An Interview with Bob Metcalfe on the Future of AI, Knowledge Graphs & More

It’s rare to meet a mind like Robert Metcalfe’s. 

Someone nearly six decades deep into their career – with nearly six decades of achievements to match – is worth listening to when they muse on the future of their field. “Bob,” as friends and colleagues call him, has certainly had to eat his words when it came to past predictions, but those days have been in the rearview mirror for almost 30 years. Today, he’s still learning, building, and scanning the horizon for what’s next.

Metcalfe helped pioneer the internet in the 1970s – pushing the envelope such that even Harvard rejected his first doctoral thesis on ARPAnet – before he went on to co-invent Ethernet and co-found 3Com (later acquired by Hewlett-Packard). That work alone earned him the IEEE Medal of Honor, followed by a tidal wave of other awards still being handed out as late as 2023. 

Along the way, he penned Metcalfe’s law on the value of networks. Initially proposed to describe the value of a telecommunications network, Metcalfe’s law numerically predicts the value of any network – telecom, internet, social, and knowledge. 

But Metcalfe isn’t ready to hang his hat up just yet. He’s still researching, still teaching, still learning, and still connecting. The value of that growing network is proportional to the square of the number of people, ideas, and technologies he’s helped connect.

Today, his learning journey has led him to knowledge graphs. In this interview with KGC, Bob Metcalfe talks about the future of artificial intelligence, the value of knowledge graphs, the potential for the Semantic Web (still a thing) and blockchain (also still very much a thing), and  of course, why he’s attending this year’s Knowledge Graph Conference. 

Given your distinguished career of inventing and pioneering new technologies, where and what are you working on these days? 

Bob Metcalfe: I am a research affiliate at MIT Computer Science & Artificial Intelligence Laboratory (CSAIL), working with the MIT Julia Lab. I am also a Professor Emeritus of Computer Science and Electrical & Computer Engineering at the University of Texas at Austin. 

Finally, I’m an enthusiastic advisor to OriginTrail. My secret motivation for hanging out with them is that I’m counting on them to teach me everything I need to know about knowledge graphs. 

What projects are you most focused on in particular? 

Metcalfe: I have a million projects going on, but the one I focus on the most and work hardest at is learning about computational engineering, which is my sixth career. It’s hard to change careers, and this one’s proving to be very hard.

My research at MIT focuses on using AI to develop geothermal wells. In becoming a computational engineer, I had to look for a problem that really needed solving, and geothermal energy is grossly underrated. The field is blossoming right now, so it’s fun to be in the midst of it as it’s growing out of the lab and into production. 

Geothermal energy focuses on harvesting the heat of the earth, and so you don’t have to burn fossil fuels to create heat as it’s naturally occurring. The trick is to harvest it economically, which is what the AI modeling efforts are all about: looking for economic ways to harvest the earth’s heat energy. 

What are your thoughts on the value of knowledge graphs: past, present, and future? 

Metcalfe: I’m not an expert on knowledge graphs – I’m still learning – but as an internet engineer, I have a hunch that knowledge graphs are a key to unlocking the power of artificial intelligence. 

Right now, AI struggles with hallucinations. My initial hypothesis is that knowledge graphs are gonna fix that aspect of artificial intelligence by providing a systematic way to gather and apply knowledge. 

I spoke briefly about the future of knowledge graphs at the sidelines of KGC 2022 with the OriginTrail team.

You’re famous for proposing Metcalfe’s law, which states that the value of a network is proportional to the square of the number of connected users of the system (n²). Any reflections on how Metcalfe’s law applies to knowledge graphs? 

Metcalfe: If there were a field in which Metcalfe’s law applies, it would be knowledge graphs.

I wasn’t thinking of knowledge graphs when I came up with it a couple of decades ago, but the idea that connecting facts and knowledge together increases the value of that knowledge, including its usefulness and applicability, absolutely aligns with Metcalfe’s law. 

Maybe at some point I should do a research paper on the numerical predictive power of Metcalfe’s law to knowledge graphs. I haven’t done that yet. 

An illustration of the increasing value of a network. As more nodes are added, connections increase exponentially.

 

What other KG-related technologies are out there that you believe more knowledge graph professionals should be paying attention to? 

Metcalfe: The taxonomists, the ontologists, and the knowledge graph community have had their own private little world for a long time. They haven’t been in the middle of AI or in the middle of the internet. But the opportunity to get out of that bubble is now. 

There’s all that work that they’ve put into the systemization of knowledge and the connections among knowledge. This is a very good time to look for applications of all these years of work that they’ve put into knowledge graphs.

