The future of AI models & public policy

The weekly newsletter of the Business of Tech, giving you new insights into the world of IT service delivery.

Looking for stories from the podcast stories? Check out the pod itself on Apple Podcasts, Spotify, or daily in your inbox. Stories are available to everyone for five days,and Patreon supporters forever.

Was this forwarded to you? Join the list!

The future of AI models

Of my many working theories about AI (longtime readers know I have a lot of them), I’m of the belief that we need to understand the models of AI a little differently when we talk about it. And beyond the technology that goes into AI, I’m also of the belief that legislation will inevitably descend upon the industry.

To explore both of these avenues and hear what the experts in the space have to say, I had the honor of welcoming Eric Daimler on a bonus episode of The Business of Tech. As the founder and CEO of data deluge solution Conexus, Daimler’s revolutionary mathematical work has helped companies of every size manage the growing pains of the digital era. He’s even spent time in the White House under the Obama Administration, where he took on the role of Presidential Innovation Fellow for AI and Robotics.

I wasn’t going to miss the opportunity to pick Daimler’s brain, so our conversation covered everything from AI to robotics to a theoretical future where a separate AI manages both. Here’s what we got into:

Shifts in the Robotics Industry

Although AI’s impact can be felt across every industry, Daimler’s role in robotics puts him in an interesting position. I asked what kinds of shifts AI is causing in the space, and he started by pointing out that his work actually goes beyond generative/probabilistic AI, which is the type of AI drawing everyone’s attention at the moment.

He’s more involved with deterministic algorithms, which go toward zero tolerance domains where you can’t afford to mess up: airplanes, energy discovery, energy distribution, drug discovery, highly very complex supply chains, etc. And turns out, deterministic AI has actually already had its own cycle of ups and downs. When researchers discovered in the 1980s that it lacked scalability, it lost traction until more mathematical advances enabled commercial companies to revisit it.

Enter Conexus, which commercializes the software expression of those math discoveries to help large organizations with zero tolerance applications.

I’m not much of a math guy, but if you are, the math behind this stricter form of AI is category theory. I asked Daimler to explain the basics of it, and he said this:

“Category theory is like a graph theory, but multidimensional, and the benefit of that is you can add richness to any of these connections or any of these entities. The benefit of that is that you can then have contextual relationships. This equals that, but only in a certain context that equals this, only in a certain context. We then do that for enterprise-level applications, specifically application migrations and application upgrades, to guarantee the integrity under all circumstances.”

Though complex, it does help illustrate the anatomy of different models, which can help us prepare for impending implementation services.

The Sommelier Analogy

Now if you haven’t heard of the sommelier analogy yet, I really like how it frames our future relationship with AI from an MSP perspective. In short, different AI models are the grapes behind the wine, and IT pros will soon be the sommelier: in charge of matching different clients’ pallets to different AI wines.

I ran this by Daimler and asked how much he thinks the theoretical sommelier would need to know about the models, and he shared this:

“The sommeliers in that analogy would need to understand the uses and the limitations of some of the basic building blocks underlying these large language models. What I am proposing is that there are many, many, many applications that will benefit from a hybrid AI of deterministic and probabilistic AI, not one or the other.”

He’s already seeing this prediction come to fruition with a Fortune 50 client. They can’t use the mainstream forms of AI because of the unpredictability and errors, so they have no choice but to turn to deterministic AI. But there’s no denying the value of the former, so he’s working to implement a combination of both. Within the next few years, he sees a future where this is the norm.

To get specific about what we’ll have to do in this scenario, I’ll again share it in his words:

“The large language model can be a way of suggesting code for a coder. It can be a way of suggesting connections possibly between applications. It can be doing some bulk processing for suggestions, but when it actually comes time to migrate from one database version to another database version – or maintain the integrity of a business intelligence application while you’re doing a cloud migration – you can’t afford to have any errors. You need to guarantee the integrity and the visibility of the business intelligence queries that have been developed over the years while quickly migrating and modernizing in every way possible your underlying IT infrastructure. That’s where we see a hybrid AI take hold.”

So when deterministic AI becomes more readily accessible to customers, one key skill we’ll have to develop is leveraging probabilistic AI (the AI we’re all getting to know) in order to safely implement the deterministic AI infrastructures.

And, get this: Daimler believes that this process will become so complex, we’ll actually have to build AIs to coordinate between the other AIs during implementation.

A similar near-future prediction from Daimler also caught my attention – within a couple of years, he thinks that the AIs we’ll start to use as personal assistants will start coordinating with each other on our behalves. But this prediction comes with a positive twist; he also thinks that this direct correspondence will be safer security-wise than if humans tried to handle it.

The Future of AI Policy

When it comes to legislating AI, Daimler thinks that deterministic AI will play a key role in addressing the security, accuracy, and bias concerns cropping up thanks to probabilistic AI. Again, there’s a notable sense of optimism here.

Here’s how he thinks Washington will address the remaining concerns:

“We need to have a demonstrated lineage of information, a demonstrated provenance of information in many of these complex systems, and that can be authenticated with deterministic AI. Something that has been advocated recently that I had supported and still support is some degree of auditing of algorithms.”

Luckily, Daimler seemed relatively confident that the experts on the Hill will eventually be able to execute here.

As I continue down this path of finding the right insight to keep the AI conversation grounded, I’m thrilled I got the chance to hear what Daimler thinks. We brought up a lot of interesting concepts for this, so I’m curious to hear what you have to think. As always, I’m open for takes at [email protected].

That’s it for this one – I’ll be back this time next week with more from the wide world of technology services.

More from MSP Radio

Missed Things?

How about our latest videos to catch you up?

The Daily Podcast available as videos

Why providers should care about IP & Copyright, with Steven Weigler

IT Solution Provider Compensation Report reviewed with Peter Kujawa of Service Leadership

Building high bandwidth remote teams and how AI helps with Andy Tryba

Marketplaces, Influencers, Pax8 and SMBs with Shirish Nadkarni

Want the Daily News?

All the stories from the daily Business of Tech Podcast are available in the daily digest, and stories are available to everyone for the first five days, and Patreon supporters forever. Catch the audio of the show anytime on Apple Podcasts, Spotify, YouTube, or wherever you find podcasts. Links at businessof.tech

Want to change how you receive these emails?
You can update your preferences or unsubscribe from this list.