How will artificial intelligence transform energy innovation?

Serious challenges must be overcome before AI techniques can fulfil their full potential on innovation, including data availability

Artificial intelligence will enhance the capacity and creativity of scientists in generating and testing new ideas.

This article was published by the International Energy Agency on Nov. 24, 2024.

Artificial intelligence is an innovation that catalyses further innovation

Like the steam engine and electricity, artificial intelligence (AI) is a general-purpose technology that could profoundly transform the global economy and the world’s energy system. Though key uncertainties remain, it stands to have major impacts. High on the list is its potential role in accelerating innovation.

Impressive technological advancements – both incremental and radical – have helped drive down the cost of key energy technologies in recent years. But to achieve global energy security and emissions goals, existing clean energy technologies need to keep improving, and novel energy technologies must reach the market. AI will enhance the capacity and creativity of scientists in generating and testing new ideas. But for AI-accelerated innovation to really deliver for the energy sector, policymakers and the scientific community need to build a common understanding of the most promising applications and key enablers – and address critical gaps.

This is a key focus of the IEA’s new workstream on energy and AI, which also involves analysing how the adoption of AI will affect electricity consumption by data centres and how AI can be applied to optimize complex parts of energy systems, such as electricity networks. The upcoming Global Conference on Energy & AI, which is bringing together leaders from government, the energy sector, the tech industry and civil society to discuss these topics for the first time, will provide a space to kickstart and advance public-private dialogues on these subjects at a critical moment.

Does AI represent a step-change in the speed of energy innovation?

For energy analysts, a fundamental question is whether the application of AI will cause the rate of technology progress to deviate from current projections. In the field of semiconductors, Moore’s Law – an observation from the 1960s that the number of transistors in an integrated circuit doubles about every two years, which proved startlingly accurate for several decades – is well known. Similarly, for many energy technologies, it is common to project cost reductions for each doubling of cumulative deployment, known as the “learning rate.”

However, progress in the semiconductor sector has slowed, and Moore’s Law has not been a good guide for technological development since around 2010. Experts question whether the learning rate for a technology like electric vehicle batteries, which IEA analysis projects at 15 per cent, can be maintained over future decades. Recent inflation in technology prices, partly caused by mismatches between supply and demand for critical material inputs, are a reminder that factors such as manufacturing capacity and trade can also impede the innovation process.

Some analysts see AI as a means to keep current learning rate projections on track despite these concerns. Others see it as a more disruptive force that could make today’s rates look very conservative. To inform this debate, it is necessary to take a closer look at the specific ways in which AI could boost the pace of innovation.

Early examples of AI discoveries on energy-related materials are very promising…

Finding a higher-performing material for a task, or one that does not contain certain undesirable inputs, has typically relied on human ingenuity and knowledge of how different compounds behave. But the number of possible options is often vast. AI techniques are already excellent at solving problems by optimizing for well-understood relationships across large and well-structured data sets.

In July 2024, researchers from a US government laboratory and Microsoft published results of a study that used AI to assess 32.5 million possible new solid-state electrolytes for lithium-based batteries and found 23 new ones with the right characteristics. Scientists in Sweden recently screened 45 million potential new battery cathode molecules and found nearly 4 600 promising candidates. Other teams have achieved similar results, and one has pursued their findings through to synthesis and testing. Notably, these types of techniques are increasingly attracting financing: Anionics, an AI start-up, recently partnered with the battery manufacturing subsidiary of Porsche, while Mitra Chem has raised USD 80 million with its promise of shortening the lab-to-production timeline by over 90 per cent.

Recent breakthroughs have not only been battery-related. Researchers using AI tools have also found they can engineer enzymes for biofuel synthesis, predict high-yielding biofuel feedstocks, identify industry-beating catalysts for hydrogen-producing electrolyzers and generate materials for carbon dioxide (CO2) capture. And as AI becomes an increasingly indispensable part of the research process for energy technologies, innovators will be also benefit from developments in adjacent areas, including improved robotics and automation. A recent study of the impact of using AI tools in an industrial research setting showed a 39 per cent increase in patenting by the company in under two years.

