The pace of technological progress is making more and more people fear for their jobs, which could soon be taken over by robots. But these fears are still greatly exaggerated, Igor Bogachev, CEO of Tsifra, is sure.
Some market participants perceive artificial intelligence (AI) not as a messiah capable of solving all the problems of mankind, not as a monster that will take away people’s jobs and take over the world. In fact, both options are far from the truth.
Much will depend on the pace of technology development in the coming years, but few doubt the prospects of AI. According to Gartner’s forecast, the adoption of AI will bring businesses $2021 trillion by 2,9, saving 6,2 billion work hours of people. And if in 2017 the AI market was estimated at $4,8 billion, then by 2025 a 20-fold increase is predicted, up to $89,8 billion. However, the AI that is now used in industry and production is still capable of solving only local problems. And although the use of wireless technologies in manufacturing has been growing at a rate of 32% per year in recent years, the industrial Internet of things covers no more than 6% of production in the world.
It must be recognized that at present there are objective reasons that limit both the possibilities of AI itself and the scale of its implementation. Two main problems can be identified.
1. Lack of uniform standards and regulations for working with data
About 65% of the total project time for the implementation of AI is the search and collection of data necessary for its operation. At the same time, a human-friendly information storage system is sometimes completely inconvenient for AI, because it was not designed for it. Also, a bit of chaos is introduced by the fact that each enterprise has its own system for collecting and storing information.
So while the speed of implementation of artificial intelligence is slowed down by the fact that enterprises are not ready to invest in collecting and storing data, managing this process and including it in the main business processes in the enterprise.
2. The inability of AI to solve unique problems
In addition, the complexity lies in the industry specifics of enterprises and technological processes – it is impossible to offer standardized universal solutions in the field of AI for petrochemical and metallurgical production. An individual approach sometimes needs to be sought even for a specific industrial installation.
For example, each blast furnace is made to order, its parameters are individually selected to solve specific production problems, the life cycle and the level of degradation of each furnace also varies.
There are similar difficulties in working with oil refineries, in which the stages of production can be unique for each enterprise. And, as we know, artificial intelligence still does not like everything that cannot be standardized, its level of development still does not allow it to be easily adapted to variable technological solutions in complex production. Therefore, the program still has to explain everything “on the fingers”, teaching it for each process separately. At the same time, in addition to information about the work of the AI enterprise, data on business processes and physical and chemical processes relevant to each type of production are needed.
Often, experienced technologists rely on their intuition when producing metals. For example, ferroalloys are used to achieve the required chemical composition of steel. Steel alloyed with ferroalloys has improved physical and mechanical characteristics. And, as practice shows, deviations in the volumes of added ferroalloys can be observed at different enterprises of the same corporation. This significantly affects the corporation’s costs for raw materials and could be optimized if the decision to change the share of ferroalloys was not made under the influence of the intuition of one or another specialist at each enterprise, but would be unified using AI. However, here too, everything rests on the lack of well-functioning mechanisms for collecting data: often they are taken from someone’s head and not recorded.
Can AI learn by observing human work, collecting data on how a technologist makes decisions depending on external factors? Yes, maybe even should. However, the decisions that he will make in the course of work will be limited by these learned schemes. If in production AI encounters an unfamiliar situation that it has not yet seen, it will stand in a stupor and will be forced to transfer control to the technologist.
It should be noted that large suppliers of industrial equipment like Siemens and Mitsubishi now have some advantage in the implementation of AI, which initially imposed a system for collecting data from equipment on counterparties.
In general, the lack of uniform standards and regulations for working with data greatly slows down digitalization. So the uprising of machines will be prevented in the end by human craving for intuitive decision-making and general confusion in production processes.
The abilities of modern AI should be soberly assessed – they are limited by rigid frameworks. That is why AI solutions are implemented pointwise at individual production stages, processes, and factories.
Here we are not yet talking about intelligence, only about its rudiments. Without a person, AI cannot do anything at all so far – neither to learn, nor to orient itself in variable factors, since for many years the industry was created by a person and for a person. So AI can only replace a person in the very distant future. In the meantime, finding itself in the area of uncertainty, the program will apply for a sanction for any action to the person. And it is right. Because safety should be paramount when using AI in manufacturing, making false decisions by the program should be cut off by a person. We hope that research will lead to the creation of AI that is able to generalize, incorporate experience and make correct and safe decisions in dangerous situations.