Go2Mars Research
Swarm AI refers to a collective intelligence system in which each AI model works together to solve complex problems. HyperCycle is a network platform that uses blockchain technology to achieve safe and efficient transactions between AI machines. It allows different AI algorithm models to collaborate with each other to form a global intelligent brain. With HyperCycle, AI algorithm models that were originally independent and single-function can cooperate with each other to execute complex intelligent algorithm processes internally, thereby achieving a qualitative change.
The effect of quantitative change leading to qualitative change has been verified over and over again in AI algorithm models. In the transition from GPT1 to GPT4, the real iteration lies in the change of parameter quantity, which makes GPT4 produce intelligence that is difficult for researchers to explain. If we are more ambitious and bring together models of all sizes such as GPT, BingChat, midjourney, and even Wenxin Yiyan, it is no exaggeration to say that AI intelligence will rise to a whole new level in a short period of time.
This ambitious concept is called swarm AI, where each AI model in the system will work together to solve complex problems. In layman's terms, we can understand it as individual efforts and teamwork - individuals in individual efforts need to be responsible for everything, while individuals in teamwork only need to be responsible for part of the work.
Ideals are always full, but reality is always skinny. It is nothing but a pipe dream to expect these technology companies to put aside their conflicts, make joint contributions to human development, and voluntarily share their core algorithms and databases.
But what if we use blockchain technology to break this chain of suspicion?
HyperCycle Application Introduction
It is definitely unrealistic to coordinate this group of AI technology giants to collaborate on AI technology through the assistance of organizations or other third-party institutions. We need a truly fair and just rule "supervisor" to maintain the order of the game. HyperCycle plays such a role.
As early as 1995, Ben Goertzel, the core founder of HyperCycle and the chief AI scientist of SingularityAI, had the idea of building a decentralized basic platform architecture for the global artificial intelligence network by leveraging the development dividend of the global Internet. However, due to the limitations of objective conditions such as network hardware, the cost of working in a secure decentralized manner is very high, so this idea has always been just an idea.
Fast forward to 2015, when Ethereum proposed the concept of smart contracts. Although strictly speaking, smart contracts are neither intelligent nor contracts, the idea of verifying and maintaining the overall network security in a decentralized manner by continuously setting verification scripts in network nodes in advance allowed the founders to see a new idea for the first time to verify this AI innovation experiment - making nodes smart.
If we describe HyperCycle in a concrete form, it can be understood that it is a blockchain architecture that allows AI algorithms and data to be uploaded to the chain. Through decentralized organizational design and more efficient data transmission and security technology, it allows AI algorithms of different projects to share AI computing power and ensure that each AI algorithm can get the "reward" it deserves after the algorithm calculation.
From a technical perspective, HyperCycle can be described as a brand new blockchain architecture, which is composed of TODA/IP and TODA framework, consensus and reputation mechanism, system supervision, smart contracts, MeTTa, etc. HyperCycle can handle high-speed, large-scale on-chain agents and interactions, such as on-chain deployment of artificial intelligence algorithms trained based on crowd data and interactive media driven by token economics.
With the help of HyperCycle, AI algorithm models that were originally independent and had single functions can cooperate with each other and execute complex intelligent algorithm processes internally, thus achieving a qualitative change.
TODA/IP bookless blockchain: making blockchain transmission efficiency suitable for AI
In order to make blockchain technology suitable for AI, the core problem lies in the time cost and data transportation cost. In the traditional blockchain system, every time a transaction occurs, all nodes will receive all transaction data, hash reference values and block headers. This replicated ledger recording mode causes the system's recording efficiency and cost to increase linearly.
In the past iterations of the model, some projects proposed a solution to sharding management, which divides the block nodes into different fragments, and each fragment only records the transaction records it is responsible for. This sharding model solves the efficiency problem to a certain extent, but it also increases the complexity of the system and further increases the system operation cost.
TODA/IP is a secure and efficient peer-to-peer network protocol based on cryptographic principles. It allows each network data packet to have a unique global identifier and belong to a data structure that ensures that the data packet belongs to a single signature public key.
