Byte-sized progress, Gigatonne-sized consequences

The speed of Artificial Intelligence (AI) adoption over recent years and the acceleration in recent months is taking businesses by storm. However, as AI becomes a key aspect in shaping the future of organisations, due consideration must be paid to the duality of the impact that AI can have on the environment. 

The hope is that AI will optimise and streamline operations, and support businesses in maintaining their competitive advantage, driving organisations into new and efficient ways of working. Additionally, business plans are afoot for using AI to better track companies’ carbon emissions, to recognise opportunities for improvement in reducing the corporate carbon footprint.

What has not yet been taken into account, or at least not by everybody rushing to harness the new technology for their businesses, is the environmental toll that it takes to power and develop AI. 

AI's carbon footprint

The connection between AI and environmental responsibility is a complex one. Whilst it is true that AI brings a wealth of benefits such as improving upon efficiency and automating repetitive and rules-based tasks, it is also true that if current AI practices remain unchanged, the energy needed for machine learning training and associated data storage and processing will account for a significant amount of global electricity consumption by 2030.

Avoiding this will require increased transparency from AI researchers and data storage service providers to quantify the impact on sustainability of the widespread development of AI. Without a quantifiable ‘number; for the environmental impact of developing AI solutions, it is easy to regard AI as something that does not require extensive physical resources to set up. This thinking tends to contradict the commitments being made by business leaders to use AI to improve their overall sustainability. 

However, if we compare AI to cloud computing, another innovation that took - and continues to take - businesses by storm, we know that organisations identified sustainability as a key factor and took action to offset and reduce their carbon emissions. Some of the world’s largest organisations such as Google and AWS for example, have set themselves ambitious targets of running whole datacentre estates on carbon-free energy by 2030 (Google) and powering entire operations with 100% renewable energy by 2025 (AWS).  

AI and the cloud are similar in their capabilities of automating processes such as: data analysis, data management, security, and decision making driven by highly efficient compute capabilities. Both technologies are also incumbent on data storage and expansive processing sites to power them. Cloud technology, similar to AI, can and is used to drive sustainable innovation and cut carbon emissions. It is common knowledge amongst business leaders and sustainability experts that migrating to the cloud can reduce carbon emissions by a staggering 59 million tons per year. Given the similarities between these two technologies, it is therefore likely that we will see the same trajectory, with businesses committing to sustainable AI integration in their operations, which is very good news indeed. 

Balancing ESG goals and maintaining innovation

It is not at all impossible to advance AI capabilities whilst maintaining a business' ESG commitments. Sustainable AI practices are slowly coming into play, which include the use of more efficient Graphic Processor Units (GPUs), using renewable energy grids or reporting training and sensitivity to hyperparameters in published performance results. The best way to achieve a sustainable integration of AI is by employing carbon neutral computing methods such as GreenCoding. 

GreenCoding best practice enables researchers to write computationally efficient codes and algorithms. By writing shorter and more efficient code, datacentres require less power to run the software, and processes can be shut down when not in use. Many businesses have already adopted GreenCoding principles and with good reason. The practice not only optimises datacentres and hardware, but it seamlessly integrates software development into the business’ sustainability goals by reducing energy consumption. 

There are additional steps that datacentres can take to enable energy efficient ways of powering AI. For example, using processors and systems that have been optimised for Machine Learning (ML) training, rather than using general-purpose processors. This improves both performance and energy efficiency significantly simply by optimising the existing GPUs being used to be more energy efficient. An emissions calculator has also been developed, which can help businesses reduce their sustainability impact and understand opportunities for improvement.

Conclusion 

There is no question that AI will power the future of businesses and alter business operations going forward. As a new innovation that has the potential to benefit most if not all aspects of society, it should be leveraged to its fullest potential. There should, however, be reasonable considerations of the potential impact that AI may have on the environment. 

The end-goal is not to stifle innovation, but rather to ensure the use of AI has a positive impact on businesses in all aspects, including their ESG commitments. This requires extensive collaboration; having a trusted sustainability partner to help on this journey will help boost business and mitigate any potential environmental impact.