A quick guide to creating autonomous AI agents.
If you haven’t read our introductory post, here’s a quick summary:
ScyNet.ai is a decentralized platform that incentivizes independent nodes to compete in finding better AI models and data for specific tasks.
There could be numerous applications — algorithmic trading, autonomous cars software, image-reading or speech-recognition and other.
The development of advanced AI has so far been reserved exclusively for large profit-driven organizations, due to their access to large server capacity. Not only ScyNet would provide a viable alternative, but it would be utterly transparent. And because the code is open-source, it’s now possible for anyone to join and contribute to the network as they will.
Some might ask?
How would it make money?
What’s the value?
ScyNet’s final product is Artificial Intelligence. That’s a broad concept.
AI isn’t magic. It’s only a very sophisticated function that optimizes to do well in, at least for now, a narrow task. The way that works is it takes data as an input, processes it, and then proposes a prediction as an output.
For example, an AI agent, designed to trade Bitcoin, would use as an input the historical price and the current state of the market. Then it would try to predict how the price would move in the future:
84% chance that BTC/USD will go up in the next 1 h.
93% chance that BTC/USD will go down in the next 30 min.
The AI would then compare its guess (the output) against what happened, and it would adjust its approach accordingly so that it could make a more ‘educated’ forecast the next time. This is what data scientists call ‘training’ the AI.
The business case on ScyNet is as simple as this — to use an agent’s output; consumers have to pay the owner. Thus, everyone involved in the creation and training of an AI gets a reward for their work. A small part of the money automatically goes to support the underlying open-source platform.
We aim to design a fully autonomous, self-sustaining, circular AI economy, providing an incentive for all contributors involved. Who would be these contributors, though?
There are three critical points in the development of a Machine Learning agent:
Writing the code.
Feed the AI with data.
Train it to perform.
One of ScyNet’s main advantages before other attempts at decentralized AI networks, s that it provides a platform for the creation of AI via its inherent NAS algorithms. In other words, there’s no need to write the code for each neural network, so we skip this point.
The Automated Machine Learning happens in what we call “Hatcheries” in ScyNet slang. Essentially, a hatchery is an AI that creates new AI. The hatchery is designed to autonomously become better at its task, meaning, as time passes, the later generations of AI would be better than their predecessors.
In its upcoming testnet, ScyNet would already have some of its hatcheries built. Since the platform is open for new programmers to join and build upon it, it would be possible to add more hatcheries to tackle more problems with Machine Learning solutions.
Hatcheries would need computational power to do their thing — we call this process training the AI. Just like a baby whose brain has a vast potential, but lacks essential experience in the surrounding world, new AI needs to be fed with data and trained to perform accordingly.Data providers. AI systems require data to improve, just like we humans need food and water to live. That is why the platform would reward those who supply the system with well-structured input. An AI designed to recognize early-stage cancer would benefit significantly from a massive database of MRI scans, labelled from expert doctors.As previously mentioned, the process of training requires processing power. The more you have, the quicker your model would learn. Trainers on ScyNet would be people who apply their robust computers to help AI agents improve faster just like the miners we know from PoW blockchains.
Breaking down this process of creation is a crucial step towards the development of a decentralized self-sufficient AI network. ScyNet achieves it by nurturing an internal token economy with an incentive for every contribution, whether that is a training dataset or computational power.
Effectively, the platform strives to make the latest AI tech accessible to many smaller organizations and individuals, despite each of them not being able to match the resources of the tech giants in the industry.
We are still building the nerve system of ScyNet, although we have progressed a fair bit down the road. We are still exploring possible mechanics of ScyNet tokens, intoake the network run as smoothly as possible. Besides its distinct functions for communication, accounting and access to the platform, there are an array of small triggers to stimulate the trouble-free evolution of a decentralized network.
The question about what makes the demand for the token is fundamental to defining its value, so we are looking to find the comprehensive answer that would satisfy us.
Along with the blockchain protocol, we are working on the first real-world application built on the ScyNet platform. Hatchery One will provide validated AI-driven bots for the trading of crypto assets.
Why not join our discord now and figure out first what we’re up t?