Is the stage set for emergence of Intelligence from the chaos of Internet?
How close is Internet infrastructure close to mimicking human mind?
At the turn of the century I’ve become interested in the size of the Internet and how soon could we see the emergence of intelligence from the chaos that we call the web. The premise is that chaos breads its own structure as at that time I was fascinated with fractals. This kind of became reality with big data and neural networks that at the end of the day are structured chaos that gives probabilistic output. Later, in 2008, I stumbled upon a very interesting article that rekindled my interest, Evidence of a Global SuperOrganism. I was especially fascinated by the line that reads:
For instance, in 2002 researchers analyzed some 300 million packets on the internet to classify their origins. They were particularly interested in the very small percentage of packets that passed through malformed. Packets (the message’s envelope) are malformed by either malicious hackers to crash computers or by various bugs in the system. Turns out some 5% of all malformed packets examined by the study had unknown origins – neither malicious origins nor bugs. The researchers shrug these off. The unreadable packets are simply labeled “unknown.”
If there is recent research that looks at this please comment below.
I wanted to know just how much information could our Internet infrastructure contain in a single second without it being recorded, buffered, or otherwise stored.
So, I spend some time collecting information about undersea cables and landlines from all over the world. The collection was tedious as not much information was centralized but the data collection moved along. Raw data can be found here but once I decided to revisit this idea I rechecked the interwebs for new information and was pleasantly surprised that FCC started to publish the data in 2016 about annual circuit capacity data for U.S.-international submarine cable systems. For more up-to-date information check out the FCC website.
Today, there are websites that track all underwater cables that run all over our world and make the backbone of the Internet, like submarine cable map and infrapedia. The submarine cable map is thankfully open source so the code and more important the data is on Github. All the data is broken into areas so I needed to make a small python script that created an aggregate excel file with all of the data.
But, soon I noticed that not all, in fact, the majority of cables don’t have their capacity/speed listed so I went back to FCC and my own data.
First off we need some references.
The base unit is DS-0 or a “telephone circuit” and it is 64Kbps. 24 of DS-0 comprise the T1 connection, a dream of any geek in the early 2000s. Two T1 combined make T1C and combine two of them and you get a T2 connection. Combine seven T2 and you get a T3 connection. So the largest commercial connection is a T4 connection with a speed of 274.176 Mbps and is comprised of 168 T-1 or 4032 standard digital PCM voice channels.
Landlines were, before 2000, the only way home users could access the internet. So, let’s have a look at terrestrial landlines, the original way people used to access the internet. In 2005 there were 13,572,366 64kb circuits in the world and that rose to 33,228,707 by 2015. This to me seemed a low number so there could be a huge error in it.
This amounts to about 2.1 Tbps. Which is a lot but not when compared to undersea transcontinental cables.
In 2020 there were 59 self-reported cable operators with a total planned capacity for 2022 of 633.3 Tbps.1
Americas 3,946,686.5 Gbps
Atlantic 613,270.0 Gbps
Pacific 506,632.0 Gbps
Total 5,066,588.5 Gbps
Or 5,066.589 Tbps or about 5 Petabytes per second.
The total of devices connected to the internet is about 35.82 billion devices in 2021 and estimation is that that will rise to 75.44 billion by 2025.2
Research from Standford found that “In a human, there are more than 125 trillion synapses just in the cerebral cortex alone”. Another research from the University of Texas found that there is a minimum of 26 distinguishable synaptic strengths, corresponding to storing 4.7 bits of information at each synapse. This results in about 73.43 Terabytes of data. 34 Or, slightly less than four of modern hard drives.
By this, we have long surpassed the memory capacity of the brain but still not reached the complexity of the network to rival a human mind in terms of treating a single internet-connected device as a single neuron. Now, this reasoning is flawed as a single internet-connected device can do the work of many many neurons.
So, the amount of data that can flow through the wires of the internet without the need for storage is already an order of magnitude more than a single human mind and possible interconnectivity is even greater than that.
But how could we recognize if intelligence exists in the bowels of the Internet? We cant, It could be as foreign to our view of what Intelligence is as a creature whose heartbeat takes a century. To us, it would look like a plain rock, an inanimate object. We could only recognize it when it enters our frame of reference and talks to us. It’s like a Person of Interest TV show where AI communicates only by a string of numbers. It’s a good TV show, check it out.
If you take a look at my post about Embedded Security, in 2012 a single researcher with good intentions armed with only a simple list of default passwords managed to create a botnet with 420.000 devices but could have infected 1.2 million unique unprotected devices.5 This was in 2012 when the Internet of Things was just ramping up. In Q4 2021 there were over 1000 active botnets with an unknown number of compromised devices. But if we take into consideration that only 1 in 20 IoT devices is protected the number could be significant and one device could house multiple botnets.
Why am I mentioning this? Well, a crafty person could deploy machine learning on compromised computers, something like TensorFlow Lite for Microcontrollers which
is designed to run machine learning models on microcontrollers and other devices with only few kilobytes of memory. The core runtime just fits in 16 KB on an Arm Cortex M3 and can run many basic models. It doesn't require operating system support, any standard C or C++ libraries, or dynamic memory allocation.
So instead of using expensive hardware or huge computing farms one could, in theory, create a botnet and use it like a neural net and in the process maybe, just maybe give rise to a global super organism, muahahaha.
https://www.fcc.gov/international/circuit-capacity-data-us-international-submarine-cables
https://www.statista.com/statistics/471264/iot-number-of-connected-devices-worldwide/
https://aiimpacts.org/information-storage-in-the-brain/
https://elifesciences.org/articles/10778
http://census2012.sourceforge.net/paper.html