VB AI Impact Series: Can you really control a multi -agent AI?

Individual coilots is yesterday’s news. Competitive differentiation revolves around the launch of a network of specialized agents who cooperate, with self -intelligence, and calls the correct model for each step. The latest batch of the artificial intelligence effect series on Venturebeat, presented by SAP in San Francisco, dealt with the issue of publishing and rule of multi -agent AI.

Yaad Oren, Managing Director Sap Labs US and World Research and Innovation Head of SAP, Raj Gamba, SVP and CIO with Agilent, a company for analytical and clinical laboratory technology, discuss how to spread these systems in the real world environments while staying within the cost, specifications and Guardrails for commitment. Orine said the Sap goal is to make sure that customers can expand the scope of artificial intelligence customers, but safely.

He said: “You can be almost completely independent if you want, but we make sure that there are a lot of checkpoints and monitoring to help improve and repair.” “This technology must be widespread. It is not perfect yet. This is the tip of the iceberg about what we do to make sure that agents can expand, as well as reduce any weaknesses.”

AI’s active pilots spread throughout the organization

Currently, Agilent is integrated with artificial intelligence throughout the organization. The results are promising, but they are still in the process of addressing this weakness and expanding the issues of scaling.

“We are at a stage where we see results,” he explained. “We now have to deal with problems such as, how do we enhance the monitoring of artificial intelligence? How can we improve the cost of artificial intelligence? We are definitely in the second stage of it, where we are no longer exploring. We are looking into new challenges and how we deal with these costs and monitoring tools.”

Inside Agilent, artificial intelligence is spread in three strategic columns, and Jamba said. First, on the side of the product, they explore how to accelerate innovation by including artificial intelligence in the tools they develop. Second, on the side facing the customer, they determine the possibilities of artificial intelligence that will provide the largest value to its customers. Third, they apply artificial intelligence to internal processes, and to build solutions such as self -healing networks to enhance efficiency and ability.

“With the implementation of these cases of use, the only thing that we focused on a lot is the framework of governance,” Jamba explained. This includes setting policy -based boundaries, ensuring handrails is removed for each solution, and removing unnecessary restrictions while maintaining compliance and security.

The importance of this was stressed recently when one of their agents updated the training, but they had no examination in a place to ensure that its borders were solid. Jamba said that the upgrade immediately caused problems, but the network was quick to discover it, because the second part of the column is to check, or ensure registration of each entry and every exit and can be followed.

Adding a human class is the last piece.

He said: “Small and small use cases are somewhat clear, but when you talk about natural language and large translations, these are scenarios where we have complex models.” “For these larger decisions, we add the element in which the customer says, I need a person to intervene and agree to my next step.”

He added that the issue of speed versus accuracy is early during the decision -making process, because the costs can be added quickly. Complex models for low -spread tasks pay these costs much higher. The governance class helps to monitor speed, cumin and accuracy of the agent’s results, so that they can determine construction opportunities on current publishing operations and continue to expand the artificial intelligence strategy.

Solve the challenges of the agent integration

Integration between artificial intelligence agents and current institutions solutions remain a large pain point. While local old systems can be connected through the interfaces of programming applications or architectural engineering that depend on events, best practices are to ensure the work of all solutions first in a cloud frame.

“As long as you have a cloud solution, it is easier to get all connections and all delivery cycles,” said Orn. “Many institutions have local facilities. We help, using artificial intelligence and agents, to deport them to a cloud solution.”

With the SAP integrated tools series, complications can be maintained such as the older programs easily in the cloud as well. Once everything is inside the cloud infrastructure, the data layer comes, which is equal if not more important.

In SAP, the Data Business cloud works as a unified data platform that collects information from both SAP and SAP. It is very similar to Google Web content, Business Data Cloud can index business data and add a semantic context.

Orine was added: “The agents then enjoy the ability to communicate and create commercial operations from one side.”

Treat gaps in the activation of the institution’s agent

Although many elements work in the equation, three of the importance: the data layer, the formatting layer, the privacy and safety layer. Clean organized data depends well, of course, that decisive and successful functional publishing depends on a unified data layer. The commitment of the agent’s synchronization runs, allowing strong automation to the agent across the system.

“The way it is organized (agents) is a science, but it is also an art.” “Otherwise, you can not only have failure, but also in checking and other challenges.”

Finally, investment in security and privacy-especially when a group of agents works through databases and architecture of institutions, where the mandate and identity management is very important. For example, a member of the Human Resources Team may need to access identification information or personal definition, but no one else should be able to see it.

Orine said that we are heading towards a future to which members of the Human Foundation team and members of the automated team join, when the identity management becomes more vibrant.

He added: “We started looking at the agents more and more that they are human beings, but they need additional monitoring.” “This includes on board the plane and the license. It also needs to manage the change. The agents have started facing a professional personality that you need to maintain, just like the employee, only with more monitoring and improvement. He is not independent in terms of life cycle management. You have checkpoints to know what you need to change and improve.”