Menopausal Mother Nature

News about Climate Change and our Planet


How AI is Contributing to Global Warming and What it Can Learn from Bitcoin – Security Boulevard

After Tesla announced it was buying 1.5 billion dollars of Bitcoin and would start accepting Bitcoin for purchases in the near future, the cryptocurrency’s value shot up dramatically. Not surprisingly, media outlets throughout the world have covered the story extensively as the cryptocurrency’s price continues to gyrate.

In the meantime, an equally important, but less positive story about Bitcoin has been percolating  — the massive amount of energy it takes to keep the currency operating. Bitcoin is an energy hog. Some researchers calculate Bitcoin as consuming about 50 percent of the energy used for data centers globally. To put this in perspective, according to BBC, if Bitcoin were a country, it would be in the top 30 countries in terms of energy use worldwide and use more energy than all of Argentina.

Bitcoin’s reliance on its “proof-of-work” algorithm consumes huge amounts of energy, primarily sourced from fossil fuels. To put it in context, this amount of energy equals more than the total consumption of energy of entire countries. For example, the Bitcoin network consumes as much energy per year as both Chile and Argentina.

AI is an Energy Hog, Too

What does Bitcoin’s energy consumption problem have to do with AI? At a fundamental level, it turns out, some AI applications require similarly massive amounts of computing energy to function.

Neural networks and other DARPA-identified AI technologies are trained over the course of several rounds of data processing to perform functions like facial recognition and to safely guide self-driving cars.

As consumers and government and commercial entities increasingly adopt these AI applications, the related neural network demand increases substantially. In fact, it has been estimated that neural network training costs are doubling every three to four months, with no end in sight.

Harmful Marketing Messaging Contributes to Energy Waste

Perhaps even more alarming are the examples of AI-related energy consumption that primarily exists to bolster marketing efforts around tech. Companies are incentivized to grow enormous, ever-expanding datasets to claim a marketing edge over competitors. Having the largest “data moat” means making significant financial and environmentally-damaging energy investments.

As is true for Bitcoin operations, the carbon footprint to run AI-related technologies is growing exponentially. For example, one elaborate deep learning model known as GPT-3, creates a carbon footprint equivalent to traveling 700,000 kilometers by car for a single training session.

Tech-Related Environmental Government Regulations on the Horizon

Recently, the U.S. Congress passed the latest version of the National Defense AuthorizationAct (NDAA ‘21), which includes guidelines for how the government will incorporate AI into its various agencies. Importantly, the Act calls for the prioritization of several related issues, including the technology’s environmental impact.

Going forward, we may see tech companies working within a similar framework to manufacturers and other entities that are subject to a carbon tax that can be mitigated in part through carbon offsets.

These initiatives allow participants to partially or fully “offset” the negative environmental impact of a given behavior. For example, a factory can offset the amount of CO2 it dumps into the atmosphere by buying a carbon offset to fund a project that reduces greenhouse gases. Similarly, companies that create and use certain AI platforms could seek to offset their energy consumption in this way.

Looking Ahead: Mitigating the Energy Impact of Modern Technology

At a fundamental level, it would be impossible to consider technology, including AI, outside the context of energy consumption. Technology requires energy to function.

The good news is that we have the power to limit that energy usage making decisions like using vastly more energy-efficient advanced algorithms. This is true for cryptocurrency as well as AI-related technologies.

In fact, a central goal of the adoption of a more advanced form of AI called “self-supervised AI” is to reduce the amount of data needed to train the AI and function independent of human involvement or training. Self-supervised AI, unlike less advanced forms of supervised AI which require constant feeds of data and human operator involvement to tune and train the algorithms which underpin them, can function independently, learning and operating instead completely on it’s own. This allows for exponentially less energy consumption over time v.s. Supervised AI. 

Efficient algorithms can be run on standard, off-the-shelf components rather than the expensive, limited use, and power-hungry hardware neural networks rely on. Self-supervised AI can be used to create a generative baseline of expected network behavior in about a week and uses that information to develop contextual analysis of ongoing behaviors, without a rules-based approach or a need to look back across stores of data to uncover anomalies.

Unlike Bitcoin, the energy consumption and potential environmental impact of utilizing second wave less sophisticated forms of artificial intelligence have yet to gain much attention. As the world continues to embrace AI in all facets of daily life, we must consider the potential environmental implications of this technological shift. Self-supervised AI presents a monumental leap forward for industries like cybersecurity and will have profound positive implications environmentally for our future.

Author Matt Shea is Head of Federal for MixMode, a next-generation, AI-powered cybersecurity platform that utilizes patented third-wave AI to help security teams dramatically increase productivity and efficiency while decreasing the wasted time, effort, and resources associated with legacy cybersecurity tools. Learn more about how third-wave AI platforms, like MixMode, are redefining the AI cybersecurity landscape through a lowered reliance on massive data stores and other energy-efficient solutions and set up a demo today.

MixMode Articles You Might Like:

Incremental Stacking of Correlative Analysis Platforms Will Ultimately Prove Ineffective and Costly

A Modern SOC Should Not Be Entirely Dependent On Human Operators and Their Personal Experience

Maximize ROI with Greater Efficacy Using Unsupervised AI

How Self-Supervised AI Tackles Ambiguity in Network Security

Performing Critical Cyber Diligence for Mergers and Acquisitions Using MixMode

Igor Mezic, MixMode CTO and Chief Scientist, Earns J.D. Crawford Prize

*** This is a Security Bloggers Network syndicated blog from MixMode authored by Matt Shea. Read the original post at: