AI Currency: How to combine Bitcoin micropayments with AI?

Bitcoins are mine, sats belong to AI.

Written by: Fat Garage

Sometimes I find it amazing. When mentioning Bitcoin in different environments, I will always encounter some people who express their skepticism about Bitcoin because they “don’t (want) to understand” and “have impure motives.” Their points are mostly about “time” and “value”, not about the exquisitely designed thing/technology itself.

I’m not skeptical, and Bitcoin is an irreplaceable source of novelty for me.

• The recent novelty comes from the lightning experience of continuously sending satoshis into the wallet from the other side of the ocean. “A value as small as 1 satoshi (1 satoshi = $0.00023) can also be sent through the network cable of immeasurable distance. in your hand”;
• From surprise discoveries in street restaurants in Paris and Bologna, lightning ⚡️ paid for banana ice cream balls and bolognese spaghetti;
• It comes from the small general-purpose hardware sent to me by Teacher Ajian from Yuanyang. I can assemble a happy seed signer for Bitcoin by myself;

I like what A Jian said: **Bitcoin is doing things that other technologies simply cannot do. **

Today I want to talk about some topics that have been in my notes for 3 months:

**The origin of micropayments, attempts on the Lightning Network and the possible combination of micropayments and AI. **

A brief history of micropayments

W3C

Ted Nelson coined the term micropayment back in the 1960s Ted Nelson, the founder of the Internet, proposed the concept of micropayment in the 1960s.

In 1992, Tim Berners-Lee, the creator of HTTP and HTML, released his second version of HTTP and the first reference to the now commonly used status codes. Among them is a code that Berners-Lee and others believe will one day be used to pay for digital content: 402 payment required. Unfortunately, this status code is officially “reserved for future use” because from the beginning, various attempts to make micropayments on the network failed to materialize. More than 30 years after the invention of the Internet, we’re still waiting for one of its main original visions to be realized.

Tim Berners-Lee founded the World Wide Web Consortium (W3C) in 1994 to guide the development of the web, and micropayments were a major consideration from the beginning.

In 1995, Phillip Hallam-Baker, who had written numerous RFCs about Internet security, drafted the Micropayments Transport Protocol (MPTP) [1] , but the agreement appears to have never been implemented. It provides many insights into the nature of micropayments that are as relevant today as they were when the Internet was founded:

There is a large interest in payment which support charging relatively small amounts for a unit of information. Here the speed and cost of processing payments are critical factors in assessing a schemes usability Fast user response is essential if the user is to be encouraged to make a large number of purchases.

However, a key limitation of MPTP is that the protocol explicitly requires a third party (called a broker). At the time, digital payments could not be made without a trusted intermediary, so any attempt at a micropayments protocol had to allow for some kind of escrow of funds.

The W3C continued to promote micropayments for some time, publishing an Overview of Micropayments in 1998 and recommending MPTP as a practical approach, stating:

Micropayments have to be suitable for the sale of non-tangible goods over the Internet […] With the rising importance of intangible (e.g. information) goods in global economies and their instantaneous delivery at negligible cost, “conventional” payment methods tend to be more expensive than the actual product.

This echoes Hallam-Baker’s second major concern, namely transaction costs arising from the technical or administrative costs of available payment mechanisms. His first concern is the need for “fast user response,” which is often overlooked in discussions about the feasibility of micropayments.

Until 1999, Nick Szabo in his paper micropayment and mental transaction cost [3] Continue to think deeply about “fast user response”. It is highly recommended to read Szabo’s paper. He proposed that micropayment is not only a technical exercise, but also about the existence of psychological transaction costs (cognitive costs) - the psychological payment cost of micropayment is far greater than the technical cost. How to understand the decision-making process of applying** micropayments? **Under the premise that technical transaction costs will continue to decline, how to design an interactive payment process to reduce “psychological transaction costs”? One possible scenario is that an individual’s resources/capital “automatically” match her tacit preferences. (Use micropayments to wrap APIs, or “internet connections”)

A network based on micropayments means frequent payments, which means decision fatigue. For most small payments, the psychological transaction costs of having to continually choose to buy can exceed the value of the item they are paying for.

Big companies like Compaq and IBM, as well as startups like Pay2See, Millicent, iPin, and others, tried early on to reduce the technical and psychological transaction costs of micropayments, but it was still believed that the concept was here to stay. Characteristic from the start.

Perhaps the most famous of these companies is DigiCash, led by David Chaum, and they will have a lasting impact on the Bitcoin community. Chaum had already formally proposed many ideas for blockchain-like data structures and secure digital cash in 1982, and then founded DigiCash in 1989. DigiCash implements Chaum’s proposal to allow users to withdraw funds from banks (called eCash) and make digital micropayments untraceable. Unfortunately, only one bank implemented eCash, and the company went bankrupt in 1998.

