TDWI | Training & Research | Business Intelligence, Analytics, Big Data, Data Warehousing

What’s Ahead in Generative AI in 2025? (Part Two)

Further predictions for the coming year.

• By James G. Kobielus
• December 19, 2024

The first part of this article addressed new generative AI platforms and new sources of diverse data expected in 2025. Read on for the rest of our predictions.

For more on generative AI, don't miss TDWI's free Virtual Summit in January! Even More: 4 Ways to Control Cloud Costs in the Age of Generative AI How RAG Will Usher In the Next Generation of LLMs and Generative AI

Prediction #5: More Robust and Comprehensive Generative AI Governance

Enterprises will focus on building robust governance into their generative AI pipelines in 2025 and beyond. In the recent TDWI data and analytics survey, 60% of enterprises said their organizations’ current or planned use of generative AI has increased the urgency to ensure stronger AI governance.

In the coming year, enterprises will accelerate their investments in retrieval-augmented generation and other tools for reducing hallucinations and otherwise ensuring the trustworthiness of generative AI model outputs.

Over the next several years, I predict that enterprise adoption of generative AI may slow down unless organizations can institute robust governance guardrails to assure that the technology is secure, trustworthy, transparent, ethical, and otherwise suitable for widespread scalable deployment.

Nevertheless, I predict that enterprises will have access to increasingly sophisticated penetration testing tools in 2025 that will generate malicious prompts to test generative AI’s vulnerability to threats such as “jailbreaks.”

And it’s worth noting that generative AI technologies will be an integral component of many of these guardrails that will be deployed next year and beyond. Generative AI will increasingly be used to create fake data for training machine learning models to detect cyberattacks, to generate and orchestrate realistic cyberattack scenarios, and to generate ersatz malware for the purpose of training systems and teams to recognize and respond to threats.

Prediction #6: Deepening Returns on Investment

Enterprises will see increasing returns on their generative AI investments in 2025 and beyond. These initiatives will increasingly emerge from proofs of concept, pilots, and beta tests and begin to pay back enterprise investments. TDWI research shows that many enterprises are well into their journeys to operationalize and productize generative AI applications.

The recent TDWI survey on enterprise AI readiness showed that 10% of organizations are already in production with generative AI, while 33% are building proofs of concept. At the same time, many enterprises are continuing their work-from-home initiatives, with 25% allowing employees to experiment with generative AI in these settings.

ROI from generative AI will remain elusive until the technology’s processing, storage, networking, and other resource requirements decline to cost-effective levels. High-performance generative AI uses larger models and data sets than almost any other type of application, increasing the likelihood that these applications will be budget-busters for many enterprise IT shops. Indeed, TDWI’s data and analytics survey found that 27% of enterprises say that cost has been higher than expected for managing data in the cloud, a concern that does not bode well for growing generative AI workloads.

I predict that the cost of generative AI applications will continue to drop as organizations adopt increasingly sophisticated “small language models” as an alternative to resource-hungry LLMs.

Nevertheless, sustainability issues surrounding generative AI may dampen or delay substantial ROI from this technology in 2025 and beyond. Generative AI will consume ever larger portions of the planet’s energy, water, real estate, and other natural resources. The technology’s ravenous power appetite will drive demand for renewable energy and even a return to established, issue-plagued power sources such as nuclear.

Takeaways

With these trends in mind, TDWI recommends that data and analytics professionals take the following next steps in their generative AI journeys in 2025 and beyond:

• Focus on copilots, digital assistants, and other use cases where generative AI augments staff productivity
• Evolve existing analytics and data foundations to support generative AI apps
• Leverage private, partner, and marketplace data to train generative AI apps for innovative apps
• Implement retrieval-augmented generation and other approaches for generative AI guardrails
• Control the costs of implementing generative AI through a focus on renewable energy and computational efficiency