A Review on Blockchain Technology, Current Challenges, and AI-Driven Solutions

Rollup solutions proposed off-chain computation to lower the cost of contract validation. Truebit, an Ethereum smart contract, delegates expensive computations to a trusted third part the solver and rewards a challenger who checks the solver’s work. This facilitates reliable intensive applications since Ethereum gas costs limit on-chain computations [179]. Arbitrum, an Optimistic Rollup, also uses off-chains contract verification to improve its scalability andit is designed for private contracts. It incentivizes parties to reach off-chain agreements about contract behavior, enabling miners to verify signatures without accessing the contract. This is designed for private contracts [86]. In summary, both Truebit and Arbitrum reduce computation costs and improve scalability via off-chain processing and selective data access. They incentivize proper computations through verification rewards and penalties.

It is a provisionally off-chain trade channel that is used to shift some transactions from the primary chain to this channel to minimize the transaction volume of the primary chain and to increase the system throughput. The Lightning network [84], which is used by Bitcoin, and the Raiden network [141] for Ethereum are two examples of payment channel methods.

In the last decade payment channels have been extensively studied, with many Lightning Network implementations being released. There are also numerous alternative off-chain payment channel options, such as Sprites [117] and Bitcoin DMC (Duplex Micropayment Channels) [59].

A multi-chain Blockchain system refers to the use of multiple interconnected blockchain networks working together in order to achieve a specific goal or set of goals, such as increased scalability, interoperability, and security. This can be accomplished by creating new blockchain networks or by connecting existing ones. Multi-chain has three categories which are: Side chain, Cross chain and sharding, which are explained one by one in the following

Side chain. A Sidechain is a distinct blockchain that is linked to a primary blockchain, enabling the movement of digital assets or information between the two chains [158]. This allows for the main blockchain to remain secure and decentralized, while also providing additional functionality or scalability to the sidechain. One of the main benefits of using a sidechain is that it allows for greater flexibility in terms of the types of transactions that can be performed and the types of assets that can be transferred. Plasma is an Ethereum scaling framework allowing infinite side chains in a tree structure [85]. Pegged sidechains enable asset transfers between blockchains securely and atomically [17]. Finally, Liquidity Network uses Merkleized Interval Trees to represent user balances, enabling real-time transactions without limits [99]. It ensures high throughput and growth while eliminating fees and delays. Sidechains like Plasma, Pegged, and Liquidity Network improve scalability and interoperability. They allow new features on sidechains while retaining mainchain security. Structures like Merkleized Interval Trees provide balances for fast, unlimited transactions.

Cross-chain. Cross-chain is another solution to addressing the scalability limitation. It has been attracted/used by many and has become one of the popular solutions being adopted. Cosmos [51] is an example of Cross-chain which is a blockchain ecosystem made up of interconnected blockchains. The network is made up of several different blockchains, each of which is referred to as a zone. Those zones can connect with one another with an Inter-Blockchain Communication protocol, which allows heterogeneous blockchains to communicate, thanks to consensus methods like Tendermint consensus [63]. Polkadot [40] is a cross-chain system that includes a relay chain for connecting different blockchains.

Sharding. Sharding is another solution devised for addressing scalability. It is a database partitioning strategy that splits up a huge database into faster, smaller, and easily manageable fragments, called “shards” [198]. Technically, shard is synonymous with horizontal partitioning. In blockchain, it divides the entire blockchain network into shards or committees, which are smaller groupings of nodes. The shards work in parallel to process distinct transactions and keep a separate ledger. When sharding is applied in blockchain, the network experiences a different life cycle. As depicted in Figure 2, the life-cycle is composed of: (i) a bootstrapping stage, and (ii), epochs with random duration. Each epoch is composed of two phases: consensus and reconfiguration. The consensus phase is where the calculation, the mining of the blocks, and the verifications are performed. The reconfiguration phase varies from one protocol to another. However, generally, at the end of each epoch the committees are reconfigured according to the information of the current mining committees, so the implementation of the new configuration, i.e., the reorganization of the shards and committees of the next epoch always depends on the committees and the previous state of the blockchain.

