While the future of AI seems endless, the reality is that this important tool may soon face roadblocks. Ten years ago, two Oxford professors published a seminal article about the future of AI in the marketplace (claiming, for example, that 47% of jobs in America eventually could be automated). Recently, those two professors published another article specifically addressing the limitations of AI a decade after that initial assessment.
Data Pools
The authors note two important obstacles for generative AI: first, the “learning data” the AI engines use is unlikely to change in significant ways. Think about it like this: generative AI learns from the giant pool of data available on the internet. Say that you want an AI engine to write a blog for you. The blog content and style will be based on what that AI engine has learned from that giant pool of data (matched with your input / request).
Moving forward, however, there may not be huge shifts in the pool of data. Sure, more will be added to it, and there will be improvements, but those improvements look like they will be at the margins rather than fundamental or significant shifts. At the same time, it should be noted that this conclusion may be subject-dependent, as many AI engines are still in fairly early stages of production.
Energy Consumption
As for the second obstacle, the energy demands required for larger scale AI implementation could be limiting. Bottom line: widespread, large-scale AI computing requires massive amounts of electricity. Several solutions exist to this problem. First, engineers might find ways of making AI usage more energy efficient. But that doesn’t look to be on the horizon yet. Second, companies conducting AI computing may just swallow the massive costs as use increases (or pass it on to consumers), but again, that doesn’t seem like a feasible solution, considering the estimates.
Finally, AI engines may be engineered to “learn” from smaller data sets. This seems a distinctly possible solution, but as you can imagine, this outcome is concerning as well. Relying on smaller data sets (limited input) might mean skewed output. If you ask an AI engine for a blog post, and it is relying on a limited data set, then the product it gives you has a higher chance of being flawed or inconsistent.
A notable illustration of how serious this energy situation is can be seen in the firing up of…Three Mile Island? That’s right – the location of one of the most dangerous nuclear events in history is coming back on line because Microsoft needs more power for its AI ambitions.
Or, consider that Georgia Power, having just built two new nuclear reactors at Plant Vogle, still plans to build more fossil fuel-powered plants because of increased demand. The AP attributes this increased demand to computer data centers.
Large Scale Efficiency
Related to these two obstacles is the problem of scale. One article discusses the delayed deployment of Grok3 and other next gen AI engines, arguing that “these delays signal potential limitations in the scalability of existing AI technologies, as increasing computations and data volumes are not translating into anticipated performance gains.”
If you combine these three problems – data pools, energy consumption, and scalability – we can see a situation in which AI innovations may be stalled.
How this Impacts Business Owners
Much of the AI and consulting class sells AI FOMO (Fear of Missing Out). You are being told “this is the year for AI THING X” and if you don’t do it, well, you’re behind the 8 ball. Recent news articles have attributed job losses to AI implementation, but the reality is more complicated than that. While AI has contributed to this trend, especially in particular fields.
Two things can be true at the same time: AI HAS revolutionized many aspects of business, BUT it still has significant limitations. You SHOULD be exploring how AI engines can help specific aspects of your business, from research to repetitive tasks and workflow automations.
But AI is not some magic bullet. At this point, agentic AI (think: semi-autonomous AI) isn’t robust enough to complete the critical thinking and physical tasks which your employees are doing. Instead, encourage your employees to explore how AI might make their own jobs more efficient.
Earlier this year, we heard a number of the AI prophets proclaiming that 2025 was the “year of agentic AI.” That hasn’t come to fruition – at least in a way that would directly and significantly alter how SMB owners operate.
So, again, YES, explore how AI engines can help you. Encourage your employees to find ways of saving time and effort with AI. Allocate some time to “play in the AI sandbox” for yourself. Keep abreast of coming tools and capabilities. Think creatively how you could use AI.
But don’t get the AI FOMO fever which implies you are being left in the dust if AI isn’t running your business. View it as you’d view any innovation in your field. See the positives and negatives and limitations clearly. Efficiently work it into your business model…but without a rushed panic.