From assistant to fully automated — Cisco’s AI strategy

By Catherine Sbeglia Nin June 13, 2024

Collected at: https://www.rcrwireless.com/20240613/ai-ml/from-assistant-to-fully-automated-ciscos-ai-strategy

Cisco is using AI across its entire product portfolio to simplify the customer experience

Earlier this week, Cisco Investments, Cisco’s global corporate venture investment arm, announced a $1 billion artificial intelligence (AI) investment fund to support the startup ecosystem and expand the development of more reliable solutions. To date, Cisco has already committed nearly $200 million of the fund. Internally, the company is using AI — and generative AI (gen AI) — across its entire product portfolio to simplify the customer experience and deliver deeper, more intelligent insights.

“When we talk AI, it’s not just text-based, large language models; it’s audio, video and text,” the Senior Vice President and General Manager of Cisco Collaboration Javed Khan told RCR Wireless News. “Our approach to AI is combining the large language progress the world has made the last few years with audio and video intelligence, which is really a strength of ours [and] is really applicable to collaboration use cases.”

AI as an assistant 

Khan said the goal is to have Cisco AI Assistant “everywhere — in security, networking, observability and collaboration.” The support tool, which combines gen AI technologies with the company’s data, runs across Cisco’s entire architecture, manifesting itself differently depending on the use case. “But it is smart enough to ask another platform what it might know about that particular use case or problem,” he explained. “What that means is the networking team, the security team, the collaboration team, they will all contribute intelligence to a common data lake.”

Currently, Cisco has integrated AI across it’s Webex Suite and Contact Center offerings. And while the future plan might be to fully automate some of the decision-making at Cisco, the company today is using AI to support its human agents address customer questions and concerns. “There is good reason for the concerns that most companies have [about AI] and we’re seeing that stall some adoption, frankly,” Khan acknowledged. For its part, he said Cisco doesn’t use personal data sets to train larger models as doing do might “pollute a greater population of learning.”

And of course, hallucination has become a famous problem for gen AI. “Unlike most computer code, sometimes the answers from gen AI is somewhat subjective and there is a chance that it may not be accurate, so the approach we’ve taken is when it comes to generative AI, we are starting with an assist approach where you need more definitive answers,” he said. An example might include a contact center where if a less straightforward question or comment comes in, AI composing the response, but an agent ultimately clicks the send button. “Over time, you measure the accuracy of those models and you get to a threshold where you’re ok going 100% automated,” added Khan.