Introduction
AIaaS (Artificial Intelligence as a Service) has grown as more companies adopt AI to solve business problems. From USD 6.68 billion in 2023, the market is estimated to reach USD 28.58 billion in 2025. AI models require datasets and machine learning techniques, which AIaaS supplies. Google's Prediction API and Amazon ML are becoming increasingly relied upon, raising concerns. Given the infrastructure, getting high-quality datasets and cutting-edge algorithms is hard. Algorithm and dataset owners may be hesitant to share. They charge extra to protect an asset or privacy in a cloud environment. If a malevolent provider penetrates the computing service during training, assets' confidentiality may be jeopardized. Finally, there are so many datasets and methodologies that such vendors are hard to identify. Academic research has partially solved these issues by creating blockchain-enabled data markets and monitoring blockchain-based cloud computing for off-chain processing. However, a decentralized means for people to buy the model they desire has not yet been developed. Anaphora First, AI leverages blockchain to create a trusted marketplace where datasets, ML algorithms, and cloud computing resources can be auctioned. After that, they employ Trusted Execution Environments to secure ML infrastructure code and data.
Anaphora's AI-as-a-Service Market
A marketplace for artificial intelligence as a service built on the blockchain. Clients that are looking to acquire AI models typically hold auctions where suppliers may submit bids. Providers may be broken down into three groups: those who supply data sets (Data Providers; DP), those who supply cutting-edge machine learning algorithms (Algorithm Providers; AP), and those who supply cloud resources (Infrastructure Providers; IP) for the training phase. The client's needs and associated requirements (such as model accuracy) are outlined in the auction. It will be possible for each service provider to place bids in their respective service areas. Therefore, a total of three winners will be announced, one in each respective category. The creation of the necessary model will require their combined efforts. An initial step for the AP will be to prepare a suitable computer environment. Then, through a protected connection, the DP and AP will transfer their assets to the IP. When the IP has finished its calculations, it will send the resulting model back to the user. Those who submit competitive bids will be rewarded.
• Phase of Semantic Matching
Information on what kind of data, methods, and infrastructure the customer needs is included in the request. Our goal is to create a system that can adapt to any given use case, hence we need to be able to handle a broad variety of use cases. An ontology-based data storage and retrieval infrastructure is used for this purpose. Client needs and provider resources are described in a shared ontology, allowing for efficient matchmaking to take place either on the blockchain, via the emission of an event directed at providers with the necessary assets, or off the blockchain, via the continuous monitoring of new client requests made public on the ledger. Providers can use the asset ontology to determine if they have assets that satisfy the client-published standards; if so, they can submit solution/service offers during the auction phase.
• Phase of Auction
The auction phase satisfies the client. AI providers bid on the auction smart contract. The client locks the reserve price of ANAPH tokens, the Anaphora AI ecosystem's token, along with its solution specs. The auction ends once the client adjudicates the auction contract or a specific duration. Each vendor can bid on a semantically matching AI solution. A provider's base estimate for a solution or service specification. A lower solution bid for the same solution request updates it. To foster competition, changing the basic price can decrease solution bids. After the auction, the winning triplet of semantically matching lowest-price bids is broadcast on the blockchain using an Auction Service Level Agreement (SLA). Due to data privacy and algorithm confidentiality, the platform cannot review datasets and algorithms before machine learning. Without results, consumers may be unwilling to pay for the system. A supplier can reference earlier auctions with semantically equivalent bids to avoid this issue. If a client previously supplied a dataset, the provider might include the Auction instance address in its bid. Auction specifications allow clients to set bid reference requirements. Bids without Auction SLA references should be cheaper.
• Secure Learning Phase
After identifying all providers, the DP and AP's intangible assets must be sent to the IP infrastructure for machine learning computation. Protecting sensitive information and intellectual property from a hostile or compromised infrastructure provider requires a safe and secure educational setting. Machine learning and model creation in a trusted execution environment (TEE) like Intel SGX enclaves protects sensitive data. Since the infrastructure provider cannot access TEE data, the data provider receives an attestation that the environment is safe and up-to-date, and a secure communication channel is formed for data upload. This technology makes learning more reliable. TEE and Linux containers can protect the AP's algorithm development IP. The secure route uploads AP's algorithms to a protected registry. The technique is protected against container image signature attacks when executed in Linux containers.
