В частности, zk сегодня активно применяется для работы с машинным обучением: для того чтобы использовать модели без доступа к изначальному сету данных для тренировки (я могу быть уверенным что для тренировки использовались финансовые данные компании, но не получу доступ к самим данным), либо чтобы исполнять определенные модели и иметь гарантию что были исполнены именно они без доступа к underlying данным — весам, входным данным. Либо для того чтобы построить более сложные системы, в которых выборочно разглашаются некоторые из этих данных.

В этом посте хочу разобрать несколько наиболее популярных сегодня юзкейсов для zkml систем. Каждый пункт звучит так, что там легко найдется место для нескольких компаний или протоколов по $1 млрд каждая.

1. Децентрализованный маркетплейс промтов для генеративного AI — вы можете сделать классный промтп и продавать к нему доступ (или к результату работы нейронки), не раскрывая промпт.

2. Децентрализованный Kaggle — конкурсы ML-алгоритмов, где проверка работ происходит децентрализованно и без доступа к самой модели. Публикуете модель, децентрализованно проверяется ее точность (accuracy). Если вы победили, смарт-контракт автоматом начисляет вам награду. Никто в процессе не имеет доступа к весам вашей модели.

3. Использование сложных трейдинговых ML алгоритмов или zk classifier/regressor. Может использовано для создания приватных и децентрализованных систем стратегий управления деньгами типа Yearn. Вот пример ончейн трейдингового ML-бота, работающего на zk.

4. Идентификация, особенно биометрическая, которая не требует разшлашения данных в промежуточных этапах обработки. Например, то что делает Worldcoin.

5. Децентрализованный маркетплейс AI моделей и совместная работа по из разработке.

6. Использование AI в смарт-контрактах с децентрализованной гарантией исполнения именно той модели и именно с тем инпутом, который нужен пользователю.

7. AI NPC / AI игровые миры для web3 игр. Я могу доказать какая именно модель или инпут использовался для генерации мира в метаверсе, не раскрывая весов и сорс-кода. Тут прям моя 💚

8. MLaaS / Оракулы для inference. Создаете модель, публикуете в облако и можете быть уверены что в процессе inference получены результаты работы именно той модели, которую вы хотели использовать, и что эти результаты не были изменены.

9. Децентрализованный аудит нейронных сетей или ML-алгоритмов. Например, через zk можно доказать при тренировке использовались или не использовались определённые данные (приватные, копирайт, коммерческая тайна).

10. Аудит смарт-контрактов и выявление аномалий. Доказательство что, например, смарт-контракт может быть эксплуатирован (оттуда можно вывести деньги) без разглашения самого эксплоита или ML-алгоритма, который нашел уязвимость. Сам zk-пруф может быть основанием чтобы остановить контракт до выяснения и устранения проблем, что потенциально может сэкономить миллиарды на взломах.

11. Inference очень чувствительной информации, такой как медицинская диагностика.

12. И, для тех кто уже начал надеяться что я вас просто так отпущу, — конечно, приватный скоринг репутационных (aka вообще любых персональных) данных человека, организации, робота, контракта с помощью ML алгоритма.

Если вы, как и я, все еще не понимаете как эта чертова магия работает, то приглашаю пройти офигенный и бесплатный курс: https://zkiap.com

P.S. Картинка из midjourney по запросу “zero-knowledge”

How is the metaverse economy different from the real one? By programmability and composability.

The effect is enormous: we now have an endless (latent) space for the design of various kinds of tools for bargaining. As an interface for an online game, where you can throw gold coins and things to the seller until they agree. And if you have a positive reputation with his faction, then you will need less coins and things. Actually, this is exactly how trading works in the real world. It’s just that before the crypt, we didn’t have the tools to digitize it and execute it in a decentralized but provable and verifiable environment.

Let’s imagine a few scenarios:

1. How-To DAO is a subscription-based digital university. I can buy a subscription just with tokens, or get a discount by providing proof of my reputation in the form of NFT from the CryptoEssay channel, POAP showing that I am a member of the TheGraph protocol board and SBT issued for completing the _buildspace course. So I paid for a year's access to HTD's community of experts and mentors and educational programs for building web3 businesses, but received a substantial discount (just like a scholarship or university rebate for my achievements) because I had demonstrated that I was active in this area.

