Dreaming up the Ai flywheel

The idea of an Ai flywheel first took shape in Spring 2014, on a whiteboard in a small conference room in a nondescript office building in San Francisco. It leapt forward in 2018 in another conference room at Dialpad HQ, also in San Francisco. Last week, when Dialpad asked me to write this intro to our new ebook about the future of Generative Ai, I found myself thinking back on those moments from years ago with a feeling of deja vu.

From SBAITSO to NLP

I can trace my interest in Ai back to 1991, when Creative Labs released the speech synthesis program Dr. Sbaitso (SBAITSO = Sound Blaster Acting Intelligent Speech-to-Text Operator). Even with all its limitations, this early example of natural language understanding gave me a glimpse of the future—and I can trace that glimpse directly to those early whiteboard sketches as we started TalkIQ back in 2014.

I co-founded TalkIQ with the goal of using state-of-the-art machine learning to make voice conversations easily accessible online at scale. Every day businesses generate oceans of customer call data that just sits there forever, unused. Call center managers can listen in on a certain number of calls in order to help agents and find issues. But no manager can listen to, let alone interpret, all the calls in a high volume call center.

But those quickly drawn streaks from an almost spent dry erase marker on our first whiteboard session manifested an opportunity to better understand communication at scale. It was an exciting time to be an Ai-fueled startup. Advancements in transcription and ASR (Automated Speech Recognition) accuracy in particular made the moment ripe for applying existing state-of-the-art NLP (Natural Language Processing). We were able to quickly build momentum with early features like call descriptions, extracting key moments, generating insights, and much more, all fueled by the great traction we were getting with partnerships.

Reality catches up to our daydream

Then we demoed for Dialpad in 2018 and everything changed. Dialpad’s communications platform made it easier to speak with customers from anywhere. We were using Ai to make it easier to understand what they were saying. It was perfect symbiosis. In two weeks TalkIQ was integrated in Dialpad’s production environment; a few weeks after that, we became one team and quickly grew our Ai efforts.

Google had recently introduced the transformer architecture to the world via their landmark paper, Attention Is All You Need, which highlighted the potential of Generative AI. We built a prototype, aptly named Whole Call Summary, that would generate a one-paragraph summary for a conversation. It was exciting—Look, we can do this kind of thing!—but it wasn’t real time, the analytics were limited and we were still working on accuracy improvements. We had, you might say, half a flywheel.

A few years later ChatGPT showed the world GenAi’s potential, and reality caught up to our daydream. With GenAi we can finally close the loop—proprietary models generating insights to improve communication and deliver organizational value, which lets you grow your business, feed more data to your models, adapt, build more features, earn new customers. Rinse and repeat. Ai today is a true flywheel—for developers, sure, but also for our customers, and their customers, and beyond.

The future of optimization

Don’t get caught up in “one model to rule them all”; what we’ll need in order to realize the astounding potential of GenAi tomorrow are tools that let anyone build the models and platform that best suit their company, their data, their goals. The goal for solution providers has to be to make this transition to the Ai future accessible, controllable, and above all, fundamentally human.

We keep hearing from a certain sector of the Ai community that human beings are in the way, slowing the progression of full automation. I don’t buy it. There’s no doubt that Ai will continue to disrupt the market; there are lots of simple, repetitive tasks that we’ll happily automate. What’s harder to imagine in advance are the new, better jobs that Ai will create as new industries become possible. The assembly line might have put lots of skilled craftspeople out of work, but in the case of the automobile industry it eventually created far more opportunity. Electronic calculators might have replaced human calculators, but they quickly became a force multiplier. The rise of artificial intelligence is the next chapter in the ongoing story of humans evolving with tools and accelerating at an amazing pace, ever since the dawn of the Industrial Revolution.

That story is called optimization. We create prosperity when we find ways to do things a little bit better, whether it’s writing less code to perform the same function or a little bit more code to make the process faster or cheaper. The challenge for each new generation is, do we truly understand how our processes are supposed to work? How to build and adapt those experiences for them and optimize them safely? Are there better tools for all of these jobs?

No matter how we wind up answering those questions, the Ai flywheel will continue to gain momentum, helping us build more meaningful feedback loops by sharpening and polishing our existing tools and finding new and better ones. It’s a privilege to be able to contribute to this exciting chapter in an ongoing story.