Insight of the Week: Operationalizing AI 2.0

By: Kerry Robinson

It's time to revisit the idea of Operationalizing AI.

Generative AI, in the form of Large Language Models like the now famous (infamous?) ChatGPT, and the recently announced GPT4 have shaken up the world of Conversational AI.

As one of my colleagues put it: "LLMs are the most breathtaking AI innovation I’ve seen in my entire career."

The game has changed. Users, vendors and analysts are racing to adapt.

But if you've been reading this weekly email for a while, I think you've got a head start.

For a long time I've been encouraging you - and whomever else will listen - to think about Conversational AI as more than just technology. You've probably heard me say:

"With Conversational AI, you're not just deploying technology, you're building a Virtual Agent Workforce, that needs leading and managing, just like humans do."

(And if you haven't, check out this?video)?in my series on?how to think differently about Conversational AI.

This was one of several?'aha' moments?I've had in my career in Conversational AI.

Along with some other key principles, this led me to the idea that organizations should focus on the idea of?Operationalizing AI, which I shared in this email a little over a year ago.

In that email I warned that?AI isn't very intelligent. We often overestimate its abilities.

And that it's?amazing in certain narrow situations. But fails in many others.

That?AI isn't naturally smart, helpful and resourceful, like humans are.

But that?it's constantly evolving. Improvements allow us to do more every year.

Looking back on what I said 14 months ago, I think it's fair to say that the breakthroughs in Large Language Models have broadened the situations in which Conversational AI is useful.

I think it's fair to say that ChatGPT and the services that leverage it are verging on smart, helpful and resourceful, like humans are.

But it still fails in unexpected ways. And we still overestimate it's abilities.

That's why we?still?need to work on operationalizing AI. Back then, I described it as:?building the systems, culture and the organizational muscle memory that ensures you and your customers are getting the best value from this flawed technology at every moment in this exciting journey.

With Large Language Models, I think we need to lean into the idea of a virtual agent workforce even more. Because, just like humans, LLMs are very capable, but inconsistent. To operationalize AI in the era of LLMs, we need to:

• ?Coach and train them to say and do the right things. Technically speaking that's 'prompt engineering' - figuring out what to feed into the Large Language Model to get the right thing out. But remember, customers might provide inputs that result in unexpected outputs. Just like humans do occasionally!
• ?Understand where the conversations are too high stakes for Large Language Models. Where only scripted responses are appropriate. Or where we need to hand off to a human. Just like junior agents escalate to managers in difficult situations.
• ?Measure and monitor for performance and compliance. Just like we do with human agents. Think profanity filters, side-by-side call (or chat) listening. Speech and chat analytics. And of course outcome metrics like issue resolution.
• ?Leverage the partnership of Artificial Intelligence and human intelligence. Think advanced knowledge base search. Suggested responses. Or even leveraging LLMs to help with the design and training of more scripted interactions, like I covered in last week's email:?Can you design your IVR with ChatGPT?

It's exciting times in AI. Just remember that with conversational AI, you're not just deploying technology, you're building a virtual agent workforce, that needs leading and managing, just like humans do. There, I said it again ??

Kerry Robinson?is an Oxford physicist with a Master's in Artificial Intelligence. Kerry is a technologist, scientist, and lover of data with over 20 years of experience in conversational AI. He combines business, customer experience, and technical expertise to deliver IVR, voice, and chatbot strategy and keep Waterfield Tech buzzing.

Subscribe to Kerry's Weekly AI Insights