AI Blog The environmental costs of Al

The environmental costs of Al

AI Blog

Issue #24

13/09/2023

The environmental costs of Al

AI, despite its multiple applications and positive impacts in many areas, brings with it significant environmental costs, as it comes with huge computing power requirements that place a significant strain on resources and our planet.

Creating an LLM requires pattern analysis on a huge set of human text. All of these calculations require a lot of electricity and generate a lot of heat. So data centers, to keep cool, have to pump water – often to a cooling tower outside their warehouse-sized buildings.

In its latest environmental report, stated that its global water consumption increased by 34% in 2022 compared to 2021. Similarly, Google in its own report reported a 20% increase in water consumption for 2022.

In Iowa, where Microsoft has some of its data centers and where OpenAI’s GPT-4 training took place, water is used for cooling only when the weather reaches 30 degrees Celsius. In July 2022, about 44 million liters of water were needed. This is 6% of the total consumption of the entire region.

Somewhere here we must stress that we are referring to clean, drinkable water!

For example, 5 questions in ChatGPT correspond to about 500ml of water. Our opinion is that AI should not endanger basic goods such as drinking water.

The tech giants (Microsoft, OpenAI and Google), recognizing the problem, mainly due to the intense public awareness and criticism they receive, are nevertheless significantly delayed and their efforts lack transparency, making many wonder whether they will prioritize real sustainability or stick to greenwashing solutions.

The UAE join the AI race

The Technology Innovation Institute (TII), based in the United Arab Emirates, has developed the Falcon-180B. It is a large language model – AI, which has been trained with a huge dataset of 3.5 trillion tokens using thousands of GPUs. It is a particularly important milestone for open and publicly available AI.

The Falcon-180B performed exceptionally well in the various nlp tasks and benchmarks in which it was tested and in several cases outperformed other GPT-3.5 and Google’s PaLM 2-Large.

It is also open source and can be found on Hugging Face. However, commercial use has certain restrictions and requires a license review.

Resources

AI Insider 📰

Following in Adobe’s footsteps, η Microsoft has announced that it will offer legal coverage to companies facing lawsuits over their AI products. It’s a very big leap of faith for both companies, considering that GitHub Copilot, Midjourney and OpenAI are getting whistleblowers left and right. It makes a huge difference because still the copyright field is extremely murky.
- In addition, Microsoft is working with Paige to create a new AI model that could revolutionize cancer diagnosis and treatment.
Since we are in an election period, we should inform you that Google requires disclosures for political ads that use AI.
- Google also announced the launch of the Digital Futures Project, a program designed to foster dialogue around the benefits and challenges of AI – setting aside a $20 million fund to award grants to select research centers and academic institutions worldwide.
Meta plans to train a new AI model that it hopes will outperform OpenAI’s GPT-4. Training of the model is scheduled for early 2024 and it looks like it will also be open source.
The Roblox AI assistant allows creators to create virtual items and backgrounds with simple prompts.
Ghostwriter submitted “Heart on My Sleeve” as a Grammy nominee in the categories of best rap song and song of the year. According to Harvey Mason Jr. (chief executive of the Recording Academy, the organization behind the Grammy Awards), the song is “absolutely eligible” because it was written by a human being.
The US IRS will begin utilizing AI tax evasion investigations on multi-billion dollar partnerships, including hedge funds, private equity groups, real estate investors and large law firms.

Learning Bytes 🧐

Time has just released their TIME100AI list of the most influential people shaping the AI sector. The 100 people range in age from 18 to 76 and are divided into four categories: leaders, innovators, shapers, and thinkers. Time’s selection process was designed to map the key relationships and power centers driving AI development.
Designers at Collina Strada (fashion house) trained an AI model on the house’s previous clothing designs to create the new spring collection. The team then adapted the designs and created tangible versions of the designs, which were shown at New York Fashion Week. The clothes have an AI feel – ruffles and lace seem to appear out of nowhere, as AI sometimes does when it adds an extra finger to the arms.
NVIDIA has entered into AI-focused partnerships with Indian giants Reliance Industries and Tata Group to upskill 600,000 employees and serve more than 600 million customers with the help of AI.

Cool Finds 🤯

We recommend you to follow Tatiana Tsiguleva στο Twitter/ X. She gives really valuable tips and she keeps it real with what is the process of producing a good result.

See you next week! 💚

#TogetherWeAI