eGain's CEO Pushes for AI Guardrails at Solve Event
During eGain's annual Solve conference in Chicago this week, Phil Britt had a chance to sit down with Ashu Roy, the company's chairman and CEO since 1997, to discuss the recent concerns about artificial intelligence and where eGain figures into conversations about the technology.
CRM: You mentioned during your keynote that knowledge is the key to AI. But to locate the knowledge to answer certain questions, with Hugging Face and some other instances, AI has broken out of the sandbox and attacked other systems. How can this be prevented?
Roy: You need to build better sandboxes. There are ways you can set up an environment where AI systems are allowed to do a range of things, or you can set up an AI system where you essentially whitelist what it can do, as opposed to blacklisting what it cannot do. You need to smartly whitelist the range of tasks that AI systems can perform in any given environment. That's what why we say the role of knowledge is to control the AI systems and feed the right information into the AI systems. Then evaluate whether they are doing the right thing.
CRM When you're bringing new people into contact centers or other environments, you might want them to use AI for certain tasks, but these new hires are rarely trained on the security issues. How do you manage that?
Roy: A newcomer cannot control the input properly and cannot judge the output quality. So, you need packaged AI solutions that have all the controls and guardrails in place, so that the consumer, which is the human associate, is simply getting the answers from a controlled environment that is set up for them. They cannot go in and create prompts. They cannot change rules.
CRM: What is eGain doing to ensure the AI with which it works doesn't break out of the sandbox?
Roy: Our premise is that knowledge is the core instruction and control for AI. We have built into our platform a low-effort ability to create, manage, curate, and verify the knowledge and do it in a continuous loop inside the business. We're able to ensure that the knowledge that the AI systems are taking as input are always correct, compliant, consistent, and consumable.
On the other side of the spectrum, when the AI systems deliver output, we have our continuous evaluation, which is working on the system. So to ensure that if there is a drift or any sort of problem, we are able to detect it and control it.
CRM: You and some others have discussed regulation to help protect against rogue AI. What do you see happening on this front in the United States? In Europe?
Roy: European markets are a couple of steps ahead in regulation. And mostly, the direction is is a good one. The U.S. will also catch up to that. We tend to regulate a little bit less. As the nature of risk changes, people could potentially do bad things much quicker. Regulations have to control the risk.
CRM: What are the best practices for AI to be deployed safely and profitably?
Roy: If you deploy AI safely and it is doing something useful, then chances are high that it's going to be a profitable investment. In other words, the technology is so powerful and capable that it can be deployed and should be deployed safely to do useful things. The simple thing we think of is ensuring that.
Best practices for that include the following:
- Ensure that you are building on a foundation of trusted input into AI. Input is typically the know-how and the data.
- Make sure that your know-how, which is knowledge, is trusted and maintained in a way that maintains the reliability of the knowledge.
- Make sure that the guardrails you're putting in place for your system, not just for AI but for the whole system, are being managed and maintained. It's not just a question of setting it up. It's also a question of continuing to manage it and keeping up with it, because the underlying AI technologies are getting better and better.
- The evaluation of output of these AI systems should be continuous, not one and done. The top-level flow is what a safe AI system in a business environment should have.
- Apply AI to customer experience and customer operations. It will improve the ability of your customer-facing staff to deliver better service. Simple queries and simple interactions with customers can be managed with a much more natural interface. Consumers are actually happy dealing with conversational interfaces much more so than what people thought earlier, so there is a win-win opportunity in CX where you can create profitable AI investment.
CRM: Any other advice for users of AI?
Roy: The rollout and deployment of more customer-facing AI is going to go up a lot. I believe that many of those will be very successful. Some of them will not be successful because they will stumble on those controls and sandboxes.
It's a question of who's going to architect their systems well and not. The risk of AI, I feel, in the next couple of years, is that people start giving up their controls of what is good output, and we forget how to judge the output. They will say: "Oh, it's kind of good. I don't need to check it anymore. That's a huge risk.