Funding Boost for NADIKI: New Support for a Sustainable AI Project

Within the field of Artificial Intelligence, one question remains unanswered: What are the environmental costs of training and inference of AI models? To answer this question the SDIA has partnered with the University of Stuttgart to bring the ability to measure the environmental impact of an AI model to the most common platform: Kubernetes. 

The work on a Kubernetes Plugin for transparent AI models is funded as part of the ‘AI Lighthouse’ program of the Federal Ministry of the Environment and Consumer Protection in Germany.

Germany’s AI Lighthouse initiative aims to make Germany and Europe a leading hub for AI technologies while promoting responsible development.

What Is NADIKI?

NADIKI is developing the ability for AI models to determine their own environmental costs within a Kubernetes cluster. This is especially relevant during the inference phase, when the model is being deployed into production environments and is being queried for decisions in real-time. Often this happens across many data centers, in different regions and on different hardware platforms.

To enable this measurement, NADIKI uses a three-stage approach: 1) mathematical estimation based on the formulas developed by Tom Kennes 2) measurement using hardware & software-based APIs and 3) connecting to the actual data center and retrieving external power readings, information about the hardware and the energy being used.

We aim to enable a holistic view of environmental impact – from energy used & re-used, to the server & network equipment involved, all the way to the water use, cooling systems and the data center facility itself. 

The integration will enable the transparency of AI models, especially during operation, and can be used to set new environmental standards for AI.

As with all work undertaken by the SDIA, all the results will be open-source and available for anyone to use.

Improving the Transparency of AI models & AI-powered applications

Without data available within the infrastructure of an application, it’s impossible for developers & operators to understand the environmental impact of their AI models or AI-powered functionality. With NADIKI it will become possible (and easy to do) to measure, monitor and report the environmental costs – from energy to resource usage and emissions.

Based on the metrics that NADIKI will make available, here are some ways how applications can be run:

Maximize server resources: Use existing infrastructure to its full capacity, so new data centers or ICT equipment aren’t built unnecessarily.

Understanding actual resource consumption: Signals from the physical infrastructure can be used to optimize training or stop it altogether if resource usage isn’t meeting targets – e.g., only run trainings and updates when renewable energy is available.

Make informed decisions on and where to compute: Shift training to a different location or change the speed of the training process. Or choose to run AI applications at a time when there is free capacity or when it’s best from a cooling perspective.

What’s New About NADIKI?

Most importantly, NADIKI will make environmental impact metrics available to AI models & applications which are running on Kubernetes. 

Together with the University of Stuttgart, NADIKI will provide – for the first time – a data collection software for the data center facility, which provides a unified API to Kubernetes, through which information can be retrieved, e.g., what is the current energy mix/how much renewable energy is available? Are the diesel generators running? How much water is being used right now to support the cooling system efficiency? Is there an outage in the facility?

Within Kubernetes, NADIKI aggregates the information from the facility, the underlying server infrastructure, and the orchestration platform itself, enriches them, and delivers all the relevant KPIs to the application and monitoring systems.

With this NADIKI will provide a new level of accuracy, all wrapped in an easy-to-use integration which gathers information via many paths, always delivering the ‘best available’ data on the environmental impact to the application.

NADIKI uses a Life Cycle Assessment approach to measure the actual resources consumed. This includes elements such as regional electricity generation, whether recycled IT equipment has been used, or if waste energy is being reused.

NADIKI can use both virtual and physical IT infrastructure, integrating the data into one system. In the past it’s been difficult to track the resources used by virtual servers. NADIKI will make it easier to measure overall consumption for an individual AI application.

All of the software – both for Kubernetes and the digital infrastructure itself - will be available open source and can be used within existing IT infrastructure. And of course, we will build upon all the great tools which are already out there. With more data comes more transparency and the potential for innovation across the entire sector.

“Artificial Intelligence for the Environment and Climate”

The AI Lighthouses are part of an initiative funded by Germany’s Federal Ministry for the Environment (BMUV) to use AI to meet ecological challenges. The project’s sponsor is Zukunft-Umwelt-Gesellschaft (ZUG) gGmbH.

Funding was presented to the NADIKI team by Christian Kühn, Parliamentary State Secretary for the BMUV and Corinna Enders, ZUG’s Managing Director at an event for AI Lighthouse projects from the Baden-Württemberg region.

At the event, Christian Kühn noted that Artificial Intelligence has enormous potential for helping us to live more sustainably, but we must also find ways to reduce the resources that it consumes.

We’re incredibly excited NADIKI has been given AI Lighthouse funding and can accelerate its progress in helping to achieve this goal.

Congratulations to the other three AI Lighthouse projects from the Baden-Württemberg region. Take a look at them here:

Using AI to optimize the recycling of small electronic devices – DESIRE4ELECTRONICS

Using AI to create climate-neutral sewage treatment plants – KikKa

Using AI to increase the recycling rate of plastic waste – RecycleBot