Solutions
AI, IoT: do machines need a Wallet?
Stepping back from the hype surrounding artificial intelligence, we are seeing increasingly more evidence that this technology and its applications are bringing unprecedented improvements in productivity, individual autonomy, and knowledge acquisition. These advancements manifest through natural language conversations with virtual assistants, the creation of images and illustrations, instant translations, the design and correction of computer code, as well as encyclopedic knowledge, among others.
These are the beginnings of a paradigm shift in the relationship between humans and machines, associated with huge productivity gains.
However, a major obstacle remains in the rise of this productivity: the economic autonomy of machines.
What could intelligent machines accomplish if they were equipped with a wallet?
Machines are economic agents
Traditionally, machines are not perceived as economic agents because they are considered tools or resources under the control of other economic agents, namely humans. However, they already actively contribute to our work and value creation. What then about their potential for "economic autonomy"?
In the field of e-commerce, automation is widespread for credit card transactions. However, this automation does not yet extend to other important aspects such as:
-
Cataloging products.
-
Negotiating prices or terms and conditions.
-
Creating accounts and accessing payment methods.
-
Outside these areas, the automation of purchases remains limited, except in specific cases.
-
For example, in the financial sector, some robots automate complex transactions.
-
Similarly, in supply chain management, precise rules allow intelligent machines, like sensors in a factory, to detect stock levels and automatically place orders for replenishment by connecting to pre-established payment systems.
With the rapid evolution of artificial intelligence and connected machines (IoT), it's easy to imagine many new applications that could be made possible by granting machines additional freedom: economic freedom.
In the rest of this article, we will use the terms 'Agent' or 'AI' to refer to a digital assistant. This typically represents an intelligent machine equipped with generative AI or a combination of different models, capable of acting autonomously on the internet and possessing economic capabilities. This defines what we call "the machine economy" or "the autonomous economy."
New applications could be made possible by granting economic freedom to machines.
wallets for AI
Commercial use cases
Let's now sample some hypothetical examples
-
An electric car that, in addition to locating a charging station and negotiating the price, makes the payment while the driver is busy with other things. It also manages parking fees by optimizing costs for the user and can take care of renewing the insurance or the technical inspection.
- We could ask our Agent to plan a weekend in the Pyrenees, specifying our favorite activities (hiking, type of restaurant, kayaking or canyoning). The Agent would develop the best itinerary, search for suitable accommodations, and make the necessary reservations. This task involves breaking down the problem into sub-tasks and having a means of payment.
- A connected TV could instantly acquire content licenses (movies, music) according to the user's preferences, without being limited to a specific catalog. It would autonomously and optimally manage an allocated budget. This requires fluidity in payments and paves the way for a new form of disintermediation, allowing users to directly compensate content creators.
- An Agent could explore the web for information on a specialized topic, paying small amounts to access certain articles, and then generate a customized report. This requires getting past some paywalls. How many of us would like to follow a multitude of topics without having the time to do so, and how to distinguish essential information amidst a mass of publications of varying quality?
- Dedicated Agents for negotiation and contracting for spot production, such as electricity or wheat, could be envisioned. For example, a smart building equipped with solar panels could automatically sell excess electricity to the grid, adjusting prices in real time according to demand.
- Associated with an IoT sensor, an Agent could autonomously conduct transactions for purchasing spare parts or organizing maintenance services before a breakdown even occurs.
-
An Agent could act as the digital twin of a connected device generating data, such as measuring flow, mapping an area, or analyzing congestion. It could then sell this data, even for very small amounts, to other services that need it. In this model, the economic flow accompanies the data flow, thus creating an atomic economy.
-
An Agent specialized in SEA (Search Engine Advertising), autonomous, could manage and deploy a Google Adwords campaign, incorporating a necessary payment step.
-
Faced with a time-consuming task, I could entrust it to an Agent with a budget equivalent to one-tenth of my hourly cost. Even if the Agent is only successful half the time, it would represent a time and productivity gain.
-
An Agent with a budget could work on monetizing my content through various means, such as publishing on social networks, uploading my content to different platforms, and creating promotional material.
The economic flow accompanies the data flow in real-time, thus creating an "atomic" economy.
Initial Lessons
Tomorrow's Human Skills
The integration of atomic payments enables the creation of truly continuous financial flows, akin to a fluid, but in a digital context. Imagine a class of future job skills: managing Agents and constantly monitoring the financial flows associated with them. The efficiency of a company then requires improving its sales flows (in quantity, price, frequency, and scope) and optimizing its expenditure flows (reducing costs, diversifying approaches, limiting frequencies): this management is no longer just a matter of the managerial domain, but becomes a highly operational and real-time function (devops).