2. I came to Cyprus for 6 months and I want to rent a house near Limassol. This is a short term rental and I don't have the cash to pay for the whole term at once. But I can offer the landlord my digital credit reputation (SBT + VCs), as well as sign a smart contract that would automatically (trustlessly!!) transfer a share of my revenue share of the contract. RevShare contract (Shurik Agapitov's idea) - this is possible when, as an influencer, I made a promotion for a new mobile toy, and the game developers in return closed a contract according to which 0.5% of the game's total revenue over the next 18 months automatically falls to my address.

3. I started a food delivery business in Tel Aviv. As a new project, I need marketing and attracting new customers, so I'm airdropping 50% discounts on the entire menu. Only customers who are power-users (power-devourers) of online delivery services can receive such a drop. To receive, you need to log in through Uber, Glovo or Yango, where you have collectively made more than 100 delivery orders over the past year.

An economy requires trust to operate effectively. Trust between countries, banks, companies, people and even two IoT devices or smart contracts.

Historically, trust is provided either by the state (healthcare, legal, education systems) or platforms (Uber, Airbnb, Steam, Amazon, Booking). The state is conservative, slow, inefficient. Platforms strive for monopoly, user lock-in, profit maximization.

A more efficient alternative is web3, decentralized identity/reputation.

From the application-centric world — the one where my music on the Spotify server, my medical history on the Kaiser server, my transactions on the JPMorgan server, my games on the Steam server, my resume on the LinkedIn server — we are moving to a user-centric model. Any external service writes or requests read permissions to/from a personal data pod that the user controls.

The protocol and marketplace of the scoring models ensure trust (so that there is no Chinese social rating or one-model-fits-all problems). A scoring model is an algorithm that assembles a summary of reputation in a specific context from thousands of individual data points. Even for a credit score, there can be a hundred different models. These models will be gamed and they will evolve to counteract.

There are three types of data sources: self-reported data about yourself; data from peers, other agents; data from the authority, e.g. state, network consensus, provable fact.

Today there are hundreds, and tomorrow we will see thousands of such products.

Bureaucracy: portable trusted data between jurisdictions, programmable bureaucracy. "If you have flown 100k miles, a visa to the UK is automatic"

Loyalty: all shops, restaurants, hotels write order history to your data store, you can share this data with others to get more personalization, discounts or payments for providing personal data from the advertisers

Social: all your posts, photos and videos are in your data store. For each piece of content, you make privacy settings. 1,000s of frontends and apps that generate feeds.

Professional: all my projects, work experience, public speaking, commits in the DAO, customer reviews, education are in my data store in an authenticated form. For a bank, employer or investor, I prepare a summary of key facts.

Blue checkmark: open and transparent profile validation rules, which are then reusable on any platform.

All these products need a protocol and it is the same one in terms of functionality between them. But a protocol is not a corporation. For-profit protocol will either become the next BigCorp, or immediately lose. The one which is most similar to Bitcoin will win: no CEO, no owners, no investors, a public good.

Building institutions is hard. Harder than companies. But more important for humanity.

None of the above require a blockchain. Except when we talk about financial transactions.

1. Subjective. Any reputation is a reflection of social weight and the degree of trust from the verifier to the owner. That is, by definition, subjective for the verifier (a person, a group of people or an algorithm).

2. Reflexive. The issuer reputation is just as important as your own. The fact that you give a credential to someone is just as important as the fact that you get issued one.

3. Domain-specific. There is no “universal” reputation. There is a reputation of a taxi driver in Uber, a taxi driver in Lyft and a truck driver in the traffic police system.

4. Decentralized. There is no single agent that would have sole control and influence over reputation scoring rules or algorithms.

5. Gameable. Reputation, like security, has no final point. It is an ongoing process in which attackers find new attack vectors and maintainers keep the system stable.

6. Private. Nobody but you has access to your reputation data by default. You can share this data with your employer/app/partner/government. You may not share all, but only part of this data (selective disclosure). You can share no data — just a cryptographic proof that your reputation is above the X threshold (zero-knowledge).

7. Portable. You can build reputation in the same domain through different applications and systems that follow a common standard (think verifiable credentials, DIDs, soulbound tokens). This also implies a shared and publicly edited taxonomy.

