NAMRATA-WOKE
commented
11 months ago Hi @ursvill can you support on this front? It's also been requested by our panelists and SC members.
Panel recording: https://www.youtube.com/watch?v=0CX9J3oAipM
List of remaining questions (let's talk about how we want to address these!), and some comments:
YouTube chat CVS (please access using your GSF email)
Mark Butcher: What’s your view on cloud providers like AWS and their “carbon reporting tool”? Do you think customers building AI platforms will be misinformed by that dashboard
Mark Butcher: Will potential fines be enforced and will they be substantial
Mark Butcher: How can companies truly build sustainable AI platforms when none of them understand how to measure emissions?
Ben Gurney: Does the entire paradigm of these model being so power hungry not mean that they are not sustainable in their current form?
Molly Anglin: To Chris and Naveen - both Avanade and Accenture presumably have a strong imperative to sell more consulting services involving AI, Do you bring the environmental considerations into your client conversations consistently? How?
Ben Gurney: Until the models themselves become less power hungry and more efficient to train is there any hope of AI being sustainable?
Atiyah: In terms of quantifiable metrics, how much potential is there if AI is conducted in an environmentally friendly way (from hardware and software perspective) if possible?
Tim Honisett: Are there defined software engineering methods for developing greener software?
Jiri Skopek: How important is the prompt engineering for improvement of AI efficiency?
Jon Clark: Data storage and AI will function as distinct components in this new AI architecture, with data storage handling the storage and retrieval of information while AI processes and analyzes this data to derive insights and make informed decisions. So new designs and optimizations will be required.
Ben Gurney: Can building AI with low energy requirements be a primary goal of the design process for a particular model?
kunalmcitm: How to ensure that implementation of Gen AI, leading to overall savings, does offset the reduction in emissions overall (net emissions). Can you please shed some light on the same?
Oliver Cronk: Isn't a lot of the challenge with AI that we are often solving problems with brute forcing and massive amounts of data and compute (particularly with general / foundational models). Is part of the answer doing the hard work of creating a specific model for the problem you are looking to solve and not reaching for the easy answer?
Houleye SIEGEL, Executive Director at JPMorgan Chase & Co.: I agree with Elif, one way in addition to building carbon aware applications is also to give the transparency to the end user on energy used in function of functionalities used and enable them to adapt their usage (energy efficiency mode,…)based on those metrics, i.e have end user awareness regarding energy consumption.
Oliver Cronk, Tech Director (Sustainable Software) at Scott Logic: Isn't a lot of the challenge with AI that we are often solving problems with brute forcing and massive amounts of data and compute (particularly with general / foundational models). Is part of the answer doing the hard work of creating a specific model for the problem you are looking to solve and not reaching for the easy answer?
Jia Yi Chow, Software Engineer | Ex-PayPal: Software will inherently emit carbon, its all about making it minimal and really use AI sparingly
Raymond du Toit: This groups is seeking make AI a useful tool to address some of the challenges: https://aiforgood.itu.int/
Mathieu Wolff, Consultant Numérique Responsable: What about AI's rebound side-effects ?
Jon Clark, Data Centre Sustainability Lead: Green AI, site and self-generation of power and utilization of heat and also factoring data sovereignty will be key as we move forwards with designs and AI builds. But this is not a quick fix and need a thought-out designs and builds.
Mark Butcher, Digital sustainability & GreenOps advocate: Green AI is not possible. It’s that simple. Just own the impact of your actions and report honestly with transparency
Ben Gurney, Engineering Manager at Container Solutions: Can building AI with low energy requirements be a primary goal of the design process for a particular model?
Mark Butcher: No platform has negative emissions - that is greenwashing pure and simple
Jon Clark, Data Centre Sustainability Lead: Data storage and AI will function as distinct components in this new AI architecture, with data storage handling the storage and retrieval of information while AI processes and analyzes this data to derive insights and make informed decisions. So new designs and optimizations will be required.
Mark Butcher: PUE is not a good metric anymore, it’s abused too much by the data centre industry
Mark Butcher: How do you measure the carbon impact whilst you shift regions? Given the size and cost is that even a practical idea? Nice in concept but not actually realistic at scale.
Mark Butcher: Don’t use queries per Watt, you’ll be missing up to 90% of your emissions and will make incorrect decisions as a result.
Jon Clark: How does the panel look to measure, manage and specifically focus on Scope 3 emissions in the context of AI-driven technologies and data centre operations, and what innovative strategies or technologies are emerging to address these indirect environmental impacts?
Raymond du Toit: Interesting comparison of AI models: http://ai-carbon-efficiency-observatory.ai-for-sdgs.academy/#emission_chart
Ben Gurney: Until the models themselves become less power hungry and more efficient to train is there any hope of AI being sustainable?
Ben Gurney: Does the entire paradigm of these model being so power hungry not mean that they are not sustainable in their current form?
Mark Butcher: How can companies truly build sustainable AI platforms when none of them understand how to measure emissions?
Mark Butcher: Will potential fines be enforced and will they be substantial
Mark Butcher: What’s your view on cloud providers like AWS and their “carbon reporting tool”? Do you think customers building AI platforms will be misinformed by that dashboard?
ursvill
Jenya-design
Article Details
Please enter the details here
Proposer: Namrata Narayan | Dir. Communications & Member Relations | @NAMRATA-WOKE Abstract: Recap of the panel, including key insights and observations and answers to questions Why: To continue the dialogue and build actions around the environmental sustainability of AI Audience: Software Engineers Timeline: October 23
Important Information
To be filled out later
Author: @ursvill Other contributors: (Experts, Reviewers) Article Link: (Link to the doc, must be in our official GSF drive) Graphics Link: (Link to the design assets for this article, must be in our official GSF drive) Published Link: (The link the article was published to for memory)
Checklist