I remember seeing a gentleman from the IBM Institute give a presentation at one of our events about 10 years ago. He said that his technology could drive a cement plant better than the best central control room (CCR) operator could do on the best hour of their best day of the year, but that the technology could do it 24/7/365. The technology would never be tired, would never be hung-over, would never be worried about its divorce or mending the fence that got blown down in the storm last month. Using ‘knowledge capture’ (training of the technology by experienced operators), the company could avoid ‘knowledge leakage’ (loss of experience through retirement or death). You could ask the technology to optimise for fuel price, or cement strength, or CO2 emissions, or specific electrical energy consumption, or a combination of factors. Why wouldn’t you use that technology?! He was, of course, talking about relatively early AI, but things have moved on since then.
ChatGPT was launched to the public on 30 November 2022, reaching one million users within five days. It’s just one example of generative AI, which takes a vast training data set (say, everything ever written, and huge photo, film and sound libraries), and using statistics works out what is the most likely next word, pixel or note. It has been used on images, video and audio, so that generative AI can produce you an essay, a graphic, a video or a song. Early versions of ChatGPT and other large language models (LLMs) had clearly ‘artificial’ or robot-like attributes, but now they can produce elegant prose, images and video that are hard to tell from real-world examples and even songs that real people are willing to pay to listen to.
There are many other forms of AI, also known as ‘narrow’ AI, since they are good (or at least better than humans) at very specific tasks. They may be based on a variety of underlying technologies, such as adaptive algorithms, neural networks, computer vision, and machine learning. Many companies are working on self-driving cars, which combine a few of the approaches outlined above, while ‘narrow’ AI is being applied to many areas in the cement industry by a burgeoning array of new and more established companies. We recently conducted a poll to ask people what they thought AI would be used for in the cement industry, and they told us, in order of importance, that it would be used for: Pyroprocess optimisation; Data analytics; Predictive and preventative maintenance; Cement performance prediction; Mill optimisation; Plant management; Root-cause analysis; Laboratory and QC; Sensors; Digital twinning; Safety; Logistics and despatch; Customer service; Quarry optimisation; Homogenisation/blending; Refractory management; Commerce/finance; and Air Pollution Control optimisation. We’ll be covering many of these topics in our 1st Global CementAI conference on 19 - 20 May 2026 in Brussels.
At the same time, we asked people in the cement industry when they thought that an AI could do their job. 39% said ‘never,’ 15% said ‘before my retirement,’ 36% said ‘within 10 years,’ and 9% said ‘already.’ I was surprised. No doubt an AI, or an AI ‘agent’ (hence Agentic AI), can already do chunks of people’s jobs better than they can. However, unless you are unlucky (or a CCR operator), an AI is unlikely to be able to do your entire job - yet. Perhaps an AI could have written this article better than I did (it didn’t - I wrote it myself) - but can it make tea?
AI agents are getting better at doing longer and longer tasks, equivalent to a human concentrating on jobs that would take a person a day or a week to accomplish. As shown in the film ‘The Thinking Game’ (which is strongly recommended - and it’s free on YouTube1), you will see that AIs can now do tasks that humans just can’t do, such as solving the ‘protein folding problem.’ The next tasks for the computers to work on, according to Nobel-prize winner Demis Hassabis of DeepMind, are genomics, theorem proving, quantum chemistry, food science, fusion, particle physics and climate science. With AIs, I think that the problems will be solved - and will change the world.
The next step will be Artificial General Intelligence, AGI, which will arrive when AI can achieve human-level capabilities in a wide range of tasks. Many companies have large teams striving for this - and it has been forecast to arrive in the very near future - even in 2026 or 20272, but certainly in the next few years. Whose job is safe then? And whose is unsafe?
And after that? Well, just as some AIs can outperform humans in specific areas, artificial super-intelligence (ASI?) will outperform humans in all areas. As I mentioned earlier, it doesn’t have to be brilliant - it ‘just’ has to be better than the best human.
1 https://www.youtube.com/watch?v=d95J8yzvjbQ (The Thinking Game movie)
2 https://www.theguardian.com/technology/ng-interactive/2025/dec/01/its-going-much-too-fast-the-inside-story-of-the-race-to-create-the-ultimate-ai
