Artificial Intelligence (AI) vs. Generative AI
First, it’s important to understand the distinction between AI and Generative AI. AI is described by the Directive on Automated Decision-Making as a form of information technology that’s able to perform tasks that would otherwise require biological brainpower.
Generative AI is a form of AI that is able to produce written, visual and audio content. It relies on the user inputting a prompt (a specific instruction or question.) Generative AI tools, such as ChatGPT, Microsoft Copilot and DALL-E, are used to generate written content, brainstorm ideas, conduct research and generate visual content.
Generative AI’s demand increases despite environmental concerns
Since the launch of ChatGPT in November 2022, Generative AI’s popularity has only increased; so has its environmental costs. Generative AI causes greater harm to the environment than standard AI systems and technology.
Generative AI has been found to require more resources and specialized hardware in order for it to process sets of data, and make predictions and decisions based on user prompts. It’s energy-intensity is a combination of its training, general usage and rising demand.
Natural resources are used to produce the hardware in AI systems. For example, microchips need minerals like tungsten, lithium and cobalt. Despite the small amount of these minerals used in both AI and computer systems, AI’s rapid development has incited a global demand that could result in the supply of these minerals and other rare earth elements falling short by as early as 2030.
Energy-consuming GPUs are used to train Generative AI
In computer software, a Central Processing Unit (CPU) and a Graphics Processing Unit (GPU) are both sources of energy consumption. While the CPU is responsible for simpler tasks like internet browsing, checking emails and document editing, GPUs are able to process thousands of tasks at the same time. CPUs (as found in most office computers) are said to use roughly 1.5 to 2 kilowatt-hours (kWh) of electricity every month, or the equivalent of charging a smartphone every day for seven months.
On the other hand, if a single high-end GPU were to run something like intense video game graphics for just four hours a day, it could consume up to 69 kWh, sixteen times the energy-consumption amount of a CPU.
While GPUs were developed with video editing and gaming in mind, they are now an integral part of training Generative AI models. The amount of electricity they consume in order to train Generative AI models could provide hundreds of homes with power for an entire year.
Training larger Generative AI models (such as ChatGPT) may require tens of thousands of GPUs to be constantly running for several weeks at a time. Those GPUs then remain active after training is completed in order to serve millions of Generative AI users, some of whom may not be aware that they are using or engaging with Generative AI tools at all due to AI’s unavoidability while using search engines and social media.
Image-based Generative AI causes most harm
The fact that the majority of AI companies still do not openly disclose the environmental consequences of the training and usage of their systems is concerning.
Generative AI systems are developing faster, and they are becoming more deceptive at mimicking human behaviours and real-life appearances. It’s being integrated into search engines, social media applications, schools and the workplaces of everyday people, and the extent of its environmental harm continues to be overlooked.
In online spaces, a popular form of Generative AI is image (and video) generation. Alarmingly, it’s also one of the forms of Generative AI that is estimated to consume the most energy, using roughly 0.0029 kWh per generated image.
On top of its environmental harm, there are also moral and ethical concerns that come with image-based Generative AI, with particular concern about its ability to create violent and sexually graphic images of real people (including children.)
With that in mind, it is worrying to see how many individuals (outside of school or workplace requirements) use Generative AI tools as a leisurely activity with a disregard for the larger consequences of feeding their appearances and personal data to these systems, all while causing substantial environmental damage.
What Canada’s federal institutions can do
The Government of Canada outlined a few best practices for how Canada’s federal institutions can combat Generative AI’s environmental harm. The suggestions include using Generative AI tools in net-zero data centres, advocating for Generative AI developers to be transparent about the environmental sustainability of their systems and only using Generative AI tools when it’s necessary for objectives and desired outcomes.
Additionally, they said that all users in federal institutions should understand how Generative AI’s training, mining, manufacturing, transportation and disposal all impact the environment outside of its general usage. Despite provincial regulations, only about 20 per cent of e-waste is properly recycled; corporations and billionaires’ competitiveness about creating new and improved AI models will continue to keep the number of recycled e-waste low.
Generative AI and individual responsibility
Arguments about Generative AI’s environmental harm and how people contribute to it on an individual-level are often shut down by the notion that an individual creating AI generated texts and images will never contribute the same environmental damage to the planet as AI-obsessed billionaires and developers do.
While it may be true that AI developers, institutions, corporations and billionaires are doing the most environmental harm with the extent of their Generative AI usage, that doesn’t necessarily mean that individuals who use Generative AI tools to do things like create fruit-based AI versions of Love Island, or those who use it as a replacement search engine have not contributed harm.
A single Google search is said to use around 0.0003 kWh of electricity, compared to a response to a response from ChatGPT which can use as much as ten times the amount of electricity.
Engaging with these Generative AI tools, even on an individual-level, causes unnecessary and avoidable damage to the environment.
The intensity of its energy and water-consumption, and the data centres the systems are run out of may continue to do harm regardless of whether someone uses Generative AI tools out of fascination or boredom, but there is power in choosing to opt out of something that does more harm than good to both people and the planet.
The post Breaking down the cost of AI on energy and the environment appeared first on rabble.ca.