Elastico [107] was the first to concretize the idea of sharding on the blockchain confirming the idea that as long as there are miners, we can generate blocks. But to achieve high scalability, it sacrificed the aspect of security in the network of the committees of the blockchain. RsCoin [56] is another sharding solution for computation and storage, but loosing one of the fundamental principles of the blockchain: decentralization. What Omniledger [90] has added over the public sharding protocols that preceded it, is the “trust but verify” concept and “Atmoix” protocol which solves the problem of UTXOs (unspent transaction output) blocked for cross sharded transactions that was a big flaw for Elastico. However, OmniLedger has also some drawbacks, such as forcing users to actively participate in cross-shard transactions, which represents a significant assumption for most lightweight users. Moreover, Denial of service attacks appears to be a threat to OmniLedger, i.e., for a malicious user who can exploit the atomic cross-shard protocol to lock arbitrary transactions. QuarkChain [139], founded by Qi Zhou, is a blockchain platform that employs a unique heterogeneous sharding solution. It is a two-layered ecosystem consisting of a root chain and multiple shards. This allows for the creation of custom shard chains within the network, each with its own consensus mechanism and miners or validators, resulting in increased flexibility. Monoxide [188] makes use of asynchronous consensus zones, each one can be thought of as a shard. This protocol is built on the account/balance transaction model, which is analogous to a bank account model, rather than UTXO transaction models. Monoxide is classified as a kind of blockchain-sharding method in theory. RapidChain [200], it is the first solution with communication sharding. As it allows sharding in calculation, storage, and communication, it is called a “full sharded” blockchain: a blockchain technology based on sharding that is extremely scalable to huge networks. Like the others, Rapidchain has some drawbacks as it is sensitive to partitioning attacks and has responsiveness problems. Lastly, we have Ostraka [109] which is unlike the previous sharding approaches, it shards the node rather than the entire system.

The Blockchain issues are generally related to block size in several ways. Increasing block size obviously allows having more transactions in a block. The same result may be achieved via block compression, which also reduces storage overhead. There are also many other alternatives that investigate data reduction strategies. The method of segregating the digital signature of the transaction is known as a segregated witness (SegWit). It consists of no longer saving the signature data with the transaction data but placing them in a separate structure called a witness [104]. Bitcoin also attempted to modify the block size in its network to address scalability. To achieve this, Bitcoin divided the network into two Bitcoin and Bitcoin-Cash by performing a hard fork in 2017. The block size of Bitcoin-Cash [42] raised to \(8MB\), which is significantly higher compared to the previous setting where the size was from \(1MB\). Following that, Bitcoin-Cash block size was updated again to \(32MB\). The Bitcoin-Cash average block interval has remained unchanged at 10 minutes. The transaction throughput can theoretically be considerably enhanced. The stress test, done in 2018, corroborated this. Improving block size can directly improve blockchain capacity scalability from both a theoretical and practical standpoint. However, because intra-blockchain bandwidth is limited, limitless growth increases the size of each block, making it difficult to transport. As a result, just raising the block size is not a long-term solution.

Danzi et al. [57] proposed a method whereby nodes within blockchain networks aggregate blockchain data into periodic updates, aiming to alleviate the communication overhead among interconnected devices. This synchronization process can be customized to meet real-time information needs. Adhering to Ethereum specifications, this aggregation approach reduces communication costs. Data on the application layer is gathered and shared only when required by end devices, optimizing communication expenses. However, this strategy is constrained by the limited computational and storage capabilities of systems, despite its benefits in cutting communication costs.

Consensus. The Consensus Algorithm is responsible for preserving the integrity and security of the whole blockchain. Consensus consists of achieving a particular agreement among all participants in the network. However, certain algorithms, such as Proof-of-Work, consume significant energy to achieve a single agreement by solving a hash puzzle and have limited throughput, making them unsuitable for blockchain. Hence, researchers attempted to propose different solutions to address the shortcomings of the PoW and optimize the performance of the blockchain. Alternative consensus protocols provide different criteria for block validation rights to improve aspects like sustainability, security, and fairness over PoW. We have listed some of the existing solutions below:

We summarize the benefits and limitations of each scalability solution in Table 5.

