Fetch AI aims to liberate AI for the masses with ledgers

Fetch.AI emerged from stealth-mode in March 2018, with $15mn funding and a grand ambition to liberate the benefits of AI and Machine Learning (ML) for the masses by creating a robust distributed virtual world for trading and interaction between autonomous agents.

Based in Cambridge, UK, it is easy to see how this ambition germinated since the company’s key founders were previously leading AI specialists at DeepMind and became disgruntled when Google took it over. CEO and co-founder Humayun Sheikh was particularly scathing of DeepMind’s sale to Google, arguing this made it impossible for it to develop independent AI platforms and that the technology will only deliver widespread benefits across society when freed from the big tech giants in general.

“I am passionate about commoditizing AI and ML so that you and I can deploy it when we want on very small things,” said Sheikh recently. “That’s when you will truly unlock the potential of AI.”

Work on the Fetch.AI platform therefore dates back almost to Google’s takeover of DeepMind in 2014, based loosely on the idea of blockchain to create distributed ledgers capable of orchestrating autonomous transactions. But Fetch.AI noted fatal flaws in the blockchain approach associated largely with its serial architecture, which fundamentally limits the rate at which transactions can be processed.

Blocks can only be added to a chain sequentially one at a time, which means that only one block can reference the preceding one in that chain. Some element of parallelism must be injected for blockchain to scale to large numbers of transactions at acceptable performance and that is what Fetch.AI set out to do. As CTO and second of three co-founders Toby Simpson put it, “We couldn’t scale or do the kind of computing we wanted in blockchain. We had to parallelize the execution of smart contracts and massively parallelize the operation of blockchain.”

The company noted in its inaugural white paper that payment channel systems such as Lightning or Raiden5 have been developed to address the blockchain scalability problem by taking large numbers of small payments off the chain. This does increase speed and decrease the cost of some transactions but is a fudge rather than a fundamental solution.

The company’s product, its Fetch ledger, has been designed to overcome this and also limitations of other approaches.  In essence this combines blockchain with a technique from mathematical graph theory, the Directed Acyclic Graph (DAG) – something also explored by IOTA. This yields a data structure that looks like a flow chart with all nodes pointing in the same direction but branching out into multiple lines that then converge again at various points. It resembles a file directory whose folders have subfolders that branch into other subfolders and back again in a multiple tree-like structure.

This according to Fetch.AI confers that elusive scalability where millions of agents can work alone or in groups to solve problems for themselves and for other stakeholders. It is also stable so that transactions of all types and speeds can proceed against a background of price stability. It also of course must be secure and trustworthy with risk well managed.

All this begs the question of use cases, and Sheikh gave the example of a country’s national health agency which typically invests huge sums attempting to integrate centralized but disparate systems to facilitate joined up healthcare. A system like Fetch would decentralize operation, and organize it around healthcare agents belonging to each individual, connecting to all databases including external sources, owned by the person, so that for example in the event of an emergency this would point to and rapidly obtain all the relevant information such as tolerance to drugs.

This in turn begs a big question for Fetch.AI, which is whether it can persuade such agencies to adopt its ledger. At this stage it is much about raising consciousness and staging convincing demonstrations, which is yet to come as commercial trials are about to begin. Enhanced code release incorporating some extensions based on AI and ML are scheduled for Q4 2018.