Jiritsu origin… the early lite-paper. Verifiable AI
JIRITSU AI; TAMPER-PROOF AI FOR SMART CONTRACTS
AI is the future of all software. AI is not just a revolution in what software can do. It is a revolution in the way software is written. AI writes software with software!
“Until recently, humans programmed all software. AI uses data to write software. By automating the creation of software, AI could turbocharge every industry.”
This software revolution is anticipated to add $30 Trillion in market cap over the next 10–15 years.
Jiritsu’s mission is to enable a significant portion of this market cap to accrue to the blockchain ecosystem.
THE OPPORTUNITY — AI WILL UPGRADE SMART CONTRACTS
The blockchain economies of NFTs, DeFi, DOA, decentralized social networks, games, and enterprise logistics are trustless and automated, but the automation is for very simple tasks — they cannot as-yet do complex things and remain trustless.
AI-powered automation will take blockchain economies to the next level:
- AI recommenders can help NFTs realize their full potential as “e-commerce for everything.”
- Sophisticated decision systems in DeFi will create new economic primitives, such as complex liquidity structures within AMMs.
- Sentiment analysis will help DOAs achieve consensus faster and with more transparency.
- AI can enable a provably-fair gaming economy, via trustless simulation, non-player-characters and content generation.
- Algorithms with transparent bias will guarantee fairness in decentralized social networks.
- Enterprises will have immutable records of AI on-chain for public confidence and to comply with emerging regulations.
AI CREATES NEW VULNERABILITIES IN SMART CONTRACT SECURITY
Blockchain is the foundation of Web 3.0, handling the smart-contracts behind every transaction in the blockchain economy. These contracts are autonomous and can handle large-scale transactions because they are tamper-proof — also called trustless.
Any blockchain that integrates non-trustless data will eventually be robbed by a bad-actor who can manipulate a contract at will.
This problem is well known when introducing off-chain data into a smart contract. It has been solved by utilizing Oracles such as Chainlink. AI data, however, presents a new and more difficult problem.
Progress has been made regarding decentralized AI compute (e.g. Boba and Cortex Labs), marketplaces for experts and deep learning models, e.g. Fetch.ai and SingularityNET, as well as API to data for training AI, e.g. Ocean. However all of these are susceptible to exploit.
But AI is opaque by nature — a “black box” of dynamic algorithms with hidden bias that can be altered at whim.
As they are integrated into smart contracts, AI become an attack vector, exposing protocols, and their users’ assets to new risk.
The next generation of on-chain scandals and exploits will inevitably culminate in the corruption of AI systems.
EXISTING ORACLES DO NOT SOLVE THE TRUSTLESS AI PROBLEM
Currently, oracles enable tamper-proof access to stored data that is deterministic, i.e. it doesn’t change. In contrast AI data is not only dynamic, it is also being generated within the black box where there is no oversight.
What good is an oracle if the content of the black box can be changed without oversight?
To solve this problem the Oracle needs to get into the black box that is AI.
JIRITSU ENABLES TRUSTLESS AI BY UNPACKING THE BLACK BOX
Jiritsu unpacks the “black box” of AI algorithms — illuminating the decision system that is inside — making AI transparent, trustless and tamper-proof.
In addition, Jiritsu supports “AI inferencing” a type of query specific to AI (or any queries that perform complex computations), not available with conventional oracles.
A SHORT OVERVIEW OF HOW AI WORKS
AI is made up of algorithms designed to answer questions. This is known as a “function call”. To do this, AI does two things: analyzes and detects patterns in a data set, and makes recommendations based on these patterns. It’s that simple.
This is the formula behind every search engine, whether on the web, within an e-commerce site, or behind a company firewall — data is run through an algorithmic engine that looks for patterns and creates an index of ranked answers in response to a query. This output is called an “AI inference”.
This seemingly simple task presents two challenges to make the AI “trustless”:
First, the algorithms are inherently biased, because they’re built by humans. (This isn’t a judgment about the quality of the bias, just an acknowledgement that there is bias and that this bias must be understood and quantified in order to neutralize it.)
