Decoupling Intelligence
Future is distributed
On our lineage, the chordate design emerges, with a cord of nerves down the middle of the animal’s back and a brain at one end. This design is seen in fish, reptiles, birds, and mammals. On the other side, the cephalopods’ side, a different body plan evolved, and a different kind of nervous system. These nervous systems are more distributed, less centralized, then ours. Invertebrates’ neurons are often collected into many ganglia, little knots that are spread through the body and connected to each other. The ganglia can be arranged in pairs, linked by connectors that run along the body and across it, like lines of latitude and longitude. This is sometimes called a “ladder-like” nervous system, and it does look like a ladder embedded within the body.
In an octopus, the majority of neurons are in the arms themselves— nearly twice as many as in the central brain. The arms have their own sensors and controllers. They not only have the sense of touch, but also the capacity to sense chemicals to smell, or taste. Each sucker on an octopus’s arm may have 10,000 neurons to handle taste and touch.
- Other Minds: The Octopus, the Sea, and the Deep Origins of Consciousness
The silicon world is mostly centralized. Our mobile and web apps sends data to the cloud to our smart devices at home, shipping bits back to a cloud enabled AI services. This architecture is akin to a central nervous system where all the sensory information is shipped back to a centralized brain to make most of the decisions.
A modern representation of intelligence on silicon is essentially machine learning where centralized data is used to train data in large models, really large models.
This decade will usher in a new era, where intelligence will move to the edge and decisions will be made without centralizing data. Additionally, decisions made by independent edge devices (agents) will work cooperatively to provide a larger “networked” intelligences.
The intelligence will go to where the data is.
A few reasons we can think why this will be the case:
Sensors (MCUs) have dramatically reduced in price, making edge hardware cheap
5G will bring low latency and high bandwidth leading to higher density of edge devices
Billions of edge devices to be launched over next few years
Every industry in automotive, energy, healthcare, telecom and others will be deploying edge AI to make faster decisions
We are entering into an era where we will decouple intelligence from the hardware and systems it was designed to run on.
HOTG’s Mission
We have talked about our ideas on TinyVerse. We launched our first product Rune. We shared key ideas behind localized intelligence and the future of orchestration on the edge while laying out the key idea of how we can distribute intelligence.
In this post, we want to articulate what our decade long mission would look like. Our team are very ambitious and want to play a key role in the future of distributed intelligence.
HOTG is building the distributed infrastructure to pave the way for AI enabled edge applications
The world of distributed AI enabled edge applications is weak at best today, lacking good developer tools, reliable orchestration platforms, and missing enterprise grade infrastructure to administer security, monitor deployments, and observe emergent behavior.
We are solving important problems in an Edge-First distributed future
HOTG is building towards solving these problems for the long term:
Intelligence is tightly coupled - AI models are tightly coupled to the devices they are built for.
Device Fragmentation - The future of edge also means there are multitude of devices to develop for. Costs to support and scale in a fragmented world are non-linear.
Lack of Orchestration - Infrastructure for deploying and managing edge AI systems hasn’t been built yet.
Our approach to solving these big problems (to put it succinctly) is by building the dev-tools and infrastructure for:
Develop at warp speed - Accelerate development of edge AI with portable Rune containers.
Automate deployments - Reliably deploy edge AI apps to any device using Hammer.
Device native orchestration - Tools and protocols for managing a distributed intelligent device network.
Market Opportunity
(Note: this does not include mobile and desktop markets which are big themselves and in HOTG’s addressable market)
The market of distributed edge AI will be massive, and is growing at an astounding rate. Some stats to think about:
Estimated 224 million tinyML compatible devices will ship in 2021, bringing $21.6 billion in revenue.
The market is forecast to grow at a 40.6% CAGR to $60 billion in 2024.
In IoT, software is an incremental annual revenue opportunity of 40% to 50% of hardware revenue. TinyML software market will grow to a multibillion dollar industry, as is already the case for AI & ML training software
The market of distributed edge will be a beast in this decade, and it has already started to grow. We are super excited to be playing a part in shaping and owning bits of this pie.
HOTG is at the sweet spot created by the convergence of following phenomena:
Edge Computing
5G
Cheaper Sensors
Privacy
What we are about to witness is a fork in the world of intelligence. A decoupling from centralized systems to a distributed intelligent systems.
Biology has already witnessed this:
The to and fro of sensing and acting is characteristic of all known organisms, including single-celled life. In the transition to the first animals with nervous systems, the machinery of external sensing and signaling was turned inward, enabling coordination within these new larger living units. Whatever nervous systems might have been doing initially, the transition from Ediacaran to Cambrian saw a new regime for animal behavior and the bodies that enable it. Organisms became entangled in each other’s lives in new ways, especially as predator and prey. The tree continued to branch, a few brains expanded, and two experiments in very large nervous systems arose, one on the vertebrate side and one in the cephalopods.
- Other Minds: The Octopus, the Sea, and the Deep Origins of Consciousness
Our team previously developed edge computing, federated learning, and privacy preserving technologies for healthcare, and we have witnessed how compelling these are for privacy. It is a tip of the iceberg to what is oncoming across several other industries - push intelligence to the edge, decouple it from central systems.
We are very thankful to our early and new investors, who share our vision of how big this change will be. As we grow our team, products, and market, we want to collaborate with investors and partners that believe in this domain.
If you think this is you, drop me a note at AKSHAY ‘at’ HOTG.AI
Follow us on our socials:
Twitter @hotg_ai and @hammer_otg | LinkedIn | Discord | Discourse