From the standpoint of large-scale complex systems, there is not a unified vision of artificial intelligence or machine learning-backed blockchain system energy consumption optimization as mentioned in the works below.

WekaCoin [37] is a Blockchain built on the “Proof-of-Learning” consensus protocol. By ranking the models produced for a specific task, Proof-of-Learning achieves distributed consensus. As part of the proof of learning, WekaCoin network participants specifically miners participate in the network by providing their computing power to solve complex mathematical problems. As a result, the protocol makes the computational process task easier for people having greedy training or calculating tasks, such as machine learning, where large amounts of data and computational resources are required to train effective models.

Lan, Yixiao and Liu, Yuan, and Li, Boyang [94] have presented a new consensus mechanism, the Proof of Learning (PoLe), that uses the computing capacity to reach neural network optimization (ANN), instead of wasting energy. PoLe creators proved that PoLe could reach a more stable block generation rate than PoW, which allows more efficient transaction processing.

Chenli, Changhao, et al. introduce Proof of Deep Learning (PoDL) [47] consensus, which obliges miners to train deep learning models rather than hash calculation, and they introduce qualified deep learning models as proofs. PoDL is the first consensus system that uses deep learning to maintain blockchains.

Based on the approach adopted in PoDL [47], Luo et al. presented S-PoDL [106], another deep learning-based consensus protocol. In S-PoDL, all the nodes are separated into sections for training several deep-learning models, at the same time, supplied by model requesters. Each section has a designated node that serves as the section leader, and all other nodes abide by the decisions made by this leader to train a requested model. The leader is the one who assesses the accuracy of the models submitted by the section nodes. Consequently, the communication overhead between nodes is reduced. As the leader requires the section nodes to work on specific models, S-PoDL outperforms PoDL in terms of computing performance and results.

Coin.AI [20] is based on Proof-of-Useful-Work (PoUW) consensus mechanism. The main idea behind PoUW is to incentivize computational work that is useful, rather than just wasteful computational work used in PoW to secure the network. So the nodes train deep learning models, and when their results achieve a certain accuracy, then they can add a new block. Other nodes verify a block by checking the transactions included in the block and the proof (the provided result for a training model).

PoDLwHO [118] is another AI-based solution applied to optimize the sustainability of blockchain. PoDLwHO [118] uses the inane computations of PoW for training DL models by setting hyperparameters instead of calculating purposeless hash values. The contribution of this approach is to use many Bayesian constructs consisting of the evaluated samples in order to help miners to get the best accuracy of specific deep learning problems by using the previous results in a more beneficial way. In other words, till now PoDLwHO is the only solution that retains the nodes’ results even if they do not reach the requested threshold.

BlockML [116] is an alternative system that replaces PoW hash-puzzles with the training of machine learning models. This approach distinguishes itself from other approaches by the concept of the absence of new coins being minted after block generation. In fact, in BlockML coins are not created for newly generated blocks. This helps to counter malignant entities that may attempt to complete their own task because the participant can adopt any role as a supplier or a miner.

In other solutions, e.g., Coin.AI, which uses “Miners” and “keepers”, the miners are responsible for training the models and adding blocks to the blockchain and the keepers store the models in the blockchain.

In Coin AI [20], the miners or the trainers should agree on the ML task that they will train its model before starting the process. So, when using this approach, miners cannot train multiple models of different ML tasks in the same epoch, which is unlike other approaches where miners can work on several ML tasks at the same time. After the model creation, the model must be stored somewhere. There are three model storage approaches that can be used as follows.

The article [32] proposes a Proof of Human work (PoH). It is proof that a person makes a reasonable effort to resolve a problem. A PoH problem must be fairly difficult to be completed by a human. A PoH problem must be difficult to be resolved by a computer without human aid, even the computer that created the puzzle. CAPTCHAs [33] (Completely Automated Public Turing test to tell Computers and Humans Apart), on the other hand, are just challenging to solve for other computers, not for the computer that created the problem. Also, Proof of Human work is verifiable by a machine without the need for a person’s intervention or interaction with the creator of the PoH.