This pursuit of constant optimization by machines 24/7 gives us a glimpse of a future where the cost of digital services could drastically decrease.
Tomorrow's Machine Skills
We have indirectly described the capabilities of these Agents:
-
Create subtasks and execute them: i.e., take a need expressed by a human in natural language, break it down into enough tasks so that a majority can be handled by automation and Agents.
-
Capable of “navigating” the internet and contextually managing this complexity.
-
For each task, be able to evaluate it against a predetermined budget. This might even include the ability to post “rewarded tasks” that will be performed by AIs or humans in cases of inaccessible specialization (e.g., microlancer.io or stakwork.com).
-
A generalist agent must be able to call upon multiple highly specialized and quality AI services, which are paid, to execute its tasks as efficiently as possible, as its competitiveness depends on it!
-
Evaluate “price signals” to choose the most efficient path for accomplishing a subtask.
-
Be able to make electronic payments to a web agent and a physical business.
-
Be able to communicate with other Agents, for example, those that are highly specialized.
The market in numbers
Is it possible to estimate the size of this new economy? Let's explore this question from several angles, starting with a comparison to mainstream e-commerce.
Comparison with Mainstream E-commerce
Currently, in terms of AI, we are just at the beginning, at 0%.
Recalling the early days of the Internet, it took five years for online commerce to account for 3% of total B2C retail sales (1999-2004). Applying a similar measure, we could envision that by 2027, 2% of B2C internet sales could be conducted by Agents, which would amount to 2% of 5940 billion dollars, or 118 billion dollars – an impressive figure.
This projection is not so far-fetched, especially considering that few of us remember making online purchases before 2004, due to the immaturity of systems and psychological barriers. Once we have a comfort zone with AI, this evolution will undoubtedly go much faster. Moreover, a part of this economy's figures will come from the monetization of flows that are currently given for free or with little sophistication.
Growth Linked to Machine Interactions
Currently, over 80% of the web uses 'API calls', interfaces that allow structured data exchange between machines, essential for connecting and automating enterprise software, systems, and servers. The software universe is becoming increasingly complex, and it's crucial to delegate repetitive tasks to machines as much as possible. This evolution can only be accomplished if payment barriers, often compartmentalized, are lifted.
magine if a portion of these APIs became easily monetizable, even modestly, through micro-cents? We could then move away from the dominant 'free' economic model, which often comes with data risks.
Regarding IoT: Sector reports forecast that the global IoT market will reach several hundred billion dollars in the next decade, with an estimated number of connected devices in the tens of billions. These projections highlight a rapid transition to greater automation in the economy. The contribution of machines to global economic transactions is constantly increasing, a trend reinforced by significant investments in IoT and 5G infrastructures.
A future in which machines would be economic agents 24/7 could well drastically reduce the cost of digital services. This 'autonomous economy' could be worth several hundred billions
Levers and Barriers to Growth
What types of financial decisions could be entrusted to these Agents, and within what limits?
-
Mental barrier: The reluctance to delegate tasks to machines persists. To gain trust, the system must prove itself with simple and economically low-impact applications, thus supporting the idea of gradual and controlled growth... at least until the competitive advantages become undeniable for skeptics. This is the famous 'gradually, then suddenly'.
-
Risk aversion: Decisions made by complex algorithms analyzing data in real time can be difficult to approve, especially when they involve major risks, such as critical financial losses or dangers to human life. Thus, certain highly sensitive tasks will not be entrusted to machines.
-
Trust issue: With the rise of AI, we face the paradox of 'infinite' identities: how and which machines can an Agent trust? This challenge is commonly defined as a "Sybil attack" (see appendices), more simply understood by imagining farms of spam or trolls managed by AIs. The countermeasure requires the implementation of costs associated with the creation and use of new identities, for example, by making interactions payable. A minimal cost would be "painless" on the scale of the well-intentioned user but prohibitive on the large scale of attackers. This approach is not only impossible to implement using traditional payment rails, as they are too slow and too costly, but with the amount of currency issued tending towards infinity, the problem is insoluble. On the other hand, this is one of the multiple advantages of a currency in a "finite" and "atomic" quantity: digital scarcity becomes a protective rampart.
