic research team — Kenneth Marko, John James and James Dosdall among them — filed through Ford across 1989–2002. David directed the program the work sat inside; his name is not on the patents. They are listed here because they are checkable, and because they show which road Ford actually took.
| Patent | Title | Inventors | Why it matters |
|---|---|---|---|
| US 5,041,976Granted 1991 | Diagnostic System Using Pattern Recognition for Electronic Automotive Control Systems | Marko · James · Dosdall | The foundational SBDS pattern-recognition patent — the learning road. |
| US 5,044,195Granted 1991 | Misfire Detection in an Internal Combustion Engine | James · Dosdall · Marko | Misfire detection, the diagnostic problem the learning approach was aimed at. |
| US 5,095,742Granted 1992 | Determining Crankshaft Acceleration in an Internal Combustion Engine | James · Dosdall · Marko | Crankshaft measurement underpinning the detection work. |
| US 5,528,931Granted 1996 | Engine Diagnostic Monitor with On-Board Adjustment of Correction Factors | James · Marko · Dosdall | On-board adjustment — the system correcting itself in the field. |
| US 5,531,108Granted 1996 | Adaptive Correction of Torsional and Wheel Profile Position Irregularities for Misfire Detection | L. Feldkamp · Puskorius · James · Marko | Adaptive correction from the Ford neural-network group. |
| US 5,559,285Granted 1996 | Fuzzy Logic Method for Detecting Misfires in Internal Combustion Engines | Bryant · Marko · James | Fuzzy logic in production — the third road, not just rules and networks. |
| US 6,434,541Granted 2002 | Automotive Engine Misfire Detection System Including a Bit-Serial Based Recurrent Neuroprocessor | Tawel · Aranki · L. Feldkamp · Puskorius · Marko · James | The Ford–JPL neural chip: a recurrent neural network in silicon. |
Assignee: Ford Motor Company (Ford Global Technologies for the 2002 filing). Verified against USPTO and Google Patents records.
What It Taught Us
Hand-authored logic grows combinatorially with the problem. Any AI proposal whose intelligence is a person writing branches has a ceiling, and the vendor demo will not show you where it is.
An expert system failed cleanly — ask it something off its tree and it stopped, visibly stuck. A learning model rarely stops. It answers confidently and fluently, and will not tell you when it is outside what it learned. For an enterprise making a decision, that is not a small difference.
Ford's technicians refused the expert system because answering a screen's questions in sequence is a poor fit for an expert doing time-critical work. It was not the intelligence they rejected — it was the interface. The industry has since made that same interface the default.
What shipped was deterministic enough to explain itself: run this test, because the last result showed that. The technician stayed a reasoning partner rather than being led by the nose or handed a verdict. The systems people accept are the ones they can see into and steer.
Since Then
MIT Sloan, 2018. Executive education in strategy, computer science and artificial intelligence.
Buyer side. As CIO and Chief Digital Officer at KEH, built AI applications for item identification, matching and catalog management, and prototyped an AI customer-service agent — production AI inside a working retail operation, not a pilot.
Vendor side. Senior Advisor to Gyrus on AI roadmap, positioning and go-to-market, and Advisor to BlackWillow Research, an AI-native financial intelligence platform. Both are go-to-market advisory roles with early-stage companies.
Neither Gyrus nor BlackWillow Research is in the IQ4hire vendor portfolio, and IQ4hire does not sell, resell or receive vendor compensation from either. These are go-to-market advisory relationships, disclosed here for the same reason every other compensation arrangement is disclosed before work begins.
How We Help
AI procurement is being run today the way ERP procurement was run in 1999 — on vendor claims, a compelling demo and a board that wants to be seen moving. The categories differ; the failure mode does not. IQ4hire runs AI selection the same structured way it runs every other category: a defined requirement, a shortlist drawn from pre-validated vendors, apples-to-apples scoring, and a recommendation you can defend.
Where AI differs is in what has to be scored. Not just price and function, but what the system does at the edge of its knowledge, whether its reasoning is legible to the people who must act on it, where the training data came from, what happens to your data, and whether the workflow is one your staff will actually accept — the thing Ford's technicians settled thirty-five years ago.