Scalability in the context of blockchain refers to a system’s ability to scale as the number of users increases. Latency (transaction confirmation time), bootstrap time (transaction verification time), and the cost produced for every validated transaction are all examples of scalability issues. Overall, one or more of these scaling limitations restrict the efficiency of a blockchain system. Because each block contains a certain quantity of transactions, traditional centralized data mining approaches are failing to deal with this condition. However, new ML consensus algorithms based on federated learning as POFL achieved solutions that learn from dispersed data sources, which provided a worldwide optimal solution for the target blockchain system.

To address PoW’s shortcoming, Proof of Federated Learning [138] (PoFL) re-invests the energy consumed in PoW for the meaningless hash task into federated learning. To reach a consensus, PoFL uses federated learning to tackle meaningful challenges with practical value. Requesters post challenges like analysis and image recognition as jobs, coupled with the accompanying prizes to incentive miners. To reach a consensus, miners must complete these tasks as proof of work.

A novel dynamic sharding blockchain framework named SkyChain [202] addressed the limitations of static sharding in existing blockchain systems using DRL to adjust the resharding interval, the number of shards, and the block size in real-time. By doing so, SkyChain ensures a long-term balance between performance and security in the constantly evolving blockchain environment.

To address the high throughput demands of blockchain-enabled Industrial Internet of Things (IIoT) systems, Liu et al. [102] introduced a novel performance optimization framework utilizing deep reinforcement learning (DRL). This advanced framework dynamically tunes critical parameters, including the selection of block producers, optimization of consensus algorithms, adjustment of block sizes, and modification of block intervals. By employing DRL, the system can adaptively enhance these elements to boost system scalability, allowing IIoT networks to effectively manage increasing data volumes and operational complexities. There is also Reference [199] that employed the deep Q-network (DQN) algorithm to dynamically adjust the already mentioned parameters and also the number of shards. This approach seeks to optimize these settings while maintaining the security of the system. In a similar vein, the study in Reference [137] introduced a DRL-enabled adaptive blockchain scheme, which not only improved scalability but also addressed the diverse needs of users. By leveraging a DRL algorithm, the system can select the most suitable consensus protocol based on the users’ quality of service requirements.

Blockchain suffers from some security limitations, such as application layer vulnerabilities, data encryption techniques, such as: vulnerabilities that can result in a successful attack, brute force, ciphertext-only, and known-plaintext attacks; additionally, implementation errors caused by improper implementation of encryption algorithms can result in security vulnerabilities and weaknesses.

Computational Intelligence is commonly employed in security systems to improve the security of computer systems and networks, it is used for Intrusion Detection and attack prevention. Computational intelligence is also applied to keep blockchain systems secure. The benefit of applying it in blockchain systems is to have strong and secure encryption algorithms, which play a crucial role in safeguarding sensitive blockchain data. Also, it can be applied to enhance the system attack-defence mechanism to guard against attacks [38, 75].

The work presented in Reference [150] proposes a method for detecting anomalies based on a deep learning encoder-decoder model trained using aggregated data taken from blockchain activity monitoring. The suggested anomaly detection method employs a neural encoder-decoder model capable of encoding information about the ledger’s state into a latent space and then reconstructing the original data from this space. The fundamental principle is that the encoding/decoding procedure retains the data’s essential features whenever the current condition is consistent. Anomaly situations, on the other hand, have inconsistencies in their values, which leads to a failed reconstruction. There are also, other works that adopted the same concept for detecting anomalies like [35, 73, 173] and [62] that discussed the challenge of blockchain behavior pattern categorization and proposed solutions as PeerClassifier [173], a deep learning-based technique to handle the issue.

Blockchain technology is seen as a viable option for maintaining privacy in applications due to its ability to securely and decentralize storage and transfer of confidential data [181]. Examples of privacy-preserving applications of blockchain technology are: Zero-knowledge proofs like Zcash [23], Homomorphic computation [204], and Multiparty computation (MPC) [9], these methods enable processing and verifying data without disclosing it. Also, on the side of privacy-preserving by smart contracts [22], we have the automation of the execution of transactions, particularly private transactions, like sharing confidential information [182]. As blockchain technology becomes increasingly prevalent, it is being integrated into more and more projects. However, as its popularity grows, the potential security issues also increase. It is critical to process transactions in a secure manner. Rachit Agarwal et al. [3] offers a method for detecting fraudulent accounts based on the temporal structure of blockchains. In this solution, they used supervised and unsupervised Machine learning algorithms.

