The early excitement around generative artificial intelligence (AI) has recently been tempered with a heavy dose of risk-related concern and growing disappointment. Among the problems cited are a lack of killer apps, weak return on investment (ROI) and the general struggle to meet high expectations.
This cooling presents a renewed opportunity for more mature AI technologies such as machine learning, as companies that have invested heavily in the infrastructure to exploit generative AI experience the friction that comes from both seeking and worrying about a new technology.
Machine learning has been around for decades and is considered a branch of AI. Having spent the last two years firmly in the shadow of its flashier cousin, machine learning may seem a bit old and dull. But that would be selling it short. It could be argued that machine learning remains the more impactful, more understood and safer form of AI.
In effect, generative AI has entered a “trough of disillusionment,” a pit stop the consulting firm Gartner describes in its graphical representation of the maturity and adoption of technologies and applications.
According to Gartner’s technology “Hype Cycle,” in this phase, interest wanes as early implementations fail to deliver. There is a separating of the commercially viable wheat from the economically unfeasible chaff. Providers either shake out or fail. Continued investment in the field gathers risk and is contingent on the satisfaction and continuing support of early adopters.
How is it possible to reconcile the twin realities of the serious-minded questioning surrounding generative AI, while investors continue to plow billions into the talent and supporting infrastructure required to build it out?
There are at least two answers. First, it’s important to appreciate that the trough of disillusionment is just a way station on a journey. According to the Gartner Hype Cycle, subsequent phases include the “slope of enlightenment,” when understanding of how to adapt the innovation is growing, and the “plateau of productivity,” in which more users see real-world benefits and the innovation goes mainstream. Not all innovators make it to the plateau, but some do, and the rewards are plentiful.
A second but complementary perspective sees generative AI development as a moonshot, a perceived high-stakes race for a technological holy grail of the ages — artificial general intelligence or AGI. Rightly or not, big players in the AI developer community have lined up behind the notion that generative AI represents the critical path toward AGI. Despite the relative lack of widespread, accessible use cases that are materially impactful today, these people have faith. Or most of them do.
So, on the one hand, generative AI has tumbled into an expectations trough, with enterprise questioning its ROI. On the other hand, the technology retains a magnetic appeal for many investors, though this remains grounded in an uncertain possibility. For decision makers driven by the bottom line, continued spending may soon become difficult to justify.
It’s unlikely there will be an ecosystem collapse, as big players and investors remain committed. But there is a good chance enterprise customers will retrench. Confronted with the weak ROI of many use cases, it’s likely that companies will take stock of all the things they have been leaning in on, including data, talent, infrastructure and ethics, and reconsider how technology can support their business priorities. What might they see?
One of the frustrating experiences of watching the obsession with generative AI has been the myopia of it. Now seems like a suitable moment to remember that there are other technologies, even AI-related technologies, that can have a major economic impact.
In contrast to generative AI, which might be seen as a big tool that customers generally limit to low-stakes applications, machine learning might be seen as a smaller, more precise tool that customers can and do apply to core business challenges. There is a clearer line of sight to ROI, and techniques for understanding how the technology works, which should facilitate its evaluation on both ethical and performance grounds.
Machine learning has been around for decades. In the 1960s, early and experimental applications analyzed patterns in sonar signals, electrocardiograms and speech patterns. Today, machine learning is applied in a wide cross-section of industries, with top uses including fraud detection, medical diagnostics, product recommendations and market forecasting.
Finally, it’s hard not to be awestruck by the assessment of Anthropic CEO Dario Amodei that training large language models will cost between $10 and $100 billion in the next three years. Only a precious few companies will be able to afford the price tag of development. The rest of the ecosystem will be comprised of deployers and users.
We can expect the pent-up demand for AI that delivers tangible results to stimulate a resurgent interest in traditional forms of the technology such as machine learning. These can deliver material benefit at the core of business priorities today.