I would challenge the folks who come from the AI world to visit KGC and to tap into the knowledge graph community to look for answers to current problems. I would also challenge the knowledge graph community to look for opportunities in the AI ecosystem. 

Of course everyone’s talking about AI these days, but I think the KGC community can make a special contribution. 

You’ve previously said in 2022 that “For years, AI would rise and then it would fall, and it fell because AI ran out of data. This time, it’s not going to fall; it’s going to continue to rise, because decentralized knowledge graphs are going to give AI more and more data.” Given the current state of AI systems today, what do you think is just over the horizon for artificial intelligence?

Metcalfe: I’ve been in AI since 1968 with my undergraduate thesis on the neuron model, so I’ve had a long exposure to the comings and goings of AI. It’s really gone up and down a lot over the years. It’s certainly rising to a new point right now.

Previous model collapses were due to the lack of data, but we have the internet now. That hopefully means this latest AI push won’t run out of data. We have to pay attention to the data and make the most of collecting it. Knowledge graphs make that possible.

This time, AI might go further than usual because it has found ways to get new data. The danger is if it starts ingesting its own AI-generated data. That’s when you get model collapse. 

Do you think the connectedness of knowledge graphs will make a difference in terms of that classic rise and fall pattern you’ve seen with AI? 

Metcalfe: In previous times of AI blossoming, we didn’t have the Internet, but we have it now. The other thing we didn’t have is the connectivity part, finding ways to connect the data. Both of those are different this time around.

A lot of people think that facts are either true or false, but I don’t think that’s how it works. There are degrees of true and degrees of false. This reality challenges the basic logic for a lot of AI systems out there. 

But still, we have the internet, and if we overlay it with a knowledge graph to give it structure and connectivity, then that’s how AI saves itself this time. 

Talk to us about your connections with the Semantic Web and Web3 – both of which use the term Web 3.0 – and how some of that ties back to knowledge graphs. 

Metcalfe: The Semantic Web aims to create a web of data, connecting the data to each other and thereby enhancing its value. Tim Burners-Lee is the main Semantic Web guy. He actually held the 3Com founder’s chair at the MIT CSAIL which I endowed. Tim’s been working for decades to get the Semantic Web to blossom.

The folks at OriginTrail, where I’m an advisor, are also pursuing a Semantic Web angle, but they’re also working with blockchain technology. Blockchain was the hot buzzword before AI, but have you heard anything about blockchain recently? The same is going to happen to AI. Buzzwords come and go. However, these technologies still have potential, even if they’re not the current fad. 

At OriginTrail, they’re exploiting the blockchain in their knowledge web. They have a special focus on supply chains, which is where you have connected systems for producing goods and services, and then tracking them and adding facts to the delivery and so on. I first met OriginTrail at the University of Texas, but they’ve broadened their charter beyond just supply chains and now they’re combining blockchain with a huge knowledge graph.

Having attended the Knowledge Graph Conference the past two years, what brings you back to KGC for a third year in a row? 

Metcalfe: I’m attending KGC to immerse myself in the language, structure, and current frontiers of this field. Knowledge graphs, ontologies, taxonomies have been flourishing at the edges of technology for a while, and now they’re coming into center focus with their applicability to AI. 

At KGC, I’m hoping to help knowledge professionals discover AI and then solve AI’s most pressing problems so we can have more powerful AI tools at our disposal. I’m not on the cutting edge myself. Personally, I’m still learning the language and who the people are. I’m looking for the knowledge graph breakthrough that will enable AI. 

2025 is my third year at KGC, and I’ve already made my hotel reservations next to the conference on Roosevelt Island. I’m a native New Yorker, so it’ll be fun to go home. 

Anything else you want to add or share?

Metcalfe: The nerd factor at KGC is pretty high, and I mean that with all respect. However, the AI world is not the same as the knowledge graph world, so they often have contending vocabularies for the same concepts. It’s important that we work out a shared language between AI and knowledge graphs.

In addition to the language issue is networking. Many times I’ll run into colleagues who should know about each other, but they don’t. That’s a sign of an early-stage network that hasn’t yet coalesced. In any early network, you want people and ideas to mix as much as possible. 

I think KGC is a good forum for both these conversations to take place.

 

The world of knowledge graphs awaits: Get your ticket to KGC 2025 today and mix with AI and KG professionals – like Robert Metcalfe – from all around the world.

Get My Ticket to KGC

 

***

Photo of Bob Metcalfe: Andreu Veà, WiWiW.org, CC BY-SA 3.0