Today we’re witnessing an interesting contradiction unfolding when it comes to AI – it is both overhyped and underhyped. There is an unprecedented wave of eagerness to capitalize on this technology, which thanks to the “coming of age” of foundation models has now reached a new, significant maturity threshold. While the excitement is partially driven by the success of pilots and sandbox tests that tease the material impact AI can deliver, to date only few companies have deployed successful generative AI (gen AI) use cases at scale and are seeing the positive impact of that investment on their bottom line. Many remain in the experimentation phase – for now.
But why is that? Complexity roadblocks.
Much of today’s AI fever isn’t grounded in a clear and fundamental understanding of the technological breakthrough, or its implementation demands. And that introduces risk and missed opportunity for enterprises seeking to scale AI competitively and responsibly.
Add to that, the hype has led to a false perception of AI as effortless, which couldn’t be farther from the truth. Simply accessing foundation models doesn’t unlock business outcomes. The real challenge lies in building robust enterprise applications on top of them. Behind a seamless AI application is an immensely complicated technology stack that developers need to harness – that’s where the magic happens.
This is no small feat either. There is a whole web of processes and critical decisions within the AI development lifecycle that ultimately determine the AI’s performance and trustworthiness. And the industry is betting on a group of largely inexperienced professionals to deliver on this.
The hype risk
Hype can be a threat to problem solving. And in this case, hyping up the wrong things is a threat to solving AI complexity.
The AI industry is marketing well ahead of its current capabilities, blurring business focus and value. Some are already touting timelines to achieve AGI, all while incidents of chatbot hallucinations and faulty outputs increase. We’re glorifying 200 to 500 billion parameter large language models (LLMs) which are undoubtably impressive, yet we overlook that small language models (SLMs) are capable of a dynamite effect at a fraction of the cost.
As a result, many enterprises are overlooking surgical use cases and precision applications of gen AI that don’t require flashy or exorbitant investments. I’d argue that those use cases are severely underhyped.
More importantly, we’re overlooking that enterprise adoption of gen AI is reliant on developers with limited AI expertise. If enterprises underestimate this – one of the biggest challenges in the industry that they must overcome – then only a privileged, few of them will benefit from AI at scale.
The need to uncomplicate the AI stack
Performant AI systems – whether chatbots, assistants, agents or other apps – depend on developers’ ability to consume foundation models, embed them into the massive matrix of technologies that make up today’s IT environments and then architect various systems around it all. This is easier said than done for a few reasons.
There remains a significant knowledge gap preventing application developers from becoming proficient AI developers. That inadvertent learning curve has a ripple effect on speed of innovation, increases latency and error rates, and introduces the risk of more security blind spots. On top of that, these professionals are trailing constant innovation cycles. New techniques are being introduced weekly, the stack is becoming more complicated, and limitations in popular techniques such as RAG (Retrieval Augmentation Generation) and fine-tuning create building constraints. Ultimately, this impedes production and has an upstream effect on costs and business growth.
If AI is to be built by non-technical or relatively technical professionals, not only do we need to simplify its development lifecycle, but we also need to operationalize it for scale. The open source community has been endeavoring to relieve this pressure for developers, simplifying the process of taking an AI use case from idea to production, but this must go even further. The demands of large enterprises warrant we go beyond solving for the sake of solving, we must solve for scale.
This is possible by arming developers with out-of-the-box toolkits that developers of all skills levels can use to create AI apps for production and enterprise scale, without needing to dive into the complexities of the AI models themselves. In fact, this is why IBM has been investing in enterprise scale initiatives and frameworks to simplify AI, and IT automation as a whole.
In a market defined by constant and swift innovation, the more we commoditize the AI stack, and simplify today’s complex AI infrastructure, the more value enterprises can crank out of these innovations. If we downplay this need, then I fear hype will overtake AI substance, and few will benefit from these innovations as growth multipliers for their business.
We've featured the best AI website builder.
This article was produced as part of TechRadarPro's Expert Insights channel where we feature the best and brightest minds in the technology industry today. The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/news/submit-your-story-to-techradar-pro
