OpenAI’s Latest
OpenAI’s recent announcement of its new model OpenAI o1, code named “Strawberry“, a model capable of handling math, physics, and chemistry tasks has stirred a great deal of excitement in certain quarters. The claims are impressive, equally, skepticism is warranted about the applicability and value of benchmarks surrounding claims that the new model “performs equivalent to a PhD student.” Previous iterations struggled with basic arithmetic due to their training focus and design. This announcement suggests progress in that area, and as always more details and hands on analysis is needed to meaningfully evaluate the model’s capabilities. Conspicuous in its absence from the broader conversation was any discussion or acknowledgement about what the new models energy demands will be like. Alongside the ethical concerns about the data used to train GenAI, the energy demands and carbon footprint of Cloud Computing and specifically where GenAI is concerned remains a significant risk associated with the technology, and conversely for all of us.
Progress demands Power demands Progress ad infinitum
Excitement surrounding LLMs is flagging, including notably in the Stock Market and among Venture Capitalists. Real world examples of GenAI and its usage bump up against concerns over plagiarism, the quality of the output, especially misleading and outright incorrect responses. And perhaps most dangerously for the technology’s adoption, the imposition of AI where users never asked for it. Response to Google’s AI generated answers in Google Search Results, and the technology’s intrusive insertion into MS Office products are two examples that have been met with mixed reviews by the broader public. The leadership at OpenAI likely saw timing this announcement as an opportunity to reinvigorate the broader public and the investor community’s enthusiasm. While the hype surrounding OpenAI grows in both in volume and frequency once more, we risk overlooking another crucial factor, particularly for people not directly involved in Software Engineering: energy consumption. Studies have highlighted the significant power demands of training and running these models . This trend aligns with projections of increased energy needs, particularly in regions like the US mid-Atlantic, where data center expansion coincides with electric vehicle growth and building electrification .
If the past is anything to go on, increased energy demand leads to price hikes. These disproportionately impact retail, residential and small hold agricultural or industrial customers who lack negotiating power or relocation options of companies like Microsoft, Amazon or Google. There are more public discussions about the potential ways to accelerate renewable energy adoption. Even here in Ontario after the Conservative Government actively disincentivized renewables as a matter of policy, they are talking about funding them again, urgently. but I digress. New energy supply, distribution grids and delivery infrastructure are time, labour and capital intensive, so no matter the jurisdiction the economic impact on consumers remains unclear. Public perception will be crucial if rising energy costs are not seen as aligned with the needs of the broader public.
Cleaner, Greener Technology requires regulation
The technology industry may have long positioned itself as holding the promise of a greener, less energy and materials intensive future. Remember the paperless office? The reality is that there are tonnes of e-waste representing out-moded, un-reparable devices and technology that are even uneconomical as recyclables. The industry needs to seriously evaluate the use of and investment in sustainable solutions to avoid a situation where the public resents footing the electricity bill, and that AI doesn’t come at the expense of de-carbonizing the economy, or slowing, stopping and eventually reversing climate change. There’s very little chance of the technology whales contributing meaningfully to the common good without their either financial incentive carrot or a regulatory stick. Financial incentives tend to come with the size of near monopolies, like Microsoft, Google, and Amazon, this may make choosing the carrot a regulatory and political necessity.
Curtis Palmer, CC BY 2.0 <https://creativecommons.org/licenses/by/2.0>, via Wikimedia Commons
