AI assists human intelligence – it doesn’t replace it

Technological developments bring forth visionary predictions of questionable value. This reminds me of an opinion in the Talmud that prophecy in our times had degenerated into the occupation of fools – such as predictions of flying cars or swarms of drones delivering merchandise.

But here we turn our attention to a serious subject. Given the increasing power of “artificial intelligence” software, will automation replace human workers at a rate that leads to social instability? I don’t think so. In fact, the question implies a mistaken notion of what artificial intelligence does.

There is hardly an area of industrial activity that has been (or will be) immune from the use of sophisticated computer algorithms and systems. The improving use of masses of data is an underlying enabler of new software. Like many other technologies, the theoretical basis of what is called computer artificial intelligence has been developed over many years. However, it has reached enormous value because of the massive and economical increase in computing power that made statistical techniques practical and enabled the application of neural network programming to make possible the analysis of huge amounts data and complex pattern matching of video and audio signal data. The result is that many industrial situations that relied solely on worker judgment can be automated.

With such success comes a problem. The perceived problem is not automation as such. Historically automation has been a key driver of industrial efficiency and economic growth. This is no surprise to the public as familiar jobs have disappeared. Anybody remember manual-switch telephone operators? The concern is the rate of change of job requirements that calls in question the task of retraining people for the new requirements. A recent IBM study predicted that 120 million workers will need to be retrained because of AI over the next three years in 12 leading national economies.

I believe that notwithstanding the amazing capabilities of the new generation of software, we should not overestimate the ability of computer intelligence to replace the human kind. Nor should we underestimate the ability of people to adapt to a changing industrial environment. In fact, as processes have been automated, new products and services were enabled over time because they became economically viable, creating new jobs. And judging by the full-employment situation in the US currently, jobs eliminated by automation have been replaced by others created with new technology – new products and services with big markets. The reason is that human intelligence is unique.

Despite the better software, the role of human intelligence remains key because it is unique in its capacities. Intelligence as defined in the Merriam-Webster dictionary: “1a. The ability to learn or understand or deal with new or trying situations. 1b. The ability to apply knowledge to manipulate one’s environment or to think abstractly as measured by objective criteria. 2. mental acuteness.”

Consider dealing with new or trying situations. The computer “thinking” consists of accessing data that is matched to the situation that it is asked to act on. Finding the match is only possible within the limits of the program and the data from which it draws the operational steps. Anything “new” outside the program’s parameter and the database that is being accessed will baffle the machine. The idea of “abstract” thinking is not relevant to computer operation.

I have had personal experience in seeing how the potential capabilities of workers can be developed in dynamic industries. I remember my experience in a chip plant when I was responsible for implementing rapidly improving automated production methods replacing manual operations. I found that the workers welcomed the change and rapidly learned the new operations that required a much higher level of initiative. Within a short time, I discovered that formerly manual operations were running smoothly in automated equipment supervised by the same workers. With the new equipment, they were asked to use a higher level of judgment compared with their previous manual work and they enjoyed the challenge.

The key power of computer ‘intelligence’ centers on the ability to deal with events that are repetitive and predictable

The key power of computer “intelligence” centers on the ability to deal with events that are repetitive and predictable. New software techniques are designed to have computers self-generate changes in programs based on experience. But in general, action is predicated on rules that are programmed. And given that computers do not get fatigued with failing awareness, this means that they outperform humans in repetitive tasks or in finding parameters in a huge mass of data – like image analysis.

Two examples illustrate how the computer’s ability to process huge amounts of data can increase production efficiency and enhance the productivity of workers. These two examples show the immense value that highly sophisticated software brings to industrial productivity, because in both cases they address problems that left to human detection alone would be slow and less effective. But these new systems are helping human operators’ productivity, not eliminating their role.

The first example is in chip manufacturing. With more than a billion transistors on a chip, eliminating production defects is critical. The quality-control technique includes viewing the surface of the chip with a high-power microscope. The inspection operator is trained to identify the defects and help trace their origin in the production process. With automated systems, the image is digitally analyzed and compared to model defects stored in the computer database.

With ever more sophisticated software able to access a huge database, and sophisticated algorithms, the computer reduces the choice of defects down to a small number, and the operator can then use personal judgment and expertise to select the most appropriate answer. That answer is linked back to the operating production equipment and remedial steps can be taken to eliminate such defects. Because of the sophistication of the visual-analysis software able to access and sort through a huge data base quickly, such remedial action can be taken in a very short time period. This means higher product yields and lower cost.

A second illustrative example is in preventive maintenance of production equipment. One way to detect impending equipment failure is by monitoring the sound of the mechanical equipment. Acoustic sensors are attached and their output monitored by computer equipment programmed to select and sort the sounds on the basis of massive stored information to which the sounds are compared. Sounds that match stored failure precursors are flagged and a monitoring operator has to decide when it is actionable. As with the earlier example, a huge amount of data has to be analyzed quickly, but the role of the operator remains essential because that is how ambiguities are sorted out and actionable steps decided.

In summary, ever more powerful computers enabled with new software will do a better and better job of complementing human intelligence – freeing humans from humdrum and mind-numbing repetitive activities and accessing massive amounts of data that enable better solutions. But the growth of economies follows from human creativity and intelligence. And this robots will not substitute for.