Around the same time, other micropayments initiatives also disbanded, and the W3C itself ended support for micropayments in 1998.

The dot-com bubble is in full swing, and micropayments are one of the ideas that’s crashing the most. This is a great time to be a critic. Author Clay Shirky wrote “The Case Against Micropayments,” in which he boldly declared:

Micropayment s have not failed because of poor implementation; they have failed because they are a bad idea. Furthermore, since their weakness is ic, they will continue to fail in the future.

In 2000, his main argument for its fundamental flaws was not technology or infrastructure, but echoing Nick Szabo a year earlier: decision fatigue. He continued:

In particular, users want predictable and simple pricing. Micropayments, meanwhile, waste the users’ mental effort in order to conserve cheap resources, by creating many tiny, unpredictable transactions. Micropayments thus create in the mind of the user both anxiety and confusion, characteristics that users have not heretofore been known to actively seek out.

Shirky goes on to predict that three payment methods will dominate the web and will not suffer from decision fatigue issues: aggregation (bundling low-value stuff into a single high-value transaction), subscriptions, and subsidies (let others do it) It’s not that users pay for content – today this is expressed as an advertising model).

By the time the dot-com bubble burst, Shirkey’s predictions looked even more prominent. The infrastructure cost of credit cards has resulted in payments below $1, so it has become the de facto payment method, while enthusiasm for micropayments projects is losing steam. The self-evident and exciting future of the web dimmed against the backdrop of its increasingly centralized, surveillance- and advertising-driven predecessor, Web 2.0.

Bitcoin and Decentralized Networks

We have to trust them to protect our privacy and trust them not to let identity thieves access our accounts. Huge administrative costs make micropayments impossible. — Satoshi Nakamoto

The idea driving 402 is that it’s obvious that support for payments should be a first-class concept on the web, and it’s obvious that a lot of direct commerce should be happening on the web […] In fact, what emerges is that the single dominant business model is advertising . This leads to a lot of centralization because the cost per click is the highest and the platforms are the largest.

— John Collison, President, Stripe

Satoshi Nakamoto released the Bitcoin white paper in late 2008, coinciding with the U.S. housing crisis. Soon after he released its original code. Bitcoin was a huge breakthrough in the history of computer science and money, and sparked a new wave of interest in the possibilities of the Internet. For the first time there is a permissionless way to transfer value with the internet’s native currency without all the inelegant, bloated infrastructure required by credit cards.

For a while, the price of Bitcoin was so low that some actually advocated using it for a micropayment system, although Satoshi admitted that this was not (yet) a good solution to the problem:

Bitcoin is currently not practical for very small micropayments. It does not apply to content that is pay-per-search or pay-per-page view without an aggregation mechanism, nor does it apply to content that requires a fee of less than 0.01.

But the limitations imposed by cost haven’t stopped people from dreaming about the new possibilities it would bring. Marc Andreessen, the creator of the first popular web browser, gave the example of monetizing content and fighting spam:

One of the reasons it’s difficult for media companies like newspapers to charge for content is that they need to either charge it all (pay the full subscription fee for all content) or none (which results in those horrible banner ads all over the web). Suddenly, with Bitcoin, there was a financially viable way to charge an arbitrarily small amount per article, per section, per hour, per video playback, per archive access, or per news alert. cost.

Of course, this statement isn’t true today (at least as far as Layer 1 is concerned), but fees in 2014 were low enough that it could actually be built around the concept of micropayments. An interesting project built around this time was Bitmonet, which allowed users to choose a subscription level and pay just 10 cents for an article, 15 cents for an hour of unlimited access to the website, or 20 cents for a Day pass. Unfortunately, transaction fees are no longer low enough to allow for arbitrarily small micropayments, and although Satoshi Nakamoto clearly had this problem in mind from the inception of Bitcoin, it was not designed specifically to solve the micropayments problem. Designed.

Shirky’s predictions about content monetization are spot on, especially when it comes to subscription and advertising models.

