Connectivity is a thing, is THE thing

The most important new fact about the human condition is that we are now all suddenly connected. To make a long story short, connectivity is the apparatus for moving stuff from point A to point B. Apparatus might include pony express, airmail, roads, radios, optical fibers or the Internet. Stuff might include mass, energy, signals or Not Alone. If you would like to argue about what connectivity really, really is, please take that over to a Facebook group. Suddenly? The Internet began switching data packets only yesterday: on October 29, 1969. On that day, a grad student at UCLA tried to use several Internet packet switches to log into a computer at SRI several hundred miles up the coast of California. He typed “LO,” but before he could type the next letter, G, the computers crashed. Since then, lo and behold, 5 billion humans have connected to the Internet — two thirds of the human race.

Suddenly. With much better results.

By the way, people too often argue about what to call the 1969 version of the Internet. Purists call it ARPAnet, with ARPA being the US Defense Department’s Advanced Research Projects Agency. The evolving Arpanet didn’t become an “internet” until a protocol upgrade in 1983. From Arpanet NCP to Internet TCP/IP. People also too often argue about the capitalization of Arpa. While working on the Arpanet in the 1970s, I would regularly drive the couple hundred miles between our family home on Long Island and MIT in Boston. My Mom, Ruth, would say, “Bobby, call us when you get to MIT to let your Dad and I know you’ve arrived safely. Let our phone ring three times, and then hang up.” My mother loved me, but not that much. AT&T was still monopolizing telephone connectivity in the 1970s. They introduced the Model 500 telephone in 1950 and continued selling it until 1984. Perhaps they finally saw the Macintosh coming — or the iPhone or George Orwell.

Internet pathologies then …

Pointing out that the arrival of Internet connectivity was sudden is how I explain what caused the Internet’s several major pathologies. We got connected so fast that we were overwhelmed by connectivity, did not know how to manage it, how to harness it. We are learning. The first pathology I recall was hacking. In 1973, two high school students in Los Angeles broke into the Arpanet. They got somebody’s root password. We had not anticipated that people would want to do that. Fortunately, there was not much Internet to hack back then and not much of it valuable. Things have changed. The second major Internet pathology that I recall was pornography. The US Congress was so impressed by the early arrival of packet-switched porno that they almost shut the Internet down. Instead, in 1996 they passed the Communications Decency Act, thereby eliminating Internet pornography. A year later, its job done, the act was declared unconstitutional. The third major Internet pathology was advertising. We engineering professors and students who built the early Internet were outraged when advertising began to appear — crass commercialization of our beautiful connectivity. Not too much later, it was discovered that advertising would pay for the Internet. For example, Google took off, not because of its search engine but because of its monetization of online advertising. We don’t talk much about spam anymore.

… and now

Today we are focusing on new, more challenging connectivity pathologies. Our epidemic of loneliness is often attributed to spending too much time on social media. We worry that polarization will lead to violence. And there’s fake news, which gives us reason to censor what’s not true. I am optimistic we’ll find ways to mitigate these and other pathologies. But the next connectivity tsunami is upon us as we transition from the megabit to the gigabit Internet. New pathologies ahead.

The next connectivity tsunami

The Internet was originally built to carry text among time-shared minicomputer hosts. Today something like 91% of Internet traffic is video. Most of it is entertainment video. Some of it is a substitution of communication for transportation. Zoom. I recall before COVID that most people would reject substituting video for being there. Too many professors would reject teaching by video despite the overwhelming economies. Too many salespeople would insist on travel to close business, pressing the flesh to do their deals. Then COVID hit in 2019, which happened also to be the 50th anniversary of the Internet. It was almost as if we built the Internet in anticipation of COVID. Suddenly everybody got a taste of video connectivity. COVID’s taste of video connectivity has established trends that will be playing out for many years. Universities are hoping that all their new buildings will prove instrumental in new connectivity-intensive learning. Commercial real estate is hoping that workers will want to return to their offices. WeWork ran out of hope. Anyway, I want to propose that we re-purpose the COVID acronym. COVID is now short for collaborative video. The Internet is of course not the only kind of connectivity. And since artificial intelligence is so hot, it should have a connectivity story. Yes, indeed. Neural networks are a key technology in AI. The word networks is a connectivity word. Advanced AIs are reported to have neural networks with a trillion parameters. Parameters are connections, analogous to our brain’s synapses. It is the brain’s connectivity that explains why brains made out of pitiful neurons are way smarter than computers made out of modern transistors. Neurons are a thousand times larger and a million times slower than transistors, but still they out-compute transistors. How do we explain this paradox? Connectivity. The current leader in neural network connectivity is GPT-4 with 176 billion parameters. It is rumored that the next generation will have 100 trillion parameters. But still nobody knows if that will be enough connectivity to yield artificial general intelligence (AGI). Even if we soon get to 100 trillion parameters, that’s only 10^14 neural connections. Our brains have 10^11 neurons each with on average 10^4 synapses (connections, parameters), which suggests that 10^15 parameters will be needed to achieve AGI. Just looking at connectivity, GPT has another factor of 5,700 to go to match brain connectivity and perhaps achieve AGI. Connectivity is a thing, is THE thing. Start there.