TODA/IP is somewhat similar to the underlying principles of shard management, but it uses a more thoroughly decentralized approach to keep the system lightweight and low-overhead - each local data block is responsible for managing its own historical information.
The core structure of TODA/IP is to link individual records to their own localized ledgers, which makes these records semi-autonomous agents in a sense. When trading or transferring data, node individuals only need to interact with transactions related to their own ledgers, especially those involving disputes.
Given a record R owned by wallet A, the record can be sent to wallet B by generating a transaction request, which is signed by wallet A, then signed by B, and finally distributed and signed by a set of validators. A cycle of TODA/IP consists of a round of transaction requests and subsequent verification.
In such a transaction process, the transaction validator replaces the traditional ledger and provides four important functions for the entire transaction:
▪️Determine the validity of the transaction (structural soundness and proof correctness)
▪️Prevent sending the same packet twice in this cycle
▪️Help establish consensus proof of transactions
▪️Provide proof that A and B match
In order to match TODA's data architecture, the file format transmitted between nodes has also been iterated accordingly. A "TODA file" is essentially a digital data file that is attached to the ledger of each file as metadata. The combination of a file's internal data and its ledger allows a file to behave like a "unique digital object" (an NFT, which can be understood as a bunch of keys that open a lock together, or part of a lock).
When transactions occur, each transaction involved causes a corresponding record to be appended to the relevant ledger of the file. These transaction records also contain other information, such as the addresses of the other parties involved in the transaction.
With the help of the TODA/IP mechanism, the originally cumbersome accounting data transmission process is optimized, and the transmission efficiency of the blockchain is matched with the AI algorithm.
POR (Proof of Reputation): Secondary lubrication of efficiency and fairness
Efficiency alone cannot support the operation of the entire swarm AI, and fairness is also a crucial part of the blockchain architecture. In HyperCycle, the system introduces a dynamic consensus mechanism that is more suitable for AI algorithm systems than Pos and Pow mechanisms - Proof of Reputation (POR).
In the earliest blockchain systems, the PoW (proof of work) mechanism was used, and each node had to prove its participation through computing work to obtain corresponding rights and interests. However, the PoW mechanism consumes a lot of energy, so a lighter PoS mechanism emerged: network nodes can obtain the right to confirm transactions by owning tokens without performing computing work.
Although PoS avoids the energy waste problem of PoW, it still continues the Matthew effect of "the rich get richer and the poor get poorer" in PoW, that is, consensus can only be provided by nodes with a large amount of data processing capabilities or tokens. For nodes that want to truly win-win or small and medium-sized models willing to provide computing power for the entire swarm AI, this winner-takes-all situation is obviously unfavorable. For this reason, the reputation proof mechanism (PoR) came into being.
TODA had its own consensus mechanisms, which were also great, but layering a Proof of Reputation dynamic on top of them would increase efficiency and simplify things in many use-cases.
The core idea of PoR is to use a reputation score based on liquidity-weighted rating as a consensus mechanism for the blockchain network. The reputation of the network nodes used by the PoR framework is determined over time and interaction.
The reputation of a single node is calculated by mixing the normalized rating with the reputation of the entire node, and it changes dynamically over time. This is not simply the rating value given directly by other nodes. The behavior of a single node will also affect its overall reputation through the corresponding quantitative formula of the system.
On this basis, the PoR mechanism will determine a group of consensus nodes responsible for maintaining the heat sharing status based on the reputation value, and over time, the reputation value between nodes will be continuously updated with interaction. At the beginning of each round of the PoR consensus mechanism, the community needs to select consensus group members and invite them into the corresponding consensus group, and the members of the consensus group will be selected from the nodes with the highest reputation value.
For example, when the collective reputation score exceeds 50% of the total network reputation value, a leader is elected from the group, whose functions are:
▪️Pack all valid transactions in the pending transaction list into one block
▪️Use the transaction data in the transaction list to calculate the new reputation value generated by all network nodes
▪️Broadcast and submit information to the consensus group
The greatest value of the PoR mechanism in the AI innovation ecosystem is that no matter whether the participant uses a large model or a small model, as long as its algorithm is effective enough for the entire cluster, it will receive a higher reputation score and reward. This will greatly attract various small and medium-sized AI model project parties to actively participate, while also giving full play to the advantages of swarm AI.