In the advertising model, content is subsidized by advertisers (usually through third parties). From 2014 to 2022, Google and Facebook essentially held a duopoly in the online advertising market, serving as third-party mediators between advertisers and content creators. These two companies (and indeed most big tech companies) collect vast amounts of personal information and simply ask users to trust them with the security of their data, despite numerous breaches. This information is used to show targeted ads for products that people are more likely to buy. Companies often refer to this model as “free with ads.” But in reality, users do pay a price. The advertising model forces users to exchange content for two things:

1. User data is forced to be given to third parties, which, as Nick Szabo said, is a security breach.

2. User attention. The more time users spend on ad sites, the more money advertisers, ad platforms and content creators make. Creators are therefore financially incentivized to show as many ads as possible without annoying users so much that they leave the platform. The currency of the “free advertising” network is the user’s attention. You are the product. The advertising model clearly shows that consumers have become second-class citizens. Because of the abstraction layer between the creator’s revenue and the end user, creating a good user experience is not the highest priority. As more consumers use ad blockers, content creators are forced to serve ads more aggressively, making the web experience worse for everyone.

Subscriptions are also growing in popularity. Users say they would rather pay regularly for bulk access to licensed content such as movies and music than to own individual songs. While this is a more honest business model, it can also be very problematic when they become the only payment option. As competition from these services has grown in recent years, more and more people have found themselves suffering from subscription fatigue. Not being able to access a specific news article (or a few) at any given time forces us to make suboptimal choices, try to pay in bulk, and optimize the bulk of a given subscription.

Take streaming services, for example. Today, there are so many streaming services competing for content licenses that users end up paying multiple subscription fees to catch more movies and TV shows they want. But all they really want is to watch a small portion of what any given service has to offer. When they choose a service for a movie or show they want, the service often doesn’t stick around for long and jumps unpredictably from company to company as licenses expire and renew.

News articles are another example. Companies like The New York Times or The Economist attract readers by allowing them to read only a few seconds of an article before blocking the content through subscription paywalls. This is even more true for newspapers than for movies, where customers are more likely to be willing to pay a small fee for a single article of their choice rather than a package deal for articles they don’t want.

While subscriptions offer a more direct approach than advertising, using them in practice often results in an increasingly costly and stressful management game.

When Clay Shirky wrote about the problem of psychological transaction costs, he was writing before the psychological costs of subscriptions and advertising began to weigh on people as much as they do today. Bitcoin provided a solution to the problems of the internet’s native currency, but slow processing speeds and high fees quickly became a prohibitive issue for systems supporting micropayments. Before micropayment technology can truly take off, a major innovation is needed.

Lightning Network

In the Lightning Network white paper, the idea of micropayment appears somewhere in the C position.

“A decentralized is proposed whereby transactions are sent over a network of micropayment channels(aka, payment channels or transaction channels)” ——Lightning White paper

Current lightning micro payment projects

Currency of AI

Just like humans need passports and currency to cross national borders, AI agents may need some form of authentication and payment mechanisms to use different services and resources on the Internet.

From 402 error to L402

What is HTTP error code? 200 OK, 404 NOT FOUND, and 500 INTERNAL SERVER ERROR are all included.

• 4xx is a client error, indicating that an invalid request was sent to the client. The most common ones are
• * 401 Unauthorized: Authentication required or authentication failed.
• * 403 Forbidden: The request was rejected by the server.
• * 404 Not Found: The requested resource was not found.
• HTTP status code 402 is “Payment Required”, which means that the client request needs to pay to access the resource.

In 1992, Tim Berners-Lee, the creator of HTTP and HTML, released his second version of HTTP and the first reference to the now commonly used status codes. Among them is a code that Tim Berners-Lee and others believe will one day be used to pay for digital content: 402 payment required. Unfortunately, this status code is officially “reserved for future use” because from the beginning, various attempts to make micropayments on the network failed to materialize.

At the beginning of the design of the Internet, HTTP 402 error prevented the Internet from becoming a network that supports (micro) payments. The L402 protocol on the Lightning Network was designed to support authentication and payment in a distributed network. In the Internet, it is: ** Used to pay for Internet native applications or services (egAPI, login, digital resource access). Such services rely on unit economics. **

macaroon is not sweet this time

The macaroon here is not a French dessert, but an advanced authentication mechanism for distributed systems. They are designed to combine the advantages of bearers and identity-based authentication systems in a single token that can be quickly issued and verified without access to a central database.

Macaroons are s with Contextual Caveats for Decentralized Authorization in the Cloud [4]

The representative entities of AI are intelligent LLM and AI agents. There is no native relationship between them and the legal currency system (you cannot register an account and show your ID card). Macaroons can give identities to AI entities in distributed systems (authentication mechanism).

Bitcoin is mine, sats belongs to AI

I am reminded of a question asked by a friend of mine. No matter what her expectations for the future of Bitcoin are, she is personally unwilling to use Bitcoin for micropayments (buying coffee, banana ice cream waffles). Indeed, it’s too late to save up, so I don’t want to waste the pie. A barrage suddenly came to mind: What if Bitcoin is not for human use to some extent (sats)?

