There are plenty of reasons to get concerned, really concerned, about what’s going on in AI. There seems to be a growing disconnect between AI press releases and stories, and what enterprises are telling me. That disconnect seems linked, as is often the case, with the concept of “washing” announcements with mentions of new technology when the premise of the linkage is tenuous at best. We may be fueling an AI bubble, based not on what AI could really do but on how exciting you can make it sound. That’s bad, but there’s also encouraging news.
What is the core value of AI? The answer is that it can do stuff a person or group could do, but faster. “Artificial intelligence” is the key term; maybe our use of the acronym has made us forget. The point here is that AI does what we do, but faster/better. We can enter something into a search engine and AI can digest the results, summarize the points if you like. I get AI results all the time when I search, which is often. I sometimes find them helpful, sufficient for my purpose, but most of the time I need to look at actual search results to get the details I need. Is this use of AI transformational? Not to me, nor to enterprises I talk with. Businesses don’t run on the Cliff Notes of economic activity, they run on the details. The most common use of AI fills our appetite for instant answers, but that’s not going to be enough to justify massive investment.
Where do enterprises find AI helpful? In most cases, what they really like is something like the notion of the AI agent, which enterprises believe is an AI model that’s trained to operate on a very specific set of information. While we hear about agents mostly in the context of autonomy, meaning AI acting alone, enterprises are generally not comfortable with having AI act on information without human supervision. So the value here is specialization, and the reason that’s valuable is that AI can quickly analyze something that a person or people would take longer to analyze, and in applications where time is critical, AI offers a real advantage.
Enterprises say that they also like AI in business analytics missions, because AI can spot patterns that people simply take too long to find. Do enterprises believe that their staff is incapable of analyzing the same data and reaching the right conclusions? No, but they think AI could do it faster and perhaps (in its agent form, not its generative form) more reliably. Can I do my taxes? Sure. Could an AI tax agent do them better? Sure, but so could a CPA. AI agents offer speed and specialization.
CIOs are getting fairly strident in their rejection of the popular “copilot” form of AI, which they classify as a kind of attempt to popularize AI and dodge actual business-case scrutiny. One told me “We have thousands using AI to help them write emails or maybe short memos. Tell me how this does anything for the company. What’s driving it is that as-a-service AI is expensed, and most companies, like us, don’t evaluate line-department use of technology delivered that way. If we did, we’d probably crush it out.”
All of this seems validated by recent comments by Broadcom. The new-age chip giant says that there’s a sea change underway in the AI space, a shift away from GPUs to specialty chips designed for machine-learning applications that sure sound like agent applications to me. If true, it could be the first silicon signal that AI focus is shifting away from the hosted chatbot model to something enterprises have said they favored.
OK, so what does this mean? I contend it means a lot of what’s claimed for AI doesn’t stand up to what those who’d have to invest in it consider a realistic assessment. I read an article early this week that claimed that AI would demand fiber and that telcos were eager to see that. Will AI demand fiber? Surely it will inside clusters of AI-GPU servers, but in the network? Our first example of AI, the typical search-enhancement example, may be nice to get an answer to a simple question, but how much is that worth? Remember my comment about running a business on Cliff Notes? Same with running a part of one, or a network. Given that, how much traffic does AI generate outside its own cluster? Enterprises have told me from the first that there is no impact on network traffic created by AI outside the AI cluster and training connections.
What would make AI transformational? Data, and more specifically, real-time data. We run businesses and enhance productivity, buy and sell products, based on information very similar in terms of timeliness to what we had when it was punched onto cards in Hollerith code. AI value demands we shift not how we process things as much as shift what things we process. Getting that real-time data to AI could increase network demand. AI could enable applications that, without it, would be difficult to create, and the running of those applications and/or review of the results could generate traffic, too. The business value of those applications could create benefits to justify investing in them, and in their traffic handling.
A lot of the value of AI, then, is linked to growth in IoT. It’s not so much in what gets media and PR attention—things like autonomous vehicles—as it is simply exploiting real-time information about business processes, not simply recording the result of those process. A sale, for example, might be a single record in the traditional handling of business results, but it might be multiple steps in real time. As real-time processing is integrated into the work itself, it generates more data and also has the potential to impact the productivity of the worker more directly.
The problem is that an IoT-real-time AI approach crosses a lot of technology lines, and few vendors are in a position to profit from the whole food chain. Given that, will any vendor see enough benefit to drive its own portion forward, especially when other essential elements may not be provided by the vendors responsible for other related technology spaces? Enterprises seem to think that progress here will come from a vendor willing to frame a kind of IoT/AI platform, and they name three candidates—Broadcom, HPE, and IBM.
I think the situation with AI is hopeful. Despite a major wave of hype on applications enterprises don’t think will make a business case, we’re seeing enterprises dig through the hype to find actual, valuable applications. We’re also seeing some companies talk about AI reality in a forum that really matters, Wall Street. Good things may be on tap for 2025.
