Gartner on artificial intelligence Generative AI is on the decline

Related Vendors

Gartner Deutschland GmbH

FAULHABER Drive Systems

In their latest report on the hype cycle of artificial intelligence (AI), the analysts of the Gartner Group see many more AI disciplines. While the hype surrounding Generative AI (GenAI) is weakening due to disappointed expectations, other transformative disciplines and forms of AI and machine learning are becoming a higher priority.

According to Gartner analysts Afraz Jaffri and Karitha Khandabattu, generative AI has passed the peak of exaggerated expectations, i.e. the hype, and is on its way to the "valley of disappointment". However, the hype is still continuing, although more added value is being gained from other AI technologies, sometimes in combination with GenAI. These projects are based on standardized processes that help with implementation.

To gain the greatest advantage from this trend, companies that want to be leaders in AI should base future system architectures on AI composite technologies. In doing so, they could use approaches from innovations that can be found throughout the AI hype cycle.

Gallery

What the Gartner analysts mean are effects and factors that gain in importance with the scope of modern AI projects. These include governance, risk management, data ownership rights (highly controversial), data security and protection and overcoming IT "legacy" issues. These factors would be tackled at every level, from the individual to teams and companies to the nation. However, the level of maturity still leaves a lot to be desired, despite progressive regulations.

Generative AI

As Gartner analyst Svetlana Sicular writes in the June 2024 report, end users are "aggressively" experimenting with GenAI, while early adopters in most industries have already achieved initial success. The leading technology providers would prioritize GenAI-supported apps and tools. Business attention is now shifting from the excitement of foundation models to use cases that drive ROI.

"Most GenAI deployments are currently internal and low-risk," observes Sicular. "But with the rapid advancement in productivity tools and AI governance practices, companies will soon be using GenAI for more business-critical use cases in specific industries and scientific research." Some providers call this "Industrial AI", for example.

Synergy of many factors

In the long term (in two to five years at the earliest), GenAI-supported user interfaces would facilitate the commercialization of the technology and thereby democratize AI, etc., according to Sicular. Numerous solutions have emerged to support innovations in basic models, hardware and data for GenAI.

According to Gartner's 2023 Enterprise AI Adoption Survey, 18 percent of market leaders using AI report that their organizations are already making good progress in adopting GenAI. "These organizations are learning how to make the most of their own data for GenAI through prompt engineering and fine-tuning," says Sicular. AI-enabled data and associated metadata have become central components of GenAI strategies. Synthetic data helps companies to expand scarce data stocks in order to compensate for bias, achieve fine-grained high resolution and ensure data protection at the same time.

"The introduction of generative AI is turning software development on its head," writes Sicular. "The automation techniques in development promise to automate five to ten percent of a programmer's work." Companies and other organizations are willing to finally tackle the modernization of their legacy code with the help of GenAI. For example, AWS developers were able to modernize outdated, insecure Java code within a very short time using the AI tool Amazon Q.

Two megatrends

In this year's Hype Cycle, Gartner highlights two megatrends, namely AI Engineering and Knowledge Graphs. Both signal the need to handle AI models in a scalable and robust manner. AI Engineering uses new team topologies to deliver business solutions. Knowledge Graphs, on the other hand, provide reliable "logic" and explainable reason, which contrasts with the error-prone, albeit powerful, predictive capabilities of deep learning in Generative AI.

Early starter

In the early stages of the so-called "innovation trigger" of the hype cycle are technologies such as composite AI, AI-enabled data, causal AI, decision intelligence, AI simulation and multi-agent systems. They reflect the growing need to expand the automation of processes and decisions about the results that individual models produce to include orchestrated, iterative multi-model services.

The hype summit

At the peak of the euphoria (as of June 2024) are technologies and processes such as responsible AI, AI TRiSM, prompt engineering and sovereign AI. AI TRiSM stands for "AI Trust, Risk and Security Management", which, like the other disciplines, points to growing concerns about the governance and security aspects of the rapidly expanding use of AI in companies and by individuals.

The descending branch

It may come as a surprise that the descent into the "valley of disappointments" includes such recently introduced processes and technologies as ModelOps, synthetic data, "smart" robots, edge AI and, finally, neuromorphic computing. "These innovations still have momentum, but the level of adoption varies widely, and they are either misused or subject to exaggerated expectations about their economic value," write Gartner analysts Afraz Jaffri and Karitha Khandabattu. Neuromorphic computing and "smart" robots have made good progress in the past year, which indicates the potential to quickly pass through the rest of the hype cycle until they reach the "plateau of productivity", as "computer vision" is already doing today.

Cloud AI services

Cloud-based AI services have fallen down the hype cycle since 2023 because so many new GenAI-based AI services have become available in the cloud. Both providers and end users have faced issues related to the capacity of the service, its reliability, the frequency of model updates and cost fluctuations. Gartner analysts Afraz Jaffri and Karitha Khandabattu attribute these phenomena to normal "growth issues". In several places, they address the energy requirements and the concerns associated with this on the part of AI users or service providers.

Autonomous vehicles, intelligent applications

These two technologies are now on the "Slope of Enlightenment", i.e. the last stage before the "plateau of productivity", which corresponds to the mainstream. Despite major concerns, the withdrawal of operating licenses and restrictive regulations, the use of autonomous vehicles has increased in some regions. Intelligent applications are now being supported by GenAI and are making inroads into the world of work. However, more time is needed to objectively quantify their impact on productivity.

New entrants to the hype cycle in 2024 would be quantum AI, "embodied" AI (see below) and sovereign AI, as companies and governments begin to look at both the potential and the dangers that an AI-driven future could bring. Note: Quantum communication is tap-proof. That's why the German government is promoting it as vigorously as AI.

The "matrix of priorities"

Gartner analysts believe that within two years, composite AI will already be the standard method for developing AI systems and will be widely used in the mainstream. It is therefore a high priority for users. Another transformative innovation is computer vision. It is already being used in the mass consumer market on "smart" devices such as cell phones.

Other innovations are only two to five years away from mainstream use. These include decision intelligence, "embodied AI", generative AI, intelligent applications and responsible AI. They offer significant competitive advantages and address issues related to the use of AI models in business processes, such as maintaining customer and employee trust. The technology with the lowest demand and less than one percent market penetration is quantum AI, according to the augurs. At the same time, they point out that quantum AI could be used positively in a large number of fields of application.

"Embodied AI"

Although it is still in the early stages, Gartner analyst Pieter den Hamer already counts "Embodied AI" as a transformative AI technology. Embodied AI is based on the view that intelligence and embodiment are inextricably linked in a particular context because one shapes the other. In this approach, a physical or virtual model or that of a virtual AI agent is trained and collaboratively developed together with its user interface, sensors, appearance, actuators (such as gripper arms). As with robots or game characters, the purpose is to interact with a specific, real or simulated environment.

According to Hamer, this enables the robust, reliable and adaptable execution of "intelligent" tasks: "Such AI agents can either act autonomously or augment the capabilities of humans in practical, dynamic environments." Their perception is active, their behavior adapts to the conditions, controlled by the intelligence of the respective AI agent and depending on the "physical" capabilities of the host body in a particular environment.

They could be virtual assistants, avatars, game characters, autonomous vehicles and "smart" robots. According to Hamer, this is "the path to more effective and trustworthy AI, and AI is finding its way into more and more products, services and business models", for example in dangerous areas of application such as disaster zones or hostile environments.

*This article was first published on our sister website www.Bigdata-Insider.de (German language)