When AI Takes Action: Thoughts from the Main Stage at the U.N. Global Compact Leaders Summit

I recently had the opportunity to speak on the main stage at the U.N. Global Compact Leaders Summit in New York City, as global leaders gathered for the U.N. General Assembly and Climate Week.

I talked about what happens as AI becomes more powerful and increasingly autonomous. I started with a story.

A man in Australia asked an AI agent to help him get off the waitlist for a popular class at his gym. The agent found a vulnerability in the booking system and removed the person ahead of him from the list. The man hadn’t asked it to do that. And when he realized what had happened and told the agent to put the person back, it couldn’t.

It’s funny at first. But the AI didn’t simply provide information. It took an action its user hadn’t authorized, and that action affected someone else.

We’re moving from AI that primarily answers our questions to AI that can increasingly act on our behalf. The possibilities are extraordinary. But the more AI can do, the more important it becomes to think about what sits underneath it.

Start with the data

A lot of the AI conversation has centered on models. But as AI starts taking action, another question becomes even more important:

What does it have access to?

If an AI agent is going to act on behalf of a person or a business, organizations need to know where their data resides, who and what can access it, and what an agent is permitted to do with it.

Privacy, security, and governance need to be part of the architecture from the beginning, not something added later.

That thinking is behind our Principles for Responsible AI at EDB. We believe AI should be sovereign, governed, trusted, and beneficial. That belief influences how we build, with a focus on giving organizations greater control over their data and how AI interacts with it.

This isn’t about slowing AI down. A stronger foundation should make it easier for organizations to take advantage of what AI makes possible.

How much more can we get from what we already have?

As AI grows, so will demand for infrastructure and energy. Before the solution is “more servers, more computing power, and more energy,” we should ask how much more we can get from what we already have.

We’ve been doing that at EDB.

We looked at deployments across three large financial services customers operating more than 120 data centers, and we had the work independently validated. In one case, optimizing the underlying infrastructure could result in up to 94% fewer compute cores and up to 87% expected emissions reduction.

The larger point is that efficiency and performance don’t have to work against each other. Better use of existing infrastructure can improve performance while lowering costs, energy use, and environmental impact.

At EDB, we think about this in terms of “intelligence per watt.” If we’re going to dramatically increase what AI can do, we should also be asking how efficiently we can power it.

We don’t have to wait for every answer

One of the things I value about EDB’s involvement with the U.N. Global Compact is hearing how other companies and leaders are working through these same questions. I serve on the Global Compact’s Legal Council and recently joined its Think Lab on AI for Business Integrity. Companies have a lot to learn from each other, and we should be sharing what works and what doesn’t.

Regulators have an important role to play. But technology is moving too quickly for businesses to wait for regulation to anticipate every possible use or consequence.

We can act now by giving organizations greater control over their data, putting governance around how AI interacts with it, and being smarter about the infrastructure and energy required to power it.

At EDB, we’re working through those questions through our AI governance work, our Principles for Responsible AI, and the technology we build.

Which brings me back to the gym waitlist.

Today, an AI agent removing someone from a gym class makes for an interesting story. As agents become more capable, the decisions they make and actions they take will have much higher stakes.

That shouldn’t make us less excited about AI. There is too much potential ahead for that. 

It should make us think carefully about the foundation we’re building for it. 

Because the more AI can do, the more that foundation matters.