• Phase of Compensation
The completed model is then sent to the client over the TEE's encrypted channel after the learning process has concluded. After evaluating the model's output off-chain, the client then makes note of the fact in the SLA contract for the auction, bringing the process to an end, releasing payments to the service providers and allowing the client to reclaim any unspent funds from the reserve. The following part will show an early version of the market being implemented, leaving the protected learning space as an open issue for further development. The central feature of our proposal is the blockchain-based marketplace, and this section provides a proof of concept for this feature. It begins with the proposal of a blockchain marketplace implementation in Solidity. The total price tag of the fix is then calculated by doing a cost analysis.
Implementation
This article implements the market on Ethereum to assess the contribution. Github has the code. Figure 2 shows market infrastructure.
Factory and Auction smart contracts support this market. Factory creates Auction instances. Any third party with a thorough description of what they require can start an auction once it's on the blockchain. Off-chain IPFS ontology files specify anticipated needs. This document describes how to submit auction specifications. Auction state variables include requirements, client address, and auction settings. The instantiated Factory contract is also recorded. This allows easy adjustments to Factory contract and requirement information without losing the link to earlier Auction instances. Bidding depends on auction length. Before executing a contract method, the auction's permitted time is verified. If so, the contract will choose the auction winner without human interaction. Participating in the auction as a supplier requires offering the desired assets and semantically matching the client's initial request. Providers can submit as many bids as they like as long as the auction is ongoing and the new offer sum is lower. The contract also limits service providers to a maximum bid value. Three winners are revealed after the verdict. A service provider that specializes in data, algorithms, and supporting infrastructure wins.
The Problem
Many businesses aim to employ AI to enhance. Consumers expect AI services to make their lives easier. AI is expensive and difficult. AI development is costly. Programmers, AI, R&D, and more are expensive. Hiring specialised expertise and developers is likewise pricey.
The Solution
Anaphora AI lets these firms pay for AI products and services as they utilize them. The service provider may monetize their goods or services, while the firm using them can avoid expensive AI solution development. No machinery, programmers, R&D, or other costs. Anaphora AI provides cloud-based marketplace and AI provider solutions. Anaphora AI delivers a cheaper, more efficient AI solution for businesses and organizations. Anaphora AI democratizes AI products and services for companies and consumers. Our marketplace lets AI developers sell their solutions.
Roadmap 2023
Conclusion
Anaphora AI's goal is to give AI service and solution providers a non-exclusive way to monetize their offerings while also making it easier for businesses, organizations, and consumers to acquire and utilize AI tools and solutions. By enabling AI developers to publish and sell their AI solutions in our marketplace, Anaphora AI is democratizing AI solutions and services, making them more accessible to companies and consumers.
For More Information Visit:
WEBSITE: https://www.anaphora.ai/
TWITTER: https://twitter.com/AnaphoraAI
TELEGRAM: https://t.me/anaphoraai
BITCOINTALK THREAD: https://bitcointalk.org/index.php?topic=5457001.0
WHITE PAPER: https://t.ly/Kry77
DISCORD: https://discord.com/invite/Kw2KbzS47Y
YOUTUBE: https://www.youtube.com/@AnaphoraAI
REDDIT: https://www.reddit.com/r/AnaphoraAI
INSTAGRAM: https://www.instagram.com/anaphoraai/
MEDIUM: https://medium.com/@AnaphoraAI
CONTRACT ADDRESS: https://etherscan.io/token/0xc58E3585C05e24bA3370e06e81c11e207F684289
GITHUB: https://github.com/AnaphoraAI
Bitcointalk Username: oprahwindfury
ERC-20 Address: 0xdE865D50CF631FEB7176D31AEc8Da115d05Bea7b