Green washing- It can cost economically, page-35

A/Prof Lucinda McKnight chat about AI and Education tonight was fun.

Pros and cons of it

Huge level of Bias build it, Huge $$$$$$$$ going to just a few, Data skimming read out right theft and more. One example she used asking a Generative AI tool to right a test article used a significant amount of her own work from news articles and her published work. It did not credit her of course.

I asked if the schools and universities can sandbox AI tools to students? Her answer as expected was “The Cats out of the Bag and there is no way to stop the juggernaut of changes in Education and Generative AI”.

News Ireland’s data centres are already using more power than all the households in Ireland today was not really surprising either

A/Prof Lucinda McKnight also used one of the first computers used in a school in 1981! The Entire school energy consumption at that time was less than 5% of todays power used.

Wind makes up about 30% of Irelands grid power.

As we are on “Generative AI power consumption” I asked AI. Copilot said

Generative AI systems, such as those used for creating text and images, have a significant impact on power consumption. Here are some key points:

1. Energy Usage: It’s estimated that a search driven by generative AI uses four to five times the energy of a conventional web search¹. Within a few years, large AI systems are likely to need as much energy as entire nations¹.
2. Data Centers: The power consumption of data centers, which host these AI systems, has been steadily increasing. Generative AI contributes to this rise, but it’s part of a broader trend of growing data center energy use².
3. Environmental Impact: Generative AI systems also require enormous amounts of fresh water to cool their processors and generate electricity¹. For example, a data center cluster serving OpenAI’s GPT-4 model used about 6% of the district’s water in West Des Moines, Iowa, in July 2022¹.
4. Comparative Consumption: Generating an image using a powerful AI model takes as much energy as fully charging a smartphone³. However, generating text is significantly less energy-intensive³.

The environmental impact of generative AI is a growing concern, and there are calls for more sustainable practices and technologies to mitigate these effects.

Is there a specific aspect of generative AI’s power consumption you’re particularly interested in?

Google version was out to lunch Gemini is currently on a break.

Please give it a moment and try again later.

Tried again and it gave

“Generative AI's Power Consumption: A Growing Concern

Generative AI is incredibly energy-intensive. Training large language models (LLMs) like those powering ChatGPT requires vast amounts of computational power, which translates to significant energy consumption.

Key Factors Affecting Power Consumption:

• Model size: Larger models require more parameters and computations, leading to higher energy consumption.
• Training process: Training involves iteratively adjusting model parameters, consuming substantial energy.
• Inference: Using a trained model to generate text or images still requires computational resources, though less than training.
• Hardware efficiency: The type of hardware used, such as GPUs or specialized AI chips, impacts energy efficiency.

The Scale of the Problem:

• Massive energy consumption: Training some of the largest language models can consume as much energy as several hundred homes for a year.
• Growing demand: The increasing popularity of generative AI is driving up energy consumption.
• Environmental impact: The carbon footprint associated with AI training and usage is a growing concern.

Efforts to Reduce Power Consumption:

• Hardware advancements: Development of more energy-efficient chips and processors.
• Algorithm optimization: Improving the efficiency of AI models and training processes.
• Renewable energy: Utilizing renewable energy sources to power data centers.
• Carbon offsetting: Investing in carbon reduction projects to compensate for emissions.

Challenges and Future Outlook:

• Balancing performance and efficiency: Improving energy efficiency without sacrificing model performance is a complex challenge.
• Transparency: The energy consumption of AI models is often not publicly disclosed, hindering efforts to address the issue.
• Regulatory landscape: Developing clear guidelines and standards for AI energy consumption is essential.

While the energy consumption of generative AI is asignificant concern, ongoing research and development are focused on mitigatingits environmental impact.

Would you like to delve deeper into any specific aspectof generative AI power consumption, such as its impact on the environment,potential solutions, or the role of regulation?

Would you like to see some statistics or data on theenergy consumption of specific AI models?

”

Chat GPT gliched out – Not enough power or lag perhaps?

Try 2 worked

“Generative AI models, especially large-scale ones like GPT-3, consume significant computational resources, which in turn translates to substantial power consumption. Here are some key points regarding the power consumption of generative AI:

1. **Computational Intensity**: Generative AI models, particularly those based on deep learning architectures like transformers, require extensive computational power to train and inference. This includes GPUs (Graphics Processing Units) and TPUs (Tensor Processing Units) which are much more power-hungry compared to traditional CPUs.

2. **Training Phase**: During the training phase of a generative AI model, vast amounts of data are processed repeatedly through complex algorithms. This process is highly parallelized and requires many computing units working simultaneously, which increases power consumption.

3. **Inference Phase**: Even during the inference phase (when the model is used to generate text or respond to queries), although less intense than training, generative AI models still require powerful hardware to perform computations quickly and efficiently.

4. **Environmental Impact**: The power consumption of large-scale AI models has raised concerns about environmental impact, particularly in terms of carbon emissions and energy usage. Companies and researchers are increasingly focusing on optimizing these models to be more energy-efficient.

5. **Research and Development**: Efforts are ongoing to develop more energy-efficient algorithms and hardware specifically tailored for AI tasks. This includes advancements in hardware architecture (like more efficient GPUs and TPUs) and algorithmic optimizations (such as pruning redundant connections in neural networks).

6. **Deployment Considerations**: When deploying generative AI models in real-world applications, considerations should include the energy footprint of maintaining and running these models over extended periods.

Overall, while generative AI brings tremendous capabilities in natural language understanding and creativity, it also comes with significant energy demands that need to be addressed for sustainable development and deployment.”

AI education is interesting from many angles.The Cats Out of the Bag for energy use growth we will see sooner than later isalso clear. While some will get the clear issues and move to all possiblesmall, medium and large-scale carbon power generation sources being needed nota competition between them.