However, these proxies will definitely need to pay for resources, whether from gated APIs or paid data sources. Additionally, they need to be able to effectively evaluate pricing signals to determine the most efficient path to accomplishing a task. These payments, evaluations, and decisions will result in thousands of AI agents making countless micropayments and microdecisions every day. With these factors in mind, it makes sense that creators of AI agents would ultimately gravitate towards the globally available, permissionless, near-instantaneous settlement, internet-native currency systems in Bitcoin and the Lightning Network, rather than traditional fiat systems that simply cannot support them. of.

If sats are the currency of artificial intelligence, my question is: What does this kind of AI agent that requires high frequency and relies on unit economic micropayments look like? Or in what kind of scenarios are these AI agents actively working?

It is not difficult to imagine that people set tasks/goals for AI agents, and then allocate funds (10,000sats) to let them complete the tasks. The agents can travel through the streets of the Internet to find the best path for you. **But what is the mission? **

iant Funds 的文章 Crypto AI Agents: The First-Class Citizens of Onchain Economies [5] There are several examples, such as:

Gnosis demonstrates this primary infrastructure through its AI mechs. Its AI Agent encapsulates AI scripts into smart contracts so that anyone (or a robot) can call the smart contract to perform Agent actions (such as betting on the prediction market) ), and can also make payments to the Agent.

AI agents need to be fine-tuned for specific industries, topics, and niches. Bittensor incentivizes “miners” to train models for specific tasks (e.g. image generation, pre-training, predictive modeling), around target industries (e.g. cryptocurrency, biotech, academia).

AI

My understanding of AI mainly comes from frequent interactions with GPT.

AI understands how AI works and the AI itself is, is not just maximizing its intelligence and problem solving capability in service, for instance, to answering Bing requests, but it’s trying to maximize its own agency.

And that means it’s maximizing its ability to control the future and play longer games.

“AI understands the work of AI. Their goal is not just to solve problems, but also to think about how to continue (play positive sum game) and be aware of what may happen in the future.” – Joscha Bach [6]

A recent Open ai paper [7] , studied how to stimulate the ability of powerful models through weak supervision. The research team fine-tuned the GPT-4 series of models on different tasks and found that models under weak supervision performed better than their weak supervisors. This phenomenon is called “weak-to-strong generalization.” In human terms: **How to make computer programs (such as chatbots) perform well when receiving less detailed instructions. Often, we need to give these programs very specific instructions to make them work correctly. **But research has found that even if the guidance is less specific, these programs sometimes do better than expected. It’s like teaching a child to do something. Although he is only given basic instructions, he can figure out how to do it better on his own.

For example, one is the recent experience of frequently calling APIs: Google scholar, semantic scholar, GPT (indirectly), which makes me feel that my role in this is to move the API KEY and copy the API deion (so that GPT can matching format)**.

The interaction during this period is interesting: people also need to learn code, but more importantly, they need to understand the role of each component and the connections between them. For example, when designing a system with xyz goals, KG organizes the skeleton of ideas, and API connects data joints. We can think more about what to connect (to create more interesting things or achieve what goals), and GPT can provide solutions on how to connect. I recently tried the role play of [Completing Tasks with GPT], in which his role is for developers to provide code solutions, and I am the porter of API keys (laughs) and provide some ideas for which APIs to assemble:

Examples have emerged of giving them specific toolkits to run on their own. One project that is frantically testing this boundary is tldraw: draw-a-UI. They are testing the ability of AI to combine with many APIs to complete different interactive tasks.

AI agents can put forward some unrealistic or unrealistic ideas through people or themselves, obtain some funds through smart contracts with certain rules, and then recruit specific people (developers, designers, memes) to participate in construction and testing. wrong. Perhaps many interesting tasks will be defined and a market will be formed to promote “human-machine collaboration into rewards.”

AI agent is an entity that is bold and imaginative, but lacks boundaries and constraints. Its cooperation with people can make more things happen that were previously unimaginable. And can some kind of corresponding currency, capital, and small rewards that can support its growth be injected with live water through Bitcoin or cryptocurrency?

Reference

[1] Micropayment Transfer Protocol (MPTP)

[2] Nick Szabo: Micropayments and psychological transaction costs

[3] Micropayments and the Lightning Network

[4] Macaroon

[5] iant fund research

[6] Joscha Bach: AI risk and the Future

[7] WEAK-TO-STRONG GENERALIZATION: ELICITING STRONG CAPABILITIES WITH WEAK SUPERVISION