Appears in:  How will artificial intelligence transform energy innovation?
Sources:  Toner-Rodgers (2024), as modified by the IEA.  To see an animated version of this graph, click here.

…but major obstacles, such as data availability, remain

Still, serious challenges must be overcome before AI techniques can fulfil their full potential on innovation. One key issue is data availability. The datasets used today have incomplete information about possible materials and represent a restricted subset of molecules or reactions.

The development of massive, structured, specialized datasets to train AI models, such as such as the Materials Project and Cambridge Structural Database, is underway, but they must be further expanded if real-world scientific problems are to be solved. While the creation of “synthetic data” to train models can overcome some of the data gaps, there is no substitute for experimental data, and the fastest route to large and reliable experimental datasets is co-operation between laboratories, including at the international level. The Mission Innovation M4E platform is an example of an international initiative that could demonstrate how governments can support common protocols and jointly curated data.

Another challenge is finding ways for AI to optimize results for more than just a narrow set of characteristics and incorporate considerations that are essential for a material to be integrated into a functional product. Today, substantial human checking and testing is still required – for example, to assess performance at different temperatures or interactions with all other components of a device. Also, working out the recipe for manufacturing the materials designed by AI can create considerable follow-on work. Having AI perform these more complex tasks appears feasible, but it leads to high computational requirements and costs that must be assessed.

If discovery is accelerated but testing and commercialization are not, then half the challenge will stay unaddressed

Identifying a new material for an energy application via a computer-based method is less than half of the innovation task. Prototyping, followed by commercialization, mass manufacturing and widespread market uptake, can take years or even decades. Yet other AI-related tools in development could compress these timetables, too.

One is known as the self-driving lab. The A-Lab at the US Department of Energy’s Lawrence Berkeley National Laboratory contains a series of robots that, since February 2024, can synthesize the energy storage chemicals predicted by computer calculations to offer major performance improvements. This self-driving laboratory can process up to 100 times more samples per day than a human-run equivalent.

For large, complex systems, a computer-based aid known as a “digital twin” can significantly reduce the costs and risks of design and scale-up. Digital twins, which are virtual representations of all the elements of a specific facility or process, have been used to optimise manufacturing for over a decade but are now being powered by AI and applied to innovation. In sectors such as nuclear fusion, they are helping design and test equipment. The hope is that the costs of complex engineering design will be sharply reduced, particularly for expensive, first-of-a-kind projects. This could be a significant fillip for innovators of industrial decarbonisation technologiesgeothermal energysynthetic fuel processes and CO2 capture and storage.

However, difficulties also persist in applying AI to this phase of the innovation process. Currently, these tools are not all widely accessible to innovators in the scale-up stage. Additionally, skills gaps could be an issue in a fast-moving field, while responsive regulatory and standards frameworks will be necessary to support and accommodate new approaches to testing and commercializing products and services.

The time to consider the policy context is now

There is clear potential for AI to enhance and accelerate innovation to tackle a wide range of energy technology challenges. There are exciting examples of this happening already, but the full potential of AI in this area will not be realised unless governments focus on some key emerging issues upfront.

To drive scientific discovery towards the most impactful outcomes, there is a need to invest in searchable databases that follow common protocols and are widely accessible, including by interconnecting laboratories across international borders. Investments in skills and equipment will also be required, and policy makers can guide efforts to the most pressing technological needs. To support commercialization, policy makers should also consider how to make new digital tools widely available to innovators and help investors to adjust to the resulting reductions in project risk. At the same time, the computing and energy needs of AI for these important tasks, as well as potential risks such as those related to intellectual property, must be discussed in multilateral fora.

If successful, AI will not only accelerate and improve innovation outcomes but also deliver economic competitiveness, too. Once new products are ready for market, analysis with AI of data generated by new products can raise their value to consumers. Better decision-making by software for controlling new technologies can likewise reduce risks and add value for their users. The benefits will be shared by all countries, their innovators, investors and firms if efforts are anticipated, directed and cooperative.

 

 

Facebook Comments

Be the first to comment

Leave a Reply

Your email address will not be published.


*