AI Network
Published in

AI Network

AINFT — Make AI accountable, reproducible, and valuable

TL;DR

AINFT is an immutable identifier that represents the connectivity of AI components. It is an accountable, reproducible, and valuable unit of AI managed by communities on the Internet for AI.

The role of communities in the development of artificial intelligence

In many dystopian movies such as Skynet, large transnational corporations develop AI in some wrong ways and become the worst nightmares of humanity. AI Network believes that no single individual or organization can fully own intelligence. Just as no human being can live in complete isolation, by definition, intelligence is social progress that can only exist by continuously interacting with data from the world.

While AI is rapidly evolving, cautious ethical approaches are being made by AI companies. For example, OpenAI provides comprehensive guidelines for the ethical use of the large model GPT3 and evaluates the use of their APIs to control their impact on AI industries.

AI Network goes a step further and believes that when these efforts are managed together as a community, we can truly create the “AI for everyone”. Humans and organizations have been wrong in predicting the future of AI over and over again, regardless of the goodness of their intentions. As Elon Musk once said, ‘AI Doesn’t Need to Hate Us to Destroy Us.” There is no guarantee that our paths and decisions will lead us to the best shot we can get.

Fortunately, we as a society know the wise process of navigating the future even when we face enormous uncertainty. By respecting each other and building a society where anyone can freely express their opinions and discuss them, we have been able to continue to step forward while embracing mistakes. If there is one evil to be wary of in the development of AI, it would be the unbalanced power dynamic where someone can have an incomparable influence on the society, and someone will not even have the opportunity to use and develop AI — regardless of whether the power is good or evil.

AI Network manages computing resources through DAO (Decentralized Autonomous Organization), and in the spirit of Open Source, we are creating Open Resource where anyone can participate in the AI ​​ecosystem without worrying about resource restrictions. We can think of three attributes as the basic building blocks for managing AI as a community. The AI ​​component to be managed must be (1) accountable, (2) reproducible, and (3) valuable. AI Network makes this possible by using AINFT — an immutable identifier for the connectivity of AI components, AIN DAO, and AIN Blockchain — the blockchain for serverless computing.

(1) Make AI accountable

One of the difficulties in managing AI is that the pipeline is very long and it consists of many modules that behave like a single organism, so it is difficult to know where the managed unit starts and ends. For example, Google says that 2 billion containers are updated every week, and they are managed by one huge code base and one huge organization. Unlike projects in closed organizations, open-source projects are fragmented in different frameworks and development environments. To connect projects scattered on the Internet, a unique number is assigned using a non-fungible token (NFT) to identify the relationships between open AI components. Since it is managed on a blockchain, an immutable database on the Internet, we will be able to better understand how things are connected over a longer timeframe (probably longer than a typical lifetime of a company). AI Network proposes the following as the schema of the first AINFT.

{
"data": "https://github.com/kmh4500/AINFT-engine/blob/main/data/ain_train.txt",
"model": {
"name": "gpt2-en-large",
"epoch": 5,
"platform": "ainize.ai/teachable-nlp"
},
"service": {
"api": "ainize.ai/kmh4500/AINFT-engine",
"demo": "tiniverse.afan.ai/ain"
},
"state": "insight.ainetwork.ai/database/values/apps/chat/AIN"
}

This schema represents the entire pipeline constituting a service and consists of the basic structural units of data, models, and services that make up AI services. Each property may be expressed as another AINFT in a more detailed form, and it is also possible to combine different AINFTs to compose a larger unit of service. The last element, state, is blockchain storage which can be used by the service specified in AINFT to receive data from the Internet and record the corresponding results. In this example, the AI ​​Network blockchain manages the state of the AIN chat-bot service, and all conversations from users are recorded in the blockchain.

(2) Make AI reproducible

AI is considered reproducible if the machine learning workflow is reproduced as described and the results of AI services or research can be recreated. No matter how good a new service or research is if it is not reproducible, it is difficult to investigate and implement. Without reproducibility, it is hard to understand, and cannot easily develop into follow-up research. Because AI studies are increasingly relying on their own frameworks or systems, it is becoming a challenge to understand exactly what is described in a paper and may produce unintended results. In addition, even if the ML framework, machine specifications, and libraries used in the study are all publicly available, as AI grows larger, it becomes too costly to reconstruct. Ainize (ainize.ai) is a cloud service that uses GPUs provided by volunteers on AI Network to run containers managed by open source developers. Using this, developers can reproduce training and inference results at an extremely low price. Above all, since the connection of containers is managed by the multiple distributed blockchain nodes, the services will be up and running and AI results will be reproducible as long as they are maintained by at least one stakeholder on the internet.

(3) Make AI valuable

For AI developers, since the business has a very long value chain, it is difficult to come up with a budget for the several years of R&D and focus on the technology, and on top of that, it takes a considerable amount of time to confirm that the service created will work in the market. This is called the long-tail problem (link) of AI, which means that for intelligence to reach the market, it needs to first “navigate problems of complex coordination” and incorporate the real-world data. Predicting and controlling user behavior in the real world is much more of a hassle for AI services, and fast prototyping and iterating through MVPs don’t give enough insights in many cases, unlike in traditional software development. Using AINFT, a long AI pipeline can be divided into smaller components and can be defined within the manageable scope that even a single developer can contribute and collaborate on. Also, since AINFTs can be assembled, if more AINFTs are combined to create a larger value, even a small AINFT a developer makes can have a big impact overall, and appropriate rewards can be given.

The first AINFT

The first AINFT mentioned in this article can be seen on Opensea (link), and it can be confirmed that it is registered as a unique transaction on the blockchain network (link). It is trained with the text from the AI ​​Network white paper, and all data, code, and models are open source. The resources for training and running models have also been managed in the open resource ecosystem by AI Network blockchain. All communications of the AI are recorded on the blockchain and AI grows as the community interacts with it. The chatbot’s name is AIN, and you can have a chat with it at https://tiniverse.afan.ai/

To sum this up, let’s ask AIN about the role of community in the future of AI.

What is the role of community in the future of AI?

AIN: “Community will be a critical part of our future AI. community is building and maintaining the infrastructure of a world computer.”

AI Network is a blockchain-based platform and aims to innovate the AI development environment. It represents a global back-end infrastructure with millions of open source projects deployed live.

If you want to know more about us,

--

--

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
AI Network

AI Network

AI Network official account. Please contact me here. info@ainetwork.ai