As we continue to accelerate the use of AI, in business and in our lives, it makes sense to ask how the development of the Internet in the 1990s informed our current journey.
Prior to having a globally connected network at no cost, there really wasn’t a way to port information in the ways that have supported the Internet age, the age of big data, the cloud, and now the AI era. In other words, suppose you had a super-smart chatbot, and no Internet. The machine would be relegated to using costly landline communications, constrained and limited in all kinds of ways that, a quarter century into the new millennium, would seem so abundantly weird to us, the stuff of steampunk.
The Internet was transformative. AI promises to be even more so. And so, we look for clues, in the dot.com era, that can help us to predict how this new wave of technology is going to game out, including in the markets.
Lessons Learned
What can we learn from the age of the Internet? A lot of things. Some of them have to do with signposts that have guided us through the last 30 years: Moore’s law, and Turing tests, both of which we’ll dig into a little later.
One such fundamental lesson is the need for universal protocols. Of course, you could argue that the Internet itself is one big universal protocol, but to be technical, it’s built on concrete protocols like TCP/IP.
“In the 1990s, when the Internet was taking shape, there were multiple competing models including ISO’s OSI (7-layer) stack,” write Mallik Tatipamula and Vinton G. Cerf at Communications of the ACM. “However, TCP/IP succeeded because it was lightweight, pragmatic, and universally interoperable/deployable. In AI, we are again faced with competing approaches and fragmented stacks. The lesson here is clear: we need common, interoperable frameworks and APIs that everyone can adopt and extend. A ‘lowest common denominator’ that allows agents, models, and tools to talk across ecosystems.”
Then there are the business takeaways, like this one presented by Steve Case of AOL fame who I have also personally interviewed.
Case gives us this nugget, among others:
“Transformative technologies often take time to mature, but when the right conditions align, adoption can accelerate at an astonishing pace.”
That’s something to keep in mind.
And then there’s this from Ryan Hughes at Bull Oak:
“Looking back at the internet’s trajectory offers insights into what we might expect from AI: swift adoption, profound societal and economic impacts, and the inevitable challenges of integrating such pervasive technology into our lives. As with the internet, AI promises to revolutionize communication, entertainment, and the job market, but it also raises concerns about privacy, security, and the potential for economic disruption.”
All of this seems to point in a pretty consistent direction: strike while the iron is hot, get into things early, and avoid having to catch up later.
Pradeep Sindhu shares this view. In a recent fireside chat at Stanford, he spoke to Navin Chaddha of Mayfield about his experience, and what the Internet age can bring to us now.
From Xerox to Juniper
Revealing how he moved from Xerox to found Juniper Networks in the 1990s, Sindhu also noted some things he found interesting as he searched for new, transformative use cases. Two of them seem to be related, in pretty direct ways, to Moore’s law. One that he enumerated was around bandwidth.
“The bandwidth was increasing, literally doubling every six months, no matter how you measured it,” Sindhu said. “The number of servers, the amount of bandwidth, the number of websites, etc. That was one thing easy to notice.”
The other thing had to do with … well, bandwidth. Specifically, the cost.
“The cost of long-distance bandwidth had been falling exponentially, and that was a big reason why communication wasn’t as fast as it could be.”
He also identified what he called a “gap” – that routers were built with generic CPUs. Sindhu and others went to work on that problem, and Juniper took off, soon becoming a billion-dollar business as the Internet bloomed and boomed.
Jensen Huang Then and Now
Here’s a very interesting tidbit in light of Nvidia’s new crown as the biggest tech company in America.
Sindhu noted that even back in the 90s, Huang was working on some of the notable technologies that laid the groundwork for Nvidia’s rise.
Specifically, CUDA or Compute Unified Device Architecture, NVIDIA’s platform for general-purpose GPU computing.
“Nvidia landed in a very nice place because they had this machine which did design for graphics, but it (also) did vectors and matrices very well,” he said. “And it turns out that it was extremely well applicable to machine learning computations.”
Then he mentioned Huang’s work.
“Jensen has done fantastically well to actually recognize this,” he noted. “And you know, he was developing CUDA back then. I met him in 1995, before founding Juniper. Even back then, he was talking about this. It wasn’t called CUDA back then, but he said, ‘Hey, there is this language, a C-like language, for parallel computations.’ And he had this vision back then. So it took a long, long time to develop this platform.”
Solving the Challenges
Later, identifying bottlenecks, Sindhu talked about memory bandwidth, and connectivity, seeming to bemoan a shortage of copper.
“If I had known all the difficulties up front, I probably would not have started,” he said, giving advice to new career professionals. “So naivety is good, in a sense, because if you saw all the difficulties that would face you, you wouldn’t get started.”
He also urged innovators to surround themselves with good people in small and agile teams, and to embrace ideas which are right for the times.
“A long, long time ago, humanity came across reading and writing, and there were people who probably poo-pooed that as well, and it made a sharp dividing line between people who could use that technology, and people who couldn’t. Folks today who ignore AI and poo-poo it are going to be left behind. They will be dinosaurs. And so, if I were growing up today, I would apply myself 120% to figuring out how this thing can help me do things better and faster. I think speed is always valuable.”
He ended with a warning:
“We must never forget that these models are being trained on abstractions, not on the real world,” he said. “So until we get to the point where they also have a complete understanding of the laws of physics, similar to what children do without actually being able to articulate Newton’s first and second laws, etc., there’ll be gaps.”
I tend to think there’s a lot of relevance here, not only in the lessons presented by those who wrote the above articles, but by innovators like Sindhu who were on the front lines back then. You can hear the echoes as they talk about the designs that created our global network, and the principles themselves seem evidently useful. Watch the AI market as this tech wave changes our world.