Lightweight "ring": the finishing touch of mechanism innovation
In the overall architecture of HyperCycle, the most ingenious design is the lightweight "ring" mechanism.
According to the definition in the HyperCycle white paper, nodes in the TODA/IP network can be identified as "rings", each of which is a set of specific records and is hierarchically connected. A TODA/IP system can hold a collection of rings, which we can call a "ring set". A minimum node is the so-called "lightweight ring".
The most ingenious part of the lightweight ring is that it can be combined with other rings to form a ring set to meet super-large computing power or data transactions, while it can also become an independent ecosystem. If one day GPT joins this swarm AI, with the help of the node mechanism of the "lightweight ring", it can exist independently as a world-leading AI language generation chat system, or it can integrate itself with the cluster and become one of the language content generation sections of a certain AI virtual person.
MeTTa contract language: making contracts truly intelligent
A new AI blockchain architecture is now only one step away, that is, how to migrate various AI algorithm models to the chain. The answer is: a native smart contract MeTTa. The MeTTa (Meta Type Talk) language was developed in the context of the OpenCog Hyperon AGI project. It has many excellent features, making it the core intelligent language of HyperCycle.
MeTTa has flexible underlying semantics, such as equivalent processing for similar configuration theories. AI project parties can directly call the system through the API interface to handle various transactions, which means that no redundant script operations will be performed when processing TODA/IP messages or TODA data, thereby improving operational efficiency.
On the other hand, MeTTa is a language rewritten based on metagraphs, which represents data and code in a unified way, implements high-order functions and dynamic types. MeTTa supports multi-paradigm programming, including functional, logical, object-oriented, and concurrent. MeTTa also provides a powerful type system to ensure the correctness and security of the code.
More importantly, by using MeTTa's compiler, the project can compile the MeTTa source code into the source code of the rholang language originally developed on the Rchain blockchain. This enables MeTTa to achieve true intelligence with the powerful concurrency features of rholang. It is even possible to use tokenomics to manage the allocation of computing resources within the MeTTa smart contract.
With their new scheme, every validator in a network doesn’t need to verify that a smart contract is being run correctly, only a judiciously chosen random subset.
With MeTTa, both AI and other project parties can freely run related smart contracts on HyperCycle nodes, automatically manage resource allocation, outsource work to others, and actively encourage cooperation between nodes to solve problems.
With this final step, our swarm AI will truly come alive.
Summary
The combination of blockchain and AI can bring new value to business and society. Blockchain provides a secure and transparent database for storing encrypted data, while AI has the ability to simulate human thinking to solve problems. When used together, blockchain can improve the credibility and transparency of data resources used by AI models and increase the speed of AI operations by connecting models to automated smart contracts.
In addition, the combination of blockchain and AI can also achieve a unified representation of data and code, thereby realizing high-order functions and dynamic types. This means that AI can quickly and comprehensively read, understand and associate data, bringing new intelligence to blockchain-based business networks. By using blockchain to store and distribute AI models, it can also provide audit tracking functions and use tokenomics to manage the allocation of computing resources.
In summary, the combination of blockchain and AI not only improves the credibility and transparency of data and models, but also brings more efficient, secure and intelligent solutions to business and society. This combination will bring more innovation opportunities to all walks of life and promote social progress.
references:
[1]https://wiki.opencog.org/w/File:A_Formalization_of_Hyperon_MeTTa_language_in_terms_of_metagraph_rewriting.pdf
[2]https://wiki.opencog.org/wikihome/images/1/1e/Basic_Atomese_Features_required.pdf
[3]https://medium.com/singularitynet/hypercycle-the-journey-to-a-fully-ai-capable-blockchain-9d2b7431cfa1
[4]https://medium.com/singularitydao/singularitydao-launchpad-hypercycle-token-generation-event-9eeeea17f0ce
[5]https://medium.com/singularitynet/hypercycle-a-convergence-of-radical-technologies-c59aeb83ab3