-
Payment method hacking problem: With the traditional banking card system, hacking is a major issue, leading to high costs, considerable time loss, and chargeback fees for merchants. The emergence of hackers using sophisticated AIs will only exacerbate these problems. Therefore, this system is unsuitable for an economy of instant flows managed by machines. A universal value exchange mechanism, immune to human or machine compromises, is necessary. Hacking of AIs also poses a significant risk to the security of user accounts and the protection of their associated data, highlighting the crucial importance of data security and sovereignty.
Summary of Needs
Competitiveness
To gain market share, digital Agents must be able to access the most efficient AI services, whether they are the most accurate, fastest, or most effective for a given task. However, the most advanced AIs, such as natural language processing (LLM) models, require costly resources like GPUs, hence the need for a means of payment.
Currently, the majority of cutting-edge AI services require the creation of a user account and the use of a credit card to access their most sophisticated models. This reliance on bank cards increases costs for both providers and users and slows down processes, which is unacceptable in sectors where speed is crucial for competitiveness.
Moreover, considering that an API call to an AI service costs only a few cents, or even less, and that the current credit card system does not support these micro-transactions, users are forced to adopt inappropriate subscription models.
Finally, it's important to note that billions of people without access to traditional banking services are de facto excluded from using these innovative tools, which could otherwise boost education, productivity, and creativity. For example, 15 million Americans, 13 million Europeans, and 1.7 billion people in the rest of the world are unbanked.
An Appropriate Payment Rail
It's crucial to break away from the traditional payment system, which is unsuitable for current needs:
-
Machines are neither legal nor moral entities and therefore cannot create accounts or hold a bank card. It's essential not only to overcome these obstacles but also, if possible, to merge them (the wallet serving both as an account and a means of payment).
-
The traditional system is too slow (with 7 to 12 invisible steps between the consumer and the online store) and costly, necessitating a transformation towards immediacy.
-
Simplify the payment flow: eliminate frictions caused by third parties, such as frequent and occasionally justified authorization refusals, and remove the time-consuming need to resolve disputes.
-
With a secure payment system, the immediate finality of transactions becomes possible, eliminating the problem of costly chargebacks ($30 billion/year) and prolonged payment terms with their associated risks. The goal is to achieve near-instant payment finality (about 500 milliseconds), relegating the "90-day payment" to the past.
-
Necessity to eliminate the risk of fraud and significantly reduce the overall risk of payment.
-
Necessity to limit spam by making it costly.
-
Need for a universal and atomic tool.
-
Need for a payment system without geographical constraints, allowing AIs from different countries to trade without currency barriers.
-
Payment capability must be adjustable as needed.
-
High availability (uptime) and resilience are essential.
-
Data security and sovereignty must be prioritized (avoiding the risk of a 'datalake' becoming a 'honeypot' for hackers).
-
Supervisable and controllable system.
Economic Models that fit
It is necessary to develop new economic models adapted to these innovative uses. Although current models will continue to exist, new needs are emerging:
-
For example, the need to set prices much more dynamically, without payment-performing agents being constrained by overly rigid rules.
-
As costs are also variable, the value of certain data can fluctuate based on their timeliness or exclusivity (e.g., yield management). The current stability of unit prices or subscriptions is dictated by the human inability to quickly adapt to changing prices, due to mental or practical barriers (such as the difficulty in consulting new prices or updating an entire catalog).
-
Machines must be able to negotiate prices, requiring a balance and agility in purchasing or pricing policies. Moreover, these negotiations imply the ability to formalize a commitment: the wallet must therefore be both a spending system and a signature mechanism (to confirm orders or service commitments).
There is a need for suitable payment rails and new economic models to realize the full potential of the 'autonomous economy'
Open protocols provide the solution
An Atomic Payment Rail Already Exists
Regarding the size of the addressable market and its anticipated growth, it appears counterproductive to rush things. A gradual, phased construction, with a long-term sustainability goal, seems to be the best approach. This reflects the philosophy of open protocols, which establish rich and inclusive ecosystems, encouraging participation and adherence from many players.
As for the payment rail, there is already an open and universal currency native to the Internet: Bitcoin and second-level protocol layers such as the Lightning Network. To delve deeper into their remarkable efficiency and the advantage of open technologies, we refer to our foundational first article, "The Interest and Power of Value Protocols."
These systems, already proven and functional, could, with their adoption by AIs, amplify their large-scale dissemination. Let's recall their main characteristics:
-
Secure (never compromised) and operational for 15 years.
-
Peer-to-peer and universal, without intermediaries or authorization needed for transactions.
-
'Atomic' payment rail (e.g., 0.003 euro cents), capable of handling a thousand times more transactions than the traditional banking system, and a thousand times faster.
