Artificial Intelligence (AI) allows companies to manage increasingly complex business operations more efficiently and cheaply. It streamlines processes, automates the most complex tasks and enables previously unimagined efficiencies. This frees up resources for higher-value tasks, while big data analysis and machine learning can uncover decision-useful information that was previously hidden in mountains of data.
Data centres are currently responsible for about 4% of global greenhouse gas emissions, a figure that is set to grow as the digital economy continues to expand. But AI-based robots can automate a range of processes to improve efficiency, while digital twins enabled with AI and machine learning cut the facilities’ carbon footprint.
“AI and robotics solutions not only help improve energy efficiency, reduce carbon emissions, provide predictive maintenance, and improve security, but also automate routine activities which reduces workforce requirements,” reports consultancy EY. “AI can predict power outages, reduce maintenance costs, and achieve higher performance metrics.”
Because data centres consume so much power to cool servers, real-time control of cooling equipment using sensors and machine learning can have a significant impact, with some operators reporting savings of up to 40% of the energy used to cool facilities.
AI can also monitor performance and predict data outages, minimizing downtimes, as well as reducing cyber risks.
Accurate atomic-scale material simulation, however, requires a vast amount of computing power, using around half of the compute of UK government-funded supercomputers. On top of this, these calculations are approximations, so the results can be difficult to trust; their findings tend not to breach the confines of the research lab.
Fusion offers the promise of almost unlimited energy using the same reaction that occurs in the sun, but because it involves hydrogen atoms smashing together to create a plasma at a temperature of about 50 million degrees C, it requires extremely powerful magnets to contain the reaction. Researchers need to make adjustments to see what configurations yield more power. Normally these adjustments require a lot of engineering and design work, but DeepMind created a deep reinforcement learning system to control the magnets and alter the shape of the plasma.
AI is also being used to predict the results of simulations, and analyze the huge amounts of data produced by the fusion experiments.
The need to be able to rapidly develop vaccines was starkly highlighted by the Covid-19 pandemic but until recently, it was a slow process. AI is helping to speed it up in a number of ways. Machine learning allows large datasets to be analzyed to identify which parts of a virus should be targeted to provoke an immune response that will have a long-lasting effect.
Machine learning can also help to predict the toxicity of compounds as part of safety and efficacy testing, and help to set correct dosages for pre-clinical testing. Further, it can speed up clinical trials by identifying the best locations for trials, and the right test subjects. AI is also being used to predict how viruses such as SARS-CoV-2 will mutate and to counter them.
Once the vaccine is on the market, AI is helping to prioritize which patients should be vaccinated first as a result of other pre-existing health conditions such as diabetes, heart and kidney disease.
AI cannot replace clinical trials, but its ability to analyze the huge volumes of data that drug development and testing generates, along with the ability to detect patterns that human analysis cannot, promises to make vaccine development and deployment much quicker.
Cement production involves complex, interdependent processes and it uses a huge amount of energy, which contributes 40-60% of the overall cost of the product as well as producing significant greenhouse gas and other emissions.
At Carbon Re, we use various kinds of machine learning and deep learning – and our core expertise is ‘reinforcement learning’: an advanced form of artificial intelligence where an AI agent learns through repeated play in a virtual environment. This process outperforms ‘supervised learning’ (which comprises 90% of all AI machine learning applications) such as ‘Model Predictive Control’ used by other ‘AI’ control systems being developed for cement production.
Reinforcement learning is the most promising branch of AI because of its ability to interact with complex systems and solve problems that require sophisticated strategies.
AI technology can now develop a ‘digital twin’ of an individual plant by analysing the sensor data gathered on a minute-by-minute basis by Distributed Control Systems. This creates a high-resolution digital model that captures the properties of a given kiln (the thermodynamics, and the physical processes of clinker production) and delivers an accurate representation of the real-world performance in that specific plant. This digital twin provides a platform for AI ‘agents’ to learn complex process control in a sophisticated simulation. The outputs are insights on how to maximise the efficiency of the plant that enable operators to adapt to the continually changing plant conditions.