AI Ethics & Policy Insights

The AI ethics & policy info we should know - simplified. Week of November 13 , 2023. Issue 16.

Nov 16, 2023

DoD Responsible AI Toolkit
- The DoD released its Responsible AI Toolkit on November 15th, 2023.
- Toolkit builds on existing RAI and AI ethical frameworks to provide a process that helps to identify, track and improve process alignment across AI projects.
- This was released and will be leveraged by the DoD but is available for public use.
- It includes the SHIELD Intake, a comprehensive and interactive intake form to evaluate and document the AI being ideated, build, used or deployed.
- It also includes the Defense AI Risk Guide (DAGR) is meant to provide a holistic approach to risk evaluation. It’s aligned with the NIST AI Risk Management Framework and aligns to seven widely held AI principles.
- One feature I love is how user friendly this is. As you can see below, you can input your responses directly into the tool so it’s all in one central place. Most RAI toolkits do not have this feature, and I think it’s a game changer.
GITHUB COPILOT ENTERPRISE and GITHUB COPILOT WORKSPACE
- GitHub Copilot Chat goes GA in December. Copilot Chat sits inside of the developers IDE and allows them to ask questions, explain code, translate code and propose fixes, among others. GitHub Copilot chat will now be available beyond the IDE, and you can use it on GitHub.com and in the CLI.
- GitHub announced GitHub Enterprise, which will be available in February 2024 for $39 per user, per month. This is an enterprise-grade Copilot subscription which includes the ability for companies to personalize Copilot chat for their codebases to get more contextually-specific code.
- GitHub showed a brief glimpse of GitHub Copilot Workspace (43:42), which is the most true example (in my opinion) that we’ve seen of human/AI partnership. Copilot optimizes the developer workflow. With any GitHub issue, Copilot Workspace automatically proposes a solution based on its understanding of the code and context surrounding the issue in question. It can build plans for how to execute on the proposed solution, and you can click “implement” to have Copilot Workspace do it for you, while also building and running tests to validate the changes.
MICROSOFT IGNITE
- Ignite is still ongoing! The biggest release so far is Microsoft’s new AI chip, and I’ve done a deep dive below!

Microsoft’s New AI Chip

Alt title: New Chip, Who Dis?

The Context:

Microchips are critical for building technology - this we know. Each chip is a set of electronic circuits on a small piece of silicon (hence Silicon Valley) and the transistors act as mini electrical switches that turn the current on or off. These chips are central to nearly every electronic device in use today - phones, cars…even some “smart” toasters.

A Microchip-Sized Microchip Primer:

To understand how critical Microsoft’s announcement is, we need understand that not all microchips are created equal. You may have heard of Central Processing Unit (CPU) and Graphics Processing Unit (GPU) chips. They’re both important and they serve different purposes, but they’re both for general use.

A CPU chip is designed to handle the tasks required for the software on the server to run correctly and is often consider the “brain” of the computer. It’s designed for a wide variety of tasks and workloads.

The GPU chip helps with graphics related work. It performs mathematical calculations very quickly to render graphics and images. The GPU can break a task down into several tasks simultaneously, which is what makes its processing so quick. GPU chips can also perform some AI-related workloads. In fact, GPUs were used for AI computing before there was a demand for stand-alone AI chips.

The AI Chip was introduced after the rise in cloud computing gave AI a boost. In 2015 the tech industry began creating dedicated AI chips to perform AI functions and tasks, which increased the capability of AI 10-fold. Before the AI chip was introduced, cloud computing companies were running AI on a mix of CPU and GPU chips. After 2015 it became clear that companies that wanted to do more with AI and empower their customers to do more with AI needed these AI-specific chips. Leaders in this space include Google’s TPU chip, China’s Cambrian deep learning processor chip, and Nvidia’s HGX H200 chip.

Microsoft’s announcement:

On November 15, 2023 Microsoft announced at its Ignite conference that it’s now in the microchip-making business. It announced the creation of its first two customer silicon chips - the Azure Maia 100 AI chip and Cobalt 100 CPU chip. This is massive for Microsoft as they move into a more empowered position as the leader in AI, and this move will allow Microsoft to prepare their datacenters for the future of AI and provide the best experience for their customers.

Microsoft is already testing the chips inside the company, and at OpenAI.

“The new Azure Maia AI chip and Azure Cobalt CPU are both built in-house at Microsoft, combined with a deep overhaul of its entire cloud server stack to optimize performance, power and cost. ‘We are rethinking the cloud infrastructure for the era of AI, and literally optimizing every layer of that infrastructure.’ says [Rani] Bokar.” (Bokar is head of Azure hardware systems and infrastructure at Microsoft.")

Why it matters:

Right now, it appears that Microsoft relies on Nvidia to provide chips. Back before the microchip shortage when the living was easy, getting chips for cloud computing and AI wasn’t as challenging (and costly) as it is today. Now that these chips are in high demand and there are only a handful of reputable AI microchip developers, it’s getting more and more challenging to get the AI chips. It’s clear that Microsoft decided to enter the game to decrease their business risk.

This is a huge win for Microsoft because it will help keep their costs lower, and provide more control in the AI space. Not only can Microsoft use their own chips, but they can help meet the demand for AI chips that will undoubtedly rise in the next few years, which is a very profitable path for the tech behemoth. This chip-purchasing process will likely give Microsoft insight into up and coming AI companies, as well, so I think this is looking like a strategic win.

The ethical implications:

It’s not a shock that the manufacturing of these chips has an impact on the environment. Silicon is made from sand, which is plentiful. But the process for creating the silicon chips requires a lot of water and energy, and uses harmful chemicals. Microsoft didn’t go into detail about the environmental impact of making the chips and if it would impact their environmental commitments to be carbon negative, water positive and zero waste by 2030, and by 2050 removing from the atmosphere the equivalent amount of carbon dioxide the company has emitted since its founding in 1975. Still, the creation of the chips for technology to be used primarily in the developed world will have an impact on those in lower income environments and developing regions through climate change and inaccessibility.

Where we go from here:

Keep an eye on Microsoft to see how it moves forward in overhauling its infrastructure for the age of AI, and keep an ear out for whispers of the company leveraging renewable energy for the creation of the chips “in house”.

Sources: Jotrin, Reuters, The Verge, Forbes

AI Policy Updates

UK AI Safety Summit:

Last week I did a deep dive on the wins and worries of the UK AI Safety Summit. You can read all about it here. It’s important to know that:

China was in the house, a move that was widely criticized.
Attendees signed the Bletchley Park Declaration on behalf of their country resulting in 28 signatures, including the EU.
A report will be created every year outlining the “State of AI Science” as an output of the summit. This report, while a collaborative effort, will establish UK thought leadership on the current and future trends and needs in AI.
The next AI Safety Summit will be in South Korea in 2024.