Second, we have no way of knowing if the inference is being changed or manipulated in some way. (In other words there is no way to audit the path from question to answer ensuring that today’s apple is not tomorrow’s orange.)
bXAI: BLOCKCHAIN EXPLAINABLE AI
Our approach is simple. We start by separating the AI into its two parts:
- off-chain compute and pattern recognition, which can be any AI developed and run by anybody anywhere, followed by
- on-chain inferencing and data traceability made via an index of answers
By using the blockchain for inferencing (B), and storing the questions and answers (i.e. algorithms and recommendations) on-chain, we make AI trustless.
bXAI achieves trustless status in three steps.
- Indexing and storing the AI on-chain, making it immutable.
- Recording an auditable ledger of each AI inference.
- Consensus mechanism to enable the independent assurance of AI, identifying critical risks such as bias.
The result is Trustless AI where algorithms are quantified in terms of programming bias, and inferences are consistent, immutable, and transparent.
THE OPENSEA NFT SCANDAL — THE ABSENCE OF OVERSIGHT
“We don’t make money when we sell things.
We make money when we help customers make purchase decisions.”
-Jeff Bezos
OpenSea, a leading NFT marketplace, provides a textbook example of the need for trustless AI to prevent bad actors from exploiting information asymmetry between buyers and sellers to create a “rigged” or unfair marketplace.
In the case of OpenSea, insiders with advance knowledge of the NFTs that were scheduled for promotion would quietly purchase the NFT, then sell it for a personal profit when the price was driven up by front page publicity. If insiders are able to profit unfairly in a closed market, how can users trust that the recommendations they receive are based on their interests and not the interests of the profiteer?
For a platform like OpenSea, a deep learning-based Recommender System (RecSys) would be a gamechanger, but such AI models are inherently opaque and vulnerable to the “black box” of manipulation and human bias.
In order to restore faith in the timing, targeting, and pricing of NFT recommendations and promotions, they need to be rendered “provably fair”.
With Jiritsu’s trustless AI engine, OpenSea’s recommendations are both transparent and auditable. The public can have confidence because independent, third-parties can certify the integrity of recommendations and test their relevance to buyers and trends.
In addition, the blockchain ensures that all buyers find out about the promoted item at the same time and can trust that no one has manipulated the recommender itself.
THE JIRITSU ROADMAP
Jiritsu’s road map addresses the next layers of trustless AI through a novel layer 1 blockchain and will be realized in three stages: queries, data management and hardware.
Jiritsu 1.0 → Making inferencing trustless
As outlined above, J1 solves the core AI oracle problem. We do this via an innovative framework for abstracting the core data structures that underpin the AI system utilizing a decentralized solution.
This allows users to generate inferences in much the same way as users request data from oracles i.e. queries are performed via “inference nodes” that function as consensus makers similar to other oracles. They are unique in their ability to run optimized AI querreis.
The data is indexed and written to chain rendering it a deterministic immutable source of AI.
Jiritsu 2.0 → Evidence of data-over-time
An important tool for AI quality assurance is the ability to examine a sequence of queries/responses.
AI is a living breathing provider of data and can get stale or drift, moving away from its original intended purpose. So for J2, we enable trustless quality-control and oversight to monitor the data flows underlying the AI service by building a “temporary immutable blockchain”.
This allows users, AI providers and regulators to conduct quality control by examining sequences of queries/responses, making it possible to detect if an AI needs to be refreshed or taken off-line.
We call this a Delta T Blockchain, for the time span 𝜟T of the cache — dTBC. This new type of blockchain acts as a cache and has the capacity to store chunks of data or a timeseries just long enough to run analytics. dTBC is very much like the Snapchat of blockchains, a data ghost.
This addresses a major challenge of cost associated with large data storage on standard immutable layer 1, most of which would become infinitely slow and infinitely expensive as huge quantities of data accumulate.
Jiritsu 3.0 → Securing the AI hardware
Deep learning lives on semiconductors and GPUs.
Our mission in J3 is to extend auditability to the full-stack, hardware and software layers, by enabling trustless monitoring of computer processing and making the AI semiconductors and GPUs tamperproof.
BLOCKCHAIN EXPLAINABLE AI — bXAI BY JIRITSUAI
Jiritsu enables trustless AI.
AI is the future of software.
You can’t have a fully autonomous blockchain without AI.
You can’t have trustless AI on the blockchain without Jiritsu.
Jiritsu’s solution is bXAI.
###
