IBM has automated the training model process to allow enterprise clients, including financial institutions, to use generative AI in a safe and compliant way that can be implemented and scaled quickly.

“Conventional machine learnings, as good as they are, take a huge amount of time. They take a huge amount of discipline and science,” John Duigenan, general manager of financial services industry at IBM, told Bank Automation News.

Not only is AI time-intensive, but there is a “gap between the appetite to use AI and the ability to deliver it safely,” Duigenan said. That’s why IBM looked to innovate using generative AI and, ultimately, unveiled watsonx in May, according to an IBM release.

Generative AI can be trained much faster than traditional machine learning because the training process is more automated and less about going through individual documents with specific tools and annotation, Duigenan said.

In the past, an AI-driven virtual assistant would take up to a year to train, Duigenan said, noting that with generative AI “being able to deliver a new capability, and new profound answers to a virtual assistant in less than a month, that’s huge.”

AI, data, governance

Watsonx is a combination of coding, information design, information sourcing and data science, Duigenan told BAN. Watsonx has three components: AI, data and governance.

“There’s a ton of technology within watsonx that really provides that ability to create AI models, test them and deploy them; use open-source data warehousing format and put those anywhere a client wants to, and then trust the AI that’s created,” he said.

Watsonx components include:

Watsonx.AI: A studio where machine learning models and deep learning models are created and tested. Large quantities of data from a company can be assembled into a foundation model that can be tested with questions, including hints and prompts when needed. Once the model is working, it will be deployed and can then be called using regular API.

Watsonx.Data: The tool has an open standard data repository and query mechanism that IBM can use to connect to existing data sources. For example, companies can create their own open-source data, without third-party involvement, reducing risk. That data can then be connected to IBM.

“You bring our AI to your data and don’t take the risk of moving it around,” Duigenan said.

Watsonx.Governance: Every firm IBM works with is regulated in some way, and adoption of AI must be explainable. Generative AI, in some cases, can create hallucinations giving wrong answers, but that isn’t the case with watsonx, Duigenan said.

“Any answer that we generate with watsonx can show the exact lineage to why we created that answer,” he said. For example, the technology can explain how AI has serviced a client, made a risk decision or hired a new employee.

Financial institutions can try watsonx here: https://dataplatform.cloud.ibm.com/registration/stepone?context=wx 

Identifying uses

Customer care is a use case that IBM clients have identified related to generative AI, he said. Sorting customer documents takes time.

“By adding generative AI, customer care expands the potential of a virtual agent massively, and also reduces the dependence that firms have had on training intense natural language,” he said.

London-based NatWest Bank is embedding watsonx into its chatbot experience to accelerate generative AI workflows and improve the customer experience, IBM Chief Executive Arvind Krishna said during the tech giant’s second-quarter earnings call in July.

Finding the tech

Financial institutions can experiment with watsonx on IBM’s website. In fact, in some cases companies work with IBM to create a custom approach to generative AI based on specific needs through client engineering, Duigenan said.

Client engineering combines IBM engineers with client engineers to create a customized tool based on what the client needs, he said. IBM clients can be up and running with watsonx in a matter of weeks or even quicker. IBM just had one client up and running in less than 10 days, Duigenan added.