8. Pseudonymous. Can accumulate value over time, but does not require instant doxxing. Sybil resistant.

9. Time-dependent. Reputation is constantly evolving over time. Recent actions are more valuable. Lack of activity leads to reduction of the score over time.

10. Ultimately, it is an economic mechanism. Social capital and financial capital are two tools available to economic systems for programming incentive-compatible mechanisms.

DAO is a new organizational form that allows coordination of resources through decentralized credibly neutral mechanisms. DAOs generate public goods similarly to traditional organization such as corporations, NGOs, governments, unions, cooperatives. But DAO is a superior organizational form that allows coordination with less overhead and better alignment of incentives. This is why some of the most amazing products we see today in the world are coming from the DAOs.

The Mission

DAO exists to coordinate people around the common goal, such as building commercial product, funding and delivering public goods, governing a community or allocating resources.

DAO is a rule-based, distributed, self-improving mechanism. It has a well-defined public process of changing its codebase. It has no single point of failure or control. No person is in charge. The codebase of a DAO can be improved using its own governance methods.

DAO starts with a culture. It takes famous Peter Drucker saying "Culture eats strategy for breakfast" to the extreme.

In pre-Internet era people congregated around common interests (church, corporation, school). Today we can work with anyone globally 24/7 on what matters the most to us. It's not about stable income anymore — it's about professional fulfilment, like-minded community and shared culture. Notice how you feel much more involved with people online who you never met than with your neighbours.

Goal of the corporation is to create profits to its owners, and culture only comes afterward to support this goal. Customers tolerate their data being sold to advertisers, paywalls, monitoring of their online behavior by the 3rd parties for the sake of using a product they need.

Goal of most of the DAOs is to build protocols by aligning interests of contributors, users and investors. All three roles become the same. DAO will only loose if it provides excessive returns to the investors (because it comes at a cost of customer satisfaction and risk of being disrupted or just forked. don’t forget most DAOs build open source).

The Structure

Typically DAOs have project leaders or "context managers" and a long tail of contributors that can have very varying level of commitment. Leads usually are working full time and have invested time into building context and relationships across the organization. Their primary goal is provide clarity and build infrastructure for contributors to do meaningful work that leads to the common goal.

Correctly designed onboarding process allows anyone to become a contributor and, given enough time and commitment, a leader within the DAO.

Instead of hierarchical management structure DAOs have emergent leadership where the whole system is designed about empowering people to step up and the lead part of the organization.

DAOs are "fractal" organizational. DAO is not a monolithic organization. DAOs typically have lightweight governance tool that controls treasury and key parameters such as multisig signers.

Then most DAOs have multiple workstreams:

• Engineering

• Product

• Community

• People operations

• Compensation

• Marketing

• etc.

Projects exist within each workstream. Projects are split into tasks that require a champion / leader who ensures completion and keeps the team accountable. Working groups have ultimate accountability for delivering the project. Web3 organizational tool stack will accommodate fluid participation and ownership.

DAOs can choose to have legal wrappers for compliance and payroll purposes. Many tools help with this: Korporatio, Fairmint.co, Etherize, Otonomous, Otoco

The People

DAOs are democratizing employment. 100 years ago people were looking for stable income, this is why corporations were created. Today people are looking for ways to be empowered to do most meaningful work.

DAOs are democratizing education. It allows to merge education and work, get paid while learning new skills and participate in creation of real products in a public, open environment.

DAOs have the lowest cost to coordinate people globally. With much lower management overhead and legal costs, DAOs are able to coordinate people from around the world. They are natively digital, global, available 24/7.

DAO's function is to empower individual contribution. Ultimately, DAO structure exists to help people lead and achieve results. DAO solves compensation: it enables anyone to become a founder, a shareholder, a board member. It is a truly meritocratic system.

DAOs employ graduated sanctions. Accountability in DAOs are similar to that of the smaller communities or commons. It can be quite strict but its goal is to help communities self-regulate rather than purely assert power.

Pseudonymity enables True Meritocracy. People don't know anything about you (age, nationality, color, name) expect for your contribution to the DAO. This enables real equality and removes discrimination.

DAOs are the workplace of the future. You won’t find 9-to-5 work schedule here. Everyone knows how much others are paid (since all transactions are done in crypto and on the blockchain). Decisions are made publicly and there’s strong pressure to avoid behind-the-curtains politics.

