Energy-thirsty AI needs to help solve the problems it's causing

Artificial intelligence: the solution to new energy problems it has also partially created.

“It is difficult to overstate the potential of AI in the fight against climate change,” as Minister of Industry and Advanced Technology and Adnoc chief executive Dr Sultan Al Jaber wrote recently in Project Syndicate.

But the sector’s rampant electricity consumption is straining grids and threatens to raise coal and gas consumption and subsequent greenhouse gas emissions.

The effect of AI on energy will play out over three timescales.

At least in the short-term, it will raise energy demand. And, if low-carbon energy and efficiency don’t keep up, that inevitably means increasing greenhouse gas emissions. At the moment, AI computing accounts for less than 1 per cent of global emissions – that could grow substantially, but by how much depends on the speed of its advance and the choices we make.

Google has a goal to reach net-zero carbon emissions by 2030.

Yet since 2019, its emissions have leapt 48 per cent – and most of that is driven by expanded data centres and AI. Its 14.3 million tonnes of carbon dioxide last year was more than the entire release of the Baltic nation of Estonia. This is despite running its operations now on 64 per cent low-carbon electricity – using renewable and nuclear power.

In the medium term, AI promises untold gains in the size, efficiency and cleanliness of energy production and use.

Dr Al Jaber’s commentary mentions gains in balancing variable renewable generation, identifying molecular structures that trap carbon dioxide, cutting water use while boosting crop yields and “breakthroughs in fusion, hydrogen, and modular nuclear power [as well as] long-term battery storage”.

He has convened a “Change Makers Majlis” in Abu Dhabi in November to discuss AI and the energy transition.

A team at Innovation for Cool Earth Forum, chaired by my Columbia University colleague David Sandalow, has extensively characterised other opportunities.

The more dramatic opportunities fall into three rough categories: optimising and automating complex systems; spotting patterns and connections in large data-sets; and designing and simulating otherwise impossible technologies.

In the first group come opportunities such as matching renewable generation in combination with weather forecasting, demand, home heating and cooling, battery storage and electric vehicle charging. Reducing the financial and mental burden of lowering emissions and saving energy costs could help ease some of the political pushback against net-zero policies.

The second group includes research, bringing together the vast published literature in ways beyond any individual scientist. AI systems can test substantial numbers of designs for advanced batteries, carbon capture materials, or catalysts for biofuels or hydrogen, and pick out the most promising for further work. Already in 2022, Deep Mind, a subsidiary of Google parent Alphabet, unveiled a system to predict the structure of 200 million proteins.

Such systems can monitor global greenhouse gas emissions and carbon storage in near real-time. They can trawl through large data sets of remote sensing, geochemistry, seismic images of the surface, magnetic, radioactive and gravitational readings, to find new deposits of hydrocarbons and critical energy minerals such as rare earths, uranium, lithium and copper.

KoBold Metals, a start-up backed by Bill Gates and Jeff Bezos, mining firm BHP and Norwegian state oil champion Equinor, claims to have used AI to find a large copper deposit in Zambia, which the country’s President said could become one of the world’s three biggest.

The third area could include such things as simulating nuclear fusion. This inexhaustible clean energy source requires us to harness plasma at temperatures of a hundred million degrees, confine it within intense magnetic fields, and keep the reaction going stably without melting the containing chamber. Models might predict room-temperature superconductors that would eliminate losses in electricity transmission.

AI computing will also grow more energy-efficient. Combining such opportunities means that in the medium term, energy use will be better directed, supply will be enhanced and reduced in cost, and potentially emissions will fall significantly – with the right choices of policy.

Beyond these and heading into the long-term – the 2050s and beyond – are the imponderables.

We hear often that it only requires “political will” to beat climate change and that we have all the technologies we need. Instead of purveying the dark arts of electoral manipulation salted with big data from social media, could AI simulate whole societies and economies, to help design and introduce climate-friendly policies that win wide acceptance?

Some of the more ambitious claims for AI suggest an intelligence that would surpass that of humans in almost all aspects by the 2030s.

Would such a creation prioritise highly effective climate action? Or would it reflect the attitudes of its creators, perhaps a trillionaire, corporation or country that cared little for the well-being of billions of vulnerable people?

But we can be certain of one thing: this AI world will require much more energy than today. This could be superintelligences, humanoid robots, globe-spanning climate clean-up systems, simulations of the whole of society, off-world industry, extraplanetary colonisation or other undreamt-of developments. The trend of technology is always to require more and better energy, even as it gets more efficient and cleaner.

The Gulf countries could be among the big beneficiaries: they have money, growing technical expertise, and the priority for AI, to apply to it to many of their key challenges.

Relatively cheap, clean and abundant energy will be helpful in building computing power, partly offset by the extra need for cooling versus more temperate climes.

With a relatively small population relative to their economies and energy resources, they will benefit disproportionately from multiplying the effectiveness of their people. Adnoc has taken steps in this regard, in its AIQ joint venture with Abu Dhabi-based AI company G42, and its cross-shareholding with big data analytics firm Presight.

Success depends on building expertise, meshing artificial with the best of human intelligence, and reaching and expanding the frontier of knowledge.

It also requires much more and better data for training systems. This is one area where the Gulf needs to improve: rather than hoarding information to preserve private empires in the name of national security, data from the energy sector and many other areas must be available, reliable, transparent and organised.

Competitors, whether rival energy producers or competing energy sources and methods are not sitting idly at their keyboards.

For now, AI represents an improvement in the industry and an opportunity for energy sales.

But in the long-term – whether 30 years or 30 months in the AI world – it will transform the energy business.

Robin M. Mills is chief executive of Qamar Energy and author of The Myth of the Oil Crisis