T–S fuzzy ANN presented by [44] can represent a traceability chain with each block and its neighbor. Their main contribution consists of minimizing the quantity of irrelevant data in order to infer more coherent linked transaction data from its computed entire history and participant nodes.

Liang et al. [98] introduce a novel approach to combat anomaly intrusions in blockchain systems using a collaborative clustering-characteristic-based data fusion technique. Their system amalgamates data from multiple sources and employs an AI model to train and scrutinize data clusters, thereby enhancing the precision of intrusion detection. By applying a mathematical model, the system detects abnormal characteristics within blockchain data and performs a weighted aggregation of information across various nodes.

To combat Ponzi scheme fraud, which is a fraudulent investment operation where returns to earlier investors are paid from the capital of newer investors rather than from profit earned, Chen et al. [46] presented a method where relevant features were extracted from user accounts and operation codes of numerous smart contracts then analyzed using the XGBoost machine learning algorithm to identify potential Ponzi schemes within these contracts.

On the Hyperledger Fabric blockchain platform, and with a relatively low tolerance for malicious activities, Salimitari et al. [147] developed an external detection algorithm using supervised machine learning, positioned before the consensus protocol as a preliminary step. This detection algorithm validated the compatibility of new data and discarded suspicious entries, thereby enhancing the network’s fault tolerance for the subsequent consensus phase.

The security solutions mentioned above generally operate based on the actions of a participant in a blockchain to detect any intrusion, malicious actions, or problems with smart contracts. However, these solutions analysis and response processes take a significant amount of time, making it difficult to quickly detect attacker behavior and intentions. Currently, there is no solution that can identify compromised identities and security flaws in real-time. This means there is still a lack of robust systems or solutions that can simulate blockchain attacks to prevent a breach or stop it before it occurs. Therefore, these solutions are missing the prediction feature that will make them respond immediately to malicious problems.

The number of available resources in a blockchain network is known in advance. As a result, as in the works below, AI can perform active and dynamic node selection to speed up the mining process and enhance overall system performance by allowing the network to select the most efficient nodes for mining at any given time. The selection process can take into account a range of factors, including the computational ability of the node, its network connectivity, and the amount of work it currently has. By closely monitoring and picking the most suitable nodes for mining, the network can guarantee that blocks are mined quickly and efficiently.

The idea of Chen et al. [43] is a Machine Learning-based super node selection algorithm “Proof of artificial intelligence” (POAI) to save resources and speed up the block generation process. A method to intelligently select a “Minor” node, is provided, by following these steps. First, computing the ATN (Average Transaction Number), a function that serves as the primary criterion for choosing nodes, based on a comprehensive assessment of the average number of transactions a node receives per time period, the value of the participants by CNN. Then under a certain threshold, the process of selection is carried out to get the list of random nodes and super nodes. Finally, generating a pool of nodes and minors (random nodes and super nodes) based on their characteristics.

The work [105] suggests a new proof-of-work consensus based on machine learning. Just a fraction of miners does the hash calculation job. In general, PoW is used to nominate a leader, who may recommend the next block on the chain. However, in this method, two control mechanisms are used to limit the PoW mining race for the next block: A Dynamic Whitelist and Cryptographic Sortition.

Decentralized Artificial Intelligence enabled Blockchain Network (DAIBCN) [92] uses a decision tree architecture, so its algorithm picks miner nodes automatically. It is no longer necessary for miner nodes to calculate sophisticated hash functions in order to qualify for reward. In addition, data pruning is done on the blocks to allow for better storage and to minimize the chain length. The blocks mined every second raise the volume of data, as a result, the pruning process will be used frequently. This study clearly demonstrates that integrating AI in blockchain system has various benefits, including lower node energy consumption, reduced time, and storage savings, all of which contribute to a more efficient blockchain network.

