The Future of AI in Industries

(Will the fourth industrial revolution (4IR) be a revolution by the industries?)

How would artificial intelligence (AI) impact the future of industries and how would industries impact the innovations in the domain of AI? Predicting the future is a fool’s errand! Many great minds have failed at it. So let us keep ourselves restricted to predictions that may happen in the next five years only. Though we should be mindful of the fact that with the knowledge half-life of Fuller’s knowledge doubling curve becoming ever shorter, even five years can bring some fundamental changes.

Let us begin our story with the main actor. This being a story about industries, our protagonist needs to be an industry. Over the past few years, I had some great interactions with interesting industries while consulting them regarding the application of AI in their facilities. Our protagonist is inspired by these experiences. It is Zircon Steels Ltd., an imaginary entity. Zircon is a small-scale steel manufacturer which gets iron ores from local mines, smelts them, and makes steel of various grades from it.

The Challenges: Zircon makes a good amount of profit and is not much bothered about applying AI in their facilities. However, they have heard of AI and wanted to check if some of the issues they have faced of late can be solved by this magic wand called AI.

• Quality of steel: One of their customers is very particular about the quality of steel and in the last few years there have been cases where there were contaminations in the steel. These were process contaminations. What they wanted is some way of telling the quality of the steel while it’s still in the process.
• Worker’s Issues: The worker in charge of the control panels is a smart and experienced person. But sometimes, he is not as attentive as he would like to be. That has resulted in fluctuations in the quality of the output.
• Efficiency: Their facility is not as efficient in terms of energy usage as the big sophisticated ones owned by larger companies.

Visions: These challenges, though specific to one industry of a particular domain, can be used to generalize some of the major needs of industries that can be satisfied by AI. From these, the following are some of the visions of industrial AI.

• Zero Waste: Can AI be used to minimize waste, viz. material waste, energy waste, and the waste of human time? It also takes us to the domain of recycling and generating revenue from it. AI will play a substantial role in circularizing the industrial ecosystem. In the industrial ecosystem every scrap saved adds up quickly and the revenue generation potentials are huge. A robust, reliable, and agile digital twin will enable industries to work more efficiently towards zero waste.
• No industries left behind: Can we transfer AI models working in other industries to new industrial setups and expect good performance? This is transfer learning on steroids! It will enable the building of industrial ecosystems where data can be shared by participating industries with the expectation that the AI algorithms should be transferable to each participating industry with minimal dev-time.
• Perception centric: Human perceptions are “personal” and subjective. In every industrial expert system, there is always a human in the loop. The way humans perceive the output of AI algorithms or graphs or plots presented to them is not well-defined. So the future industrial intelligent control systems would need to take perception into account in their development phase. Not only would the intelligent control system need to present the process parameters in easy-to-perceive formats, but it will also need to adjust the presentation according to the current state of mind of the operator.

Main Challenges: The vision presented above may sound a little bit futuristic. However, even the current-day AI algorithms are not fully utilized in every industry. Let us quickly discuss the main challenges that make Zircon slow to adopt AI-based control systems. What are the main hurdles limiting the application of AI in the facilities of Zirconia and industries like it?

• Data: Lack of a “sufficient” amount of data is an obvious challenge cited by almost every AI engineer. The lack of data is more acute in conventional industries using older equipment. However, we must deeply appreciate the fact that in the industrial ecosystem the lack of data is mostly linked with the lack of sensors. For industrial applications of AI, RDI in sensors and AI need to go hand-in-hand. Every challenge needs its own set of sensors and, together, they need new/refined AI algorithms. SensAI is going to be the major field of development to enable industries to fully benefit from the current wave of developments in the domain of AI. E.g. measuring the quality of steel while the metal is in the pipeline needs both new sensors and new AI algorithms, it needs SensAI!
• System of systems: Another big challenge for the adoption of AI revolution in industries is the way in which industries are fundamentally different from some other applications (like face recognition or AI-enabled advertising). The development of AI systems in software only is challenging. The challenge is exacerbated when we add hardware in the loop. This becomes even more challenging with humans in the loop. The challenge keeps growing manifold as we add laws and regulations, health and safety in the loop as well. The development of AI solutions that will work for industries needs systems thinking.
• Explainable AI (xAI): The blast furnaces used by Zircon were old. However, they still have a pretty good amount of sensors in them. Control and instrumentation systems for traditional industries like metal industries have been well developed for a long time. These traditional control systems are based on well-trackable equations where every result, every control step taken is well understood by the experts. Industries will need a lot more push to replace these well-developed analytical control systems with a deep-learning-based intelligent control system. The current lack of explainability of AI systems is a major deterrent for industries in adopting AI-based control systems.

Algorithmic Development Needed: Now that we have discussed what industries wish for AI and things that slow the adoption of AI in the industrial ecosystem, let us quickly discuss algorithmic innovations that will enable the wider adoption of AI systems in industries.

• eAI: AI algorithms need to be explainable, ethical, and empathetic! Explainable AI (xAI) is a major stream of development in the AI fraternity. One of the greatest developments in this direction has been the recent work towards bridging the gap between discriminative and generative AI. The development of generative AI models (GAN being one of the most popular and powerful versions of it) started using neural networks not just to predict the discriminative boundaries of a given data-point, but also to generate the data itself! So far, this has been the forte of statistical machine learning. This, for me, is the starting point of xAI.

Similarly, ethical AI is another recent area of activity. The recent stories around Google have given this field some well-deserved hype. For example, removing biases from AI algorithms is a huge challenge getting some good attention from experts.

• Perception in the loop: The importance of human perception is getting a lot of attention of late. Any real-life AI system will need to take this into account. We are not talking about making AI systems perceive things like humans. That is a great goal for hard-AI scientists. However, we are talking about quantifying human perception and using them in training our neural network models. Such perception-centric AI systems will be crucial towards the adoption of AI in complex systems like industries and smart cities.
• Intuition-based AI: The lack of a sufficient amount of data and the lack of generalized transferability are two challenges that we have discussed above. Bio-inspired computational approaches have always given us new ways to look at problems. One brain-inspired approach that may enable AI algorithms to solve both the data and the transferability challenges is “intuition”. Intuition-centered AI models can use less amount of “right kind” of data to build robust models. It can also give directions to have an architecture that can leverage fusing symbolic and non-symbolic AI to have a modus operandi to enable better transferability of models in the industrial ecosystem.

Concluding Remarks: As AI-engineers, we are living in interesting times indeed! The media is full of stories about AI. In April 2021, the EU published its proposed AI regulations. These are some of the most advanced and considered regulations I have come across so far. An AI engineer can not just sit with their computer and innovate AI algorithms. An industry cannot just take some codes and start implementing them in its control systems. Laws like these are more enabling (to push development in the right direction) than limiting.

In addition to these, interesting opportunities are coming up which will push for larger adoption of AI. For example, the EU is pushing hard to have better circular economy action plans and plastic strategies.

The fourth industrial revolution may not be a revolution by the industries (yet)! But it will be an industrial revolution fundamentally different from the previous ones, where prosperity will take the front seat and not just profit; where machines will not just assist humans but also cooperate with humans to make the world a better place!