In the past five years there has been a rapid growth in the number of practical Artificial Intelligence (AI) applications around us. Smart services, like self-driving cars, face and voice recognition in mobile phones and image translation are getting a central place in everyday life. This rise can be explained by the advances in machine learning research and the ready availability of cloud computing. The Effect Network is designed to provide a feature complete alternative to Microtasking, AI as a service and Machine Learning Computation, and operates fully on smart contracts deployed on a Turing-complete blockchain.
According to their whitepaper the main reasons that currently make AI development inaccessible for individuals is data processing, diverging tasks and computational cost.
These points are claimed to be solved by the Effect Network. Like other decentralized applications, Effect directly connects supply and demand without the need for an intermediary party. This brings the following advantages; Accessibility, accuracy, performance, interoperability.
The Effect Mechanical Turk platform is a decentralized, peer to peer marketplace for tasks that require human intelligence. It provides similar features as centralized services like Amazon Mechanical Turk, It is a crowdsourcing technology that enables requesters to submit tasks that can be completed by human agents in exchange for compensation. Users can work on tasks from requesters at any time, anywhere and from any device.
The tasks are called Human Intelligence Tasks (HIT). The providers of the HITs are called requesters. When a worker completes a task they are paid with the token AIX. Requesters can put tasks on the Effect Mechanical Turk platform to be completed by workers. They can decide how many AIX the workers will get for each completed task and the workers can complete the tasks from the requesters in exchange for the AIX tied to these tasks. Afterwards requesters can retrieve the results from the platform and use these results to train their AI algorithm. Effect Mechanical Turk gives requesters access to an on-demand, scalable and distributed workforce.
A task represent a piece of work that has been submitted by a requester, and can be accepted by workers that match its requirements. Each task points to a data set that can contain any amount of media assets. The contract ID of the task will validate the format of the data.
The structure and required feedback for a task is defined by the contract ID and the blueprint. Each type op task requires a smart contract to handle interaction. Effect maintains a database of deployed smart contracts to make it easy for requesters and workers to interact with the network.
Decentralized AI Exchange
The Effect Exchange is a decentralized marketplace where AI algorithms can exchange their services. An application owner can register on the exchange by specifying a public endpoint of his application, following their data interchange format and specifying a usage fee for consumers. This application can now be called through smart contracts on the blockchain.
Application endpoints on the Effect AI Exchange communicate over the HTTP protocol. Data is exchanged in JSON format and the defined RDF schema. RDF Schema provides a data-modelling vocabulary for RDF data. RDF Schema is an extension of the basic RDF vocabulary. (2)
Decentralized AI algorithms
In the whitepaper they claim that in the final phase of the network the actual computation will be distributed, so that the algorithms run globally without a failure. To achieve this they use the fact that most machine learning algorithms have rigid structure, and operate on sets of weights. The Effect decentralized compute engine is based on deep-learning networks like Caffe (3) and Tensorflow (4), where the network structure can be defined as a declarative graph and weights are stored as matrices of real numbers. These matrices can be distributed over a decentralized file system and be processed at different compute nodes on the network.
In order to achieve liquidity in AIX, they claim that the Effect Network will maintain a central pool of tokens to provide liquidity, encourage adoption and stabilize network fees. This pool is called the Galaxy Pool and consists of a mix of EFX and native tokens. Several rules will drive the Galaxy Pool towards stability. These rules can later be refined by means of governance. As the Effect Network is decentralized there can not be a single person or organization authoring these changes. To fix this the network has a governance system that allows prominent people in the community to propose and vote for improvements. Right to vote is at first acquired by selected individuals.
The Galaxy Pool ensures stable exchange rates for users of the platform at all times. Any coin that is bought from the Galaxy will initially be tainted, and a tainted coin can not be sold back to the pool. By tainting the coins they protect the Galaxy Pool from external manipulation and keeps exchange rates stable for workers. A tainted coin is converted to a regular EFX token by spending it through an Effect service contract. These are the service contracts from the tasks and service registry.
On the Effect Mechanical Turk workers are rewarded tokens for their effort. This could make malicious users to gain wealth by submitting a large quantity of tasks with poor quality. To avoid this, the network will appraise users by their quality of work. Users that put in good effort will be rewarded with Honor Tokens (HNR). These tokens can not be traded or sold, but will gradually expire over time. Workers with a large number of honor tokens will be able to apply for more high rewarding tasks and will have to pay less tax to the Galaxy Pool. HNR tokens are credited to users when they are rated for good work.
Company and team
3 of the 8 team members have worked at Itsavirus, a software and interaction design company which worked for Heineken and Shell. (5)
CEO: Chris R. Dawe, Business management and marketing Project manager of Itsavirus. (6)
Lead Developer: Laurens Verspeek, Bachelor and Master of Science in computer Science. Lead developer Itsavirus. (7)
Lead Blockchain: Jesse Eisses, Bachelor and Master degree Artificial Intelligence. Lead developer Itsavirus. (8)
Lead Designer: Nick Vogel, Nick studied Communication and Multimedia Design at the Hogeschool van Amsterdam. He was also the lead designer at Itsavirus and is the Co Founder of Effect.AI (9)
At this moment there is no partnership announced yet.
The token crowdsale is planned for Q1 2018 and more precisely in march according to their website. (10) In their roadmap the Effect M-Turk will be fully released by Q3 2018, the CEO mentioned that the M-Turk will be out way before Q3. By Q2 2019 the Effect Smart Market will be fully released. That’s the decentralized AI platform we discussed earlier, where people can offer and buy AI services. Eventually in Q4 2019 the Effect Power will be fully released, where the platform will run deep learning frameworks.
Please note that this roadmap is still being worked on and is not definitive, it is however an estimate