The Machine Intelligence debate goes on, with the latest topics including predictive maintenance. The implication seems to be that if a machine can predict equipment failures and eliminate them by creating messages to human operators or maintenance technicians to take preventive action, that this is a sign of intelligence. The root of the confusion is around the nature of intelligence.
It has taken hundreds of man-years of programming and several decades of programs from ARPA to produce an autonomous vehicle (car) that can navigate it’s surroundings. Having accomplished this feat, the vehicle cannot take it’s own decision about where it is to go or what it will do. The tasks are provided by the operator of the system. The behavior, however intelligent it might appear, is very narrowly defined by the programmer(s) and now matter how much the systems may acquire data about it’s surroundings, it cannot exceed its programming to spontaneously go pick up a case of beer.
Preventive maintenance technology has been emerging for several decades. The most common application was monitoring ball bearing vibration signatures in order to anticipate a catastrophic bearing failure that might damage a large piece of capital equipment. In the early days of the technology it was mostly about finding high resolution vibration sensing and using Fourier analysis to determine a threshold for pass or fail. The data was empirical, it had to be acquired, stored and analysed. From software analysis of the data a threshold can be mathematically defined for the amount of noise on the sensor that represented a bearing that was about to fail.
In today’s technology high resolution analog to digital converters, Fourier transform analysis, logic and network analysis can all be coordinated on a single chip making the application very inexpensive. The vibration sensing can be done with a low cost piezoelectric device or accelerometer. So the lower cost threshold makes this type of preventive maintenance widely affordable and easy to implement. It’s only taken 30 years to get here.
Irrespective of the means, is the a case where computer software is somehow predicting the future? Not at all. The electronic system is simply responding to signal data that has valuable information. What is present is the embedding of “intelligence” achieved over decades of observation, trial and error, into an electronic system of sensors, processors and code. The system cannot exceed its programming and spontaneously respond to some other problem.
What we call “intelligence” in machines and robots is the ability to mimic human knowledge within a very narrowly defined field and under very well defined circumstances.