The causes of disillusionment with GenAI

Nearly two years after the explosion of generative AI with the introduction of ChatGPT, the market has shifted.

The generative AI market is heading toward disillusionment, according to Gartner analyst Arun Chandrasekaran.

The letdown comes after years in which both vendors and enterprise customers expected generative AI models to perform unrealistic functions.

In this Q&A, Chandrasekaran discusses the causes of disillusionment, and how vendors can address it.

What is causing the disillusionment we're seeing with generative AI?

Arun Chandrasekaran: The first cause of disillusionment is business value, which is not close to what was anticipated by CIOs and CTOs. Enterprises had sky-high expectations. People were expecting 30% to 40% improvement in productivity out of the gate, and they're starting to recognize that that's not happening. There are things like productivity leakage.

The second cause is that these models are non-deterministic. They continue to hallucinate. In 2023 and early 2024, the vendors accelerated the delivery of these products to the point where a lot of the products promised Heaven and Earth. The marketing was sky high, but the products were not doing the things that the vendors expected the products to do.

I would also argue that some of this is a function of the vendors jostling in the marketplace and trying to position their products a little too aggressively than what the product was capable of.

Another cause of disillusionment is governance. Ensuring data privacy, explainability and model safety requires effort. I don't think enterprises anticipated the work required to govern this. Another is data. These models are advanced, but with their general-purpose models, enterprises must combine their data -- often unstructured data -- with these models. That requires an enormous amount of effort.

Does disillusionment with AI models translate to slower adoption of generative AI models and systems?

Chandrasekaran: I don't think we have seen customers moving away from it, but it's slowing adoption without a doubt. People are a lot more conscious of the value and the ROI story. They're also starting to do longer pilots with some of these products and projects. A lot of decisions that were made early were emotional decisions. People are becoming a little bit more rigorous in terms of how they evaluate it. It's slowing things, but it's not forcing customers away from AI. I don't think we've seen that happen yet.

Is there anything that vendors themselves can do to address AI disillusionment?

[Some disillusionment] is a function of the vendors jostling in the marketplace and trying to position their products a little too aggressively than what the product was capable of.

Chandrasekaran: They could market the products for what they can do. Vendors need to create better ROI stories for customers in how they sell the product and articulate the business value. They need to embed security and privacy into the product so they can ease some of these implementation rules for customers. Finally, I would argue that the data part is an area particularly for service providers as well as automation vendors.

There's a lot of market opportunity for vendors in terms of trying to solve this problem. For example, unstructured data -- how do you do more effective ETL [extract, transform, load] and automated labeling and annotation of unstructured data? That's an area where I think there's an enormous opportunity for vendors. Data classification is another area where there's an opportunity for vendors. All these challenges, in some sense, also translate into market opportunities for vendors where they can help customers navigate and overcome these challenges.

Editor's note: This Q&A was edited for conciseness and clarity.

Esther Ajao is a TechTarget Editorial news writer and podcast host covering artificial intelligence software and systems.