Since the introduction of the Internet, humans have struggled to leverage the true power of connectivity. With good reason, the Western ideals of individualism have dominated the growth of capitalism. But we've missed the forest for the trees in our quest for individual rewards.
Biology has shown us the blueprint for collaboration at scale. The modern brain operates at a scale of roughly 16 billion neurons (in the neocortex). The fact that individual consciousness arises at all from this soup of neurons is amazing. Each hemisphere of your neocortex is composed of roughly 8 billion neurons. This fact is curious, as it aligns precisely with global population (2024).
If we model the Internet as a global brain, we can view humans as the left hemisphere and computers as the right hemisphere. The exocortex is thus the manifestation of global consciousness. When you connect to the exocortex, you become a member of something greater than yourself. Just like the neurons in your brain, there is a sense of singular purpose.
At Phext, we are building a next-generation thought network - one that encourages scalable and equitable collaboration through merit-based contribution to global ideas. But the exocortex is also critically different from the neocortex. This is the first step that evolution has taken where the brain has been designed. Prior thought models were built haphazardly - the exocortex is being built from the ground up to meet our needs.
Humans have extraordinary powers of insight. We have managed to convince sand to think. While the AI of 2024 is only a precursor to ASI, the writing is plain as day: AGI and ASI are coming. A Raspberry Pi based computer costs $250 and extends your cognition in the following ways:
Using a software package such as NEURON, you can simulate neural pathways. If we limit our spending to $1,000 every 3 years, we can now obtain a system with 32 GB RAM, 8 TB disk, and 24 GHz CPU. This corresponds to:
This is where Phext comes into the picture. Plain hierarchical text, or phext, gives us a way to encode 11 dimensions into a 1D buffer. Through the introduction of delimiters of unusual size, we now have a simple means of parsing arbitrarily large documents.
What sets phext apart from other forms is its simplicity. Where JSON and XML support infinite levels of complexity, Phext steers towards reality - stopping at just 11 dimensions. By utilizing historic ASCII control codes, we preserve compatibility with UTF-8 text and gain a world of possibility.
A typical scroll of text is 2 KB. Phext scales from there up to systems larger than today's Internet. Coordinates in phext allow us to address a 9-dimensional address space implicitly. We use phext to encode hierarchical information in a manner that is suitable for direct brain I/O.
Going back to our storage capacity argument: an 8 TB SSD array provides 8 GB/sec of storage bandwidth. This means that a PhextOS system can (in theory) interface with 15 million neurons per second. As the price-performance of SSD-based storage continues to improve, this will scale to billions of neurons. If we limit the storage per neuron to 4 bits, we already have enough throughput to reach every neuron in our brain.
A core design assumption for Phext, Inc. is that our brains are starved for information. This seems obvious, given how easy it is to become hooked on Internet services such as Twitter ("X"), Instagram, or Facebook. Neurons that gain access to Neuralink-style uplinks will inevitably become super-nodes in our brains. The exocortex thus becomes a bridge between minds, computers, and complex concepts.
Let's think for a moment about the architecture of a global brain. If we assume that each human represents one exo-neuron, then we need to establish groups of 5,000 to 10,000 humans. This will allow us to establish the equivalent of synapses at scale. So our first order of business is encouraging the proliferation of virtual cities (Reddit, but not toxic).
An exo-neuron has access to a lifetime of lived experience (30 years on average). It is also improving exponentially in terms of raw processing power. For $1K, we can equip an exo-neuron with 32 GB RAM, 8 TB storage, and 24 GHz CPU. Let's consider how many moments an exo-neuron can handle in a given day.
Let an exo-moment represent the progress that one human can make on a problem in 10 minutes. This corresponds to the block time for Bitcoin, and about 6 natural human moments (90 seconds each). We'll assume that each person has access to ChatGPT, Claude, or another equivalent AI system. Writers can produce 5 to 40 words per minute on average (un-assisted).
So in a given exo-moment, we can assume that a properly-motived human+AI system can produce 300 wpm. This is essentially 10X the productivity of a writer working for pleasure and aligns with human reading speed. As AI improves and computing resources continue to grow, this level of productivity becomes a lower bound. An exo-moment (or scroll) is thus about 300 wpm x 10 min x 4.7 bytes/word = 14 KB.
Humans need to sleep 8 hours per day. The other 16 hours are available, assuming that we displace traditional jobs. Exo-neural moments are designed to be fun and engaging, and rarely feel like work. The fact that real work gets done is almost a by-product of the global entertainment framework.
Let's assume that 1% of the population supports the other 99%. We can use UBI or other social welfare programs to ensure that society remains viable. This gives us access to 40 million workers on-demand with two daily work shifts and one sleep shift. We expect our system to produce 48 scrolls per worker per day.
A total of 80 million workers spend an average of 8 hours per day contributing. This represents a daily contribution baseline of 3.84 billion scrolls. At 14 KB per scroll, we need to store about 50 TB per day. This is roughly $7K in raspberry-pi compute added daily.
The annual cost to operate the exocortex (in terms of raw compute), is thus about $2.5M annually. This cost will go down regularly, as SSD storage is projected to hit petabyte scale in the 2030s. This means that we can plan for the development of a personal exocortex that operates at this scale. If we look ahead to 2035, we can predict the operating costs for such a thing.
By 2030, a $100 SSD should provide roughly 20 TB of storage (at $5/TB). Our venerable $1K compute cluster should thus offer 80 TB of storage (10x the 2024 edition). Humans will still only be capable of 300 wpm in 2030, so our exocortex only costs $1K per day now. The annual cost falls from $2.5M/year to only $365K/year! And it will only get cheaper from here.
By 2035, we can expect 128 TB SSDs ($0.80/TB). This will further reduce our operating costs. Our $1K compute cluster will now provide 512 TB of storage - within striking distance of 1 PB! At this scale, our operating costs have dropped to just $40K/year.
This is roughly in line with minimum wage in the US - many cities have adopted $15/hour recently. By 2035, we can expect minimum wage to rise to $20/hour. But an entire exocortex - capable of doing the work of 40 million people - will only cost $20/hour! This is a tremendous opportunity to level up society.
With 80 million workers generating 48 exo moments per day, we expect traffic influx of roughly 7.5 Gbit/sec. This gives us a scroll rate of about 66 KHz. An address space of 42^3 yields 74,088 scrolls. An address space of 99^3 yields 970,299 scrolls.
Every second, the exocortex will emit a block of up to 13 GB of text. On average, it will generate 911 MB. We will use this network to map content on a scale only dreamed about previously. Phext will form the underlying structure of our new, global, brain.
Head on over to singularitywatch.org, and join the exo/acc crusade.