-
Finite monetary policy and engraved in its code.
-
Components to Develop on the AI Side
For the AI part, let's explore the necessary components for a complete solution:
-
Native method for restricting access by payment: The HTTP 402 status code, known as 'Payment Required,' is a standard of the HTTP protocol. It informs the client that a payment is required to access a resource. The HTTP 402 response is accompanied by an entity body that provides the client with additional information about payment requests. Although this code has existed since the early days of the internet for micropayments, it is still little used by modern browsers.
-
Open protocol for authentication on paid APIs: The L402 initiative, including a native Bitcoin Lightning authentication mechanism and 'Langchain,' a library to facilitate interactions with AI applications, is a promising example, albeit recent.
-
Transformation of existing APIs into paid resources: For example, Aperture, a reverse proxy server based on L402, can convert any API into a resource accessible on a pay-per-use basis.
-
Standardized method for communication between agents: It is probably necessary to have a uniform method, for the payment part, or even the presentation of the data. It is very likely that we will see a few major patterns, each with their advantages, being copied and reproduced, becoming de facto a standard. And since AIs are increasing their coding and API integration capabilities every day, the step won't be high.
-
Discovery mechanism for AI services on the web: Is it necessary to have a centralized service for this? Not necessarily, as ultimately AIs are intelligent. Several promising distributed network-based approaches exist in this regard.
-
Defining limits: Should the roaming of agents on the internet be limited for security reasons? This is an open topic for debate with arguments on both sides.
Already operational, the Lightning Network is the perfect payment rail solution. The first functional interconnections with autonomous machines have already been launched.
Conclusion
The emergence of machines with autonomous capabilities suggests a future where economic transactions are no longer limited to human interactions. Not only will they be able to assist us in our end-to-end tasks if they are equipped with payment capabilities, but we will also have to rethink many economic models and invent new professions.
However, traditional payment systems are obsolete in the face of the needs of this autonomous economy. Only the adoption of "atomic" and instant payments, made possible by the Bitcoin Lightning Network technology, can pave the way for continuous and fluid large financial flows, suited to the economy of autonomous machines.
Ironically, in a world where the number of AIs could be infinite, it appears even more essential to have a finite currency. The convergence of this universal digital currency with artificial intelligence could well be the decisive catalyst for the autonomous economy.
This evolution also raises new questions: if entire swathes of the economy were continuously optimized by a multitude of autonomous and specialized AIs, would we witness the advent of hyper-efficiency? What will be the new power dynamics and challenges?
The convergence of this universal digital currency with artificial intelligence could well be the decisive catalyst for the autonomous economy.
Appendices
The Sybil attack is a type of security threat in computer systems and networks. The term is named after the clinical case of Sybil Dorsett, a woman with multiple personality disorder, and was popularized by researcher John Douceur in his 2002 paper "The Sybil Attack."
Definition: In a Sybil attack, a single user creates multiple fake identities (often called "Sybils") in a network. The goal is to exert disproportionate influence in the network, either by overwhelming the system with actions from these identities or by manipulating decision-making or consensus processes in the network.
Examples of Impact:
-
Peer-to-Peer Networks: In these networks, a Sybil attack can allow the attacker to control a large part of the network, affecting the distribution of resources, the reliability of data, or the security of communications.
-
Voting or Consensus Systems: Sybil attacks can manipulate voting results or consensus processes, creating the illusion of massive support or opposition.
-
Social Networks and Reputation Systems: Sybil attacks can distort reputation systems or influence trends on social networks.
Prevention and Mitigation: Combating Sybil attacks typically involves implementing identity validation mechanisms or costs associated with creating new identities, making it difficult for a single user to maintain multiple active identities.
Source Elements:
-
WSBI-ESBG: Number of Unbanked Adult EU Citizens
-
CFTE Education: The World's Top 5 Unbanked Countries
-
St. Louis Fed: Economic Complexity Index
-
Precedence Research: B2C E-commerce Market
-
Voltage: Lightning Network Use Cases and AI
-
ARK Invest: Bitcoin Brainstorm
-
Voltage: Lightning Network and AI Impacts
-
Lightning Engineering: L402 Langchain
-
Lightning Network: L402 Documentation
-
GitHub: Lightning BLIPS
-
Ten31 VC: Insights on AI
-
Kinsta: HTTP 402 Code Knowledge Base
Demos and Services:
-
GitHub: Matador Project
-
Vimeo: Matador Video Demonstration
