The tragic story of Dragon NaturallySpeaking

Welcome to AI Collision 💥,

In today’s collision between AI and our world:

• A memory from the 90s
• The DNA of Dragon
• $19 billion but nothing to show for it

If that’s enough to get the Dragon speaking, read on…

AI Collision 💥

My brother is currently away in Japan. I’m not sure if he was wondering the streets of Akihabara (Tokyo’s famous tech district) which sparked his memory but he flicked me a message about a throwback to a tech job he used to have in the late 90s.

At one point he’d landed a job working for a company called Dragon Systems. Had no real idea who they were at the time, and they certainly weren’t well known.

They placed him inside a major technology retailer in Australia called Harvey Norman, at a Dragon Systems concession stand to sell their flagship product, Dragon NaturallySpeaking.

The software was a voice recognition software that allowed a person to talk to their computer and have their voice converted into text. It was a speech-to-text model.

I remember him bringing home a demo version of the software and it was absolutely fantastic. You’d put on this little headpiece (the kind you’d expect to see in a call centre) with a microphone on it, boot up the software and then just talk to the computer and watch your speech appear on the page.

It was revolutionary… but for the life of me, it never really exploded into a massive household name. And I could never figure out why. But the idea of a computer recognising speech was a real game changer.

The message he sent me from Japan said, “hey remember that NaturallySpeaking software from Dragon Systems I used to sell, what do you reckon ever happened to them?”

So, we went down the rabbit hole, and the story is incredible, a little tragic… and also the very foundation of AI as we know it today.

It’s easy to forget this whole AI “revolution” didn’t start with ChatGPT. Or Google DeepMind. Or even really Nvidia. It started in the basement of a former physicist at NASA, Dr. James Baker.

In the late 1980s, Baker had a wild idea, what if a computer could understand you when you spoke to it?

Not just transcribe your words phonetically, but actually listen, like a secretary or a friend. The kind of thing that was true science fiction at the time.

So, in 1982, he and his wife Janet founded Dragon Systems in Newton, Massachusetts.

Their mission, continuous speech recognition by a computer system. Not commands like “open file,” but naturally flowing language. Hence the name of their flagship product, Dragon NaturallySpeaking.

The first version launched in 1997, and if you were lucky enough (or nerdy enough as we were) to use it back then, you’ll remember the magic.

Sure, it needed training to set up properly. Sure, it struggled a bit with heavy accents.

But for the first time, your computer listened like a human could. You could talk to it in conversation, and it would reflect that speech. You could write novels without typing and easily get your thoughts down into a Word document without ever touching the keyboard.

It had a clear market for people with certain physical disabilities. And for use in the corporate world. But for the average home computer owner, it never really quite hit its strides.

What was not evident at the time to me (or my brother) was that Dragon was an early iteration of AI as we know it today.

Beneath the surface of NaturallySpeaking was a statistical model based on Hidden Markov Models (HMMs). That’s a core technology that forms the groundwork for many of today’s large language models (LLMs). These early HHM networks were probabilistic not the neural nets used today, but they were the bedrock of which neural nets developed from. It was in short, AI, learning patterns in data to predict and guess what comes next.

Sound familiar?

Dragon did sell a lot of units, and were quite successful, but never huge, and certainly not a household name.

NaturallySpeaking won awards, drew loyal users, and inspired a generation of researchers. But business was harder than breakthroughs.

In 2000, Dragon Systems made a tragic error. Looking to scale, cash in and exit, the Bakers agreed to a $600 million acquisition by a Belgian company called Lernout & Hauspie (L&H).

It was originally supposed to be cash and stock, but “advisers” at Goldman Sachs convinced the couple to take an all-stock deal with L&H.

The deal closed in June 2000.

However, by November 2000 L&H imploded in an accounting fraud scandal and were declared bankrupt.

Dragon’s founders lost everything. Their visionary company was swallowed by the chaos. And in the lawsuit against Goldman that followed, Goldman got off without even a slap on the wrists.

The Bakers had built a bedrock of AI, sold it for a lifechanging fortune and in months had it all ripped away.

While the Baker’s fortune was gone, the tech survived.

After bouncing around through a few corporate hands via mergers and acquisitions, Dragon’s crown jewel, NaturallySpeaking, ended up with a company called Nuance Communications.

They had the foresight to keep refining and developing it, then expanding its use into healthcare, customer service, and enterprise voice tech.

You may not have known it, but if you ever dictated a doctor’s note, navigated a phone tree, or used early Siri integrations… it was Dragon behind it all.

Then came the big deal.

Announced in April 2021, and closed in March 2022, Microsoft quietly completed a $19.7 billion acquisition of Nuance.

At the time, the headlines focused on a big deal in healthcare. But under the surface, Microsoft was absorbing one of the richest voice AI datasets and toolkits in the world… Dragon.

Decades of training data, models, speech-to-text engines, and fine-tuned applications across industries all now fed into the Azure AI stack. My take is that Microsoft knew exactly what they were buying, and it was more about Dragon, than it was anything to do with healthcare.

Which brings us to now.

Microsoft has integrated the engine of Dragon and Nuance’s technologies into their own AI stack.

Today, when you ask Copilot in Word to write an email, or dictate a message in Microsoft Teams, or use transcription in Outlook… you’re not just using CoPilot, you’re using Dragon DNA.

Great tech always finds a way to be where it’s supposed to be. While companies and fortunes might come and go, if the tech is game changing, and the future is clear, then tech always wins.

AI Scientist who worked on DeepBlue: “The AI Panic Is About to Begin” 

James Altucher – a 40-year AI veteran who helped build IBM’s Deep Blue – says AI markets are on the cusp of a panic. 

But not the kind most people are thinking of. 

James foresees a panic into AI stocks – a stampede of $15.7 trillion flooding into the sector. And right now, he says there’s a brief “wealth window” to get in ahead of it. 

Capital at risk.

Boomers & Busters 💰

AI and AI-related stocks moving and shaking up the markets this week. (All performance data below over the rolling week).

Boom 📈

• iRobot (NASDAQ:IRBT) up 23%
• C3.ai (NYSE:AI) up 7%
• Palantir (NASDAQ:PLTR) up 7%

Bust 📉

• Alibaba (NYSE:BABA) down 5%
• Arm (NASDAQ:ARM) down 5%
• AeroVironment (NASDAQ:AVAV) down 6%

From the hive mind 🧠

• More on this tomorrow… MechaHitler
• I wonder if they’re also asking if you send them one bitcoin they’ll send you two back? RubioCoin anyone?
• Geez I hope that ChatGPT doesn’t go the way of Grok! Would make my kid’s future schooling reeeeeallllly difficult.

Artificial Polltelligence 🗳️

This isn’t a poll today but a request for a trip down memory lane…

Or…

Weirdest AI image of the day

ChatGPT’s random quote of the day

Thanks for reading, and don’t forget to leave comments and questions below,

Sam Volkering

Editor-in-Chief

AI Collision