Fundamentally, DAO is about ownership. DAOs give participants significant equity, power and ability to manifest the best of their potential, not just responsibilities as it is happening in corporations.

Obviously, power can and will be misused (as it is in corps and govs today). DAOs, unlike centralized organizations, give participants autonomy to coordinate or withdraw at any moment. (to learn more read on "rage quit" in MolochDAO). As with any public blockchain, DAO participation:

1. is based on self-interest, voluntary permisionless participation;

2. is steered into the future using incentives compatibility (mechanism design and game theoretic models generally);

3. its value and previous decisions are protected by public key cryptography.

The Value Creation

DAOs create their own monetary policies. Once DAO tokens are issued they are immediately publicly traded. However, unlike a public company, DAO has full control over the monetary system: it can issue currency, use it to create value and spread this value back to its contributors.

Note that today governments are the top-1 funders of public goods. They typically issue money through debt to fund creation of public goods and expect future revenues to cover the investments. DAOs replicate this model but in a much more granular and efficient manner.

DAOs can have a business model but don't have to. Some DAOs are focused on achieving certain public benefits, others choose to maximize returns to their investors in a form of token appreciation or dividends. As with many traditional organizations, the right approach is probably in the middle where organizations can generate a positive impact while ensuring sustainability and growth.

Top-20 DAOs have ~$13B in their treasuries. This is not counting TVL of their products.

DAOs create open source products or protocols. Thanks to composability of OSS and web3 protocols, nature of competition in DAOs is different. You don’t compete by building exact stack of your rivals and try to extract more value from customers. Instead, you can build on top existing protocols. For example, when building decentralized social network you focus on one aspect (e.g. social graph) and leverage the rest of the infrastructure (e.g. data storage, decentralized identity, wallets).

Most DAO business models are token-based and employ circular crypto economy. Tokens give ability to programmatically split revenues between dozens of protocols. Costs of DAO-to-DAO contracts are negligible: they are automated, don’t have compliance or transaction costs, are always and forever auditable, permissionless (I don’t need your permission to use and pay for using your protocol).

DAOs are inevitable. Decentralized products can’t be built by the centralized corporations as products reflect organizational structure.

Types of DAOs

Areas where we use traditional corporations are just a subset of what a DAO can do. Let's examine concrete examples of the successful DAOs. For more detailed overview you can read this great article.

Ephemeral DAO

"Emergent, ephemeral swarms ready to tackle one problem and dissipate" [src]

DAOs can persist in time or can be assembled to just complete one task, such as organizing an event, launching a protocol, investing in a product or piece of art, writing a book or even buying an NBA team.

Social DAOs

Many DAOs exist in a form of a social club. This is a closed community of like-minded individuals where you need to own an NFT or a token to become part of it.

For example, Collab.land allows communities to restrict discord/telegram access to members who hold a certain amount of a particular token

Service DAOs

There's DAO for any group of professionals and freelancers, from designers and coders to lawyers. DAOs can specialize on a very specific task such as user research or code audit.

Governance DAOs

Governance tools allow communities to influence operations and budget via proposals and voting: Boardroom, Tally, Snapshot, Orca, Policy Kit, MetaGov, Paladin

Grants DAOs

One of the first actually scalable use cases for DAOs was giving grants to fund OSS development, contribution to existing protocols and other non-for-profit work. Examples include DAOHaus, Gitcoin, Meta Gamma Delta, DxD.

Investment DAOs

Metacartel, LAO and many other DAOs operate as a for-profit investment vehicles that allow to pool resources and invest into (mostly web3) projects.

Collectors DAOs

Curate, pool resources, purchase and hold physical collector items and NFTs.

Education DAOs

Some DAOs are exclusively focused on education, which can be free, paid or learn2earn.

DAO Operating System

As DAOs are growing, there's a need for tooling that simplifies spinning up and management of a DAO. Today we see dozens of services to help with this.

Protocol DAOs

Majority of DAOs today are only used to govern the development of a protocol through voting. This somewhat limits the functionality but it already has proven to be very effective in community-led decision making for multi-milion dollar deals and initiatives.

A lot of ideas here are inspired by the speakers at #MCON2021

If you’re looking to join a DAO — be it a silent observer or active paid contributor — you can stop by Deep Skills Discord channel and say hello. We would love to provide more guidance :)