Next Generation AI Advanced Search technology can make data ‘way more valuable’

AI is rapidly transforming industries, but much of its potential is still largely untapped. One major opportunity yet to be tapped is Next Generation AI Advanced Search technology – we sat down with Isabelle Pons and Paul Verhaar from Sopra Steria to find out more on the matter.

Advanced Search technology addresses the challenge of information overload by transforming vast amounts of data into valuable insights. This is the ideal job for AI technology: turning mountains of data into actionable insights.

Despite the immense promise of AI, many organizations struggle to translate proof-of-concepts into scalable solutions. Some of the lingering issues include things like poor data quality, lack of technical readiness, and unclear business value. There is also growing concern over the environmental impact of AI when it comes to power consumption.

“AI must be responsible. We must balance business needs with environmental impact,” says Isabelle Pons, AI Program COO at Sopra Steria, a pan-European technology consultancy firm. “For example, that can include optimizing model sizes, accepting a slight reduction in performance or scope.”

Data strategy in place

Enormous amounts of data can be overwhelming and without an appropriate data strategy in place, it can be difficult to extract meaningful insights. “That’s where Next Generation AI Advanced Search comes in, turning this overwhelming data into actionable intelligence by using retrieval techniques such as Retrieval-Augmented Generation (RAG),” notes Paul Verhaar, AI Practice lead at Sopra Steria in the Netherlands.

“However, moving from experimentation to successful production deployment remains a significant challenge for many RAG projects,” he adds.

RAG is a technique that enhances the accuracy and reliability of generative AI models by providing them with access to relevant information from external sources, such as a knowledge base, a document repository, or any data silo available. Using the knowledge contained within these external sources, the AI model can then generate more accurate and informative responses.

While RAG is very powerful and useful, Verhaar also noted some of the potential drawbacks. “It’s easy to experiment with RAG, often leading to: oversimplified architectures that do not scale well, lack of integration into existing systems, and potentially bad results from bad inputs.

In addition to that, many organizations will opt for closed-source language models – as it is the quick and effective option. Closed-source AI models use underlying code and training data that are not publicly accessible, which can put limits to their customization and transparency.

The correct implementation of RAG, according to Verhaar, will need to pay close attention to scalability, both in terms of horizontal scaling (adding servers) and vertical scaling (boosting individual server power).

These components can then be scaled independently based on load and the particular use case. A successful implementation of any RAG system needs to be easily integrable with current workflows, meaning that it can run both on-premise and from the cloud. This will allow a more seamless integration into an organization’s existing systems.

Another key point is to ensure that a RAG system is easily configurable by enabling component interchange without refactoring. “This flexibility also allows easy extension to new data sources and modalities as they become available,” says Verhaar.

Education and human oversight are key

Sopra Steria’s inhouse developed Next Generation AI Advanced Search technology was used among others by the Dutch Financial Intelligence Unit in its mission to become a knowledge hub in the fight against money laundering and financing of terrorism.

Sopra Steria’s has also collaborated in other similar AI projects throughout various countries in Europe, like helping to develop AI strategies and building AI assistants. One project the firm delivered focused on setting up a cloud generative AI platform for the French government.

The consultants of Sopra Steria warn against so-called ‘dirty data’. “For any AI project to succeed, it is necessary to clean and structure data with the support of both AI and human oversight”, Isabelle stresses. “Therefore it is crucial to educate employees on the use of AI and to put clear guidelines and governance in place. Only then we can leverage the potential of AI.”