The main contribution of Wang et al. [190] is finding the best permissionless blockchain mining approach based on a machine learning algorithm “reinforcement learning”. In this work, the consensus starts by choosing a leader who oversees packaging transactions, creating a block and appending this block to the blockchain. To avoid malicious nodes from gaining blockchain control, selecting a leader should be unpredictable. Because in permissionless blockchain, anonymity is intrinsically built as an objective. It should address the Sybil assault, in which an attacker generates members with different identities to maximize its chances to get selected as leader. The objective of using PoW is to randomly pick a leader from participants of each round with a probability proportional to each participant computer capacity to overcome the aforesaid difficulties.

Because PoW requires a lot of computing power and time, mobile devices are unfit to do it efficiently. Consequently, in the mobile blockchain network, edge computing paradigm was proposed by Nguyen et al. [128] to offload mining from mobile devices. The edge computing paradigm is composed of an edge computing service provider and mobile users, i.e., miners. Edge computing service provider owns resources for computing that are spread around the network to deliver computing resource services to mobile consumers. Because a single mobile user is allotted to a single edge computing resource unit, the mobile users compete for the unit. To finish block mining, mobile user is enticed to pay extra fees as much as he will continue using edge computing unit. Artificial intelligence contribution in this solution is introducing an ECSP [71] (Edge Computing Service Provider), auction based on neural network architecture. The neural network architecture used in this approach implements ECSP allocation and payment regulations; and, ensures that any auction mechanism discovered by the network is the best one.

A fork is an event in which a blockchain diverges into two separate paths, typically arising from a disagreement among miners regarding the consensus process. This divergence not only poses security risks but also escalates costs, time, and energy consumption. To mitigate these substantial risks and costs associated with blockchain forks, Mohammadi et al. study [120] employed machine learning techniques to forecast these events. The research evaluated the prediction accuracy of blockchain forks using four prominent machine learning methods: K-Nearest Neighbors, Naive Bayes, Decision Tree, and Multilayer Perceptron. By comparing these methods, the study aimed to determine the most effective approach for predicting forks, thereby enhancing blockchain security and efficiency.

References [43, 105] and [92] try to decrease latency and save resources by selecting super nodes in the network for mining. But they are centralized solutions at the mining level as it abandons the equity between the nodes in the network by having a whitelist for [105] and a list of super and random nodes for [43]. These solutions optimize the blockchain well but will be unable to handle a high blockchain throughput because they did not use all the computing power provided by the blockchain network solutions. [190] and [128] are interesting solutions as they increased blockchain throughput and latency performance for blockchains based on POW consensus protocol, but they failed to overcome the principle of using a lot of energy to finally get a useless result.

In the realm of complete governance and minimal reliance, there are solutions such as Neurochain aiming to realize the notion of an autonomous blockchain. This entails an approach governed by Artificial Intelligence, enabling a reduction in dependency on powerful nodes, as the need for complex hash puzzles to validate blocks is eliminated.

NeuroChain [48] utilizes bots, or software agents, that are present on every participant’s blockchain, creating a decision and communication ecosystem. Their algorithms introduce a novel consensus mechanism based on evaluating each member’s interactions within the network and the quality of their exchanges. By sharing and assessing this information, the network achieves a faster and more energy-efficient selection of the member responsible for confirming a block while simultaneously performing a beneficial task.

Many articles have prophesied that soon machines will take over and manage human life. That future is already here, or at least a chunk of it is. On the blockchain, smart contracts, a sort of language, have been used to negotiate contracts between users, determining who will be paid. Smart contracts, for the time being, are not all that “smart,” and are confined to only the most basic types of contracts. AI may soon be capable of dealing with increasingly complicated scenarios such as automating mortgage system, automating escrow processes, generating a person’s digital identity that cannot be counterfeited, and managing income tax return declarations to prevent any tampering with tax returns.

Therefore, having an autonomous blockchain backed by AI, intelligence is related to the evolution of smart contracts capabilities, so if we can implement all complex AI algorithms and protocol in smart contracts, blockchain which will be based on this smart contract and will be governed by an artificial intelligence entity being capable of performing tamper-proof and unbiased automated arbitration. All choices would be recorded on the blockchain, making them justifiable and transparent.

The increasing number of users and transactions on blockchain networks are putting a strain on the current blockchain solutions, and it is crucial that new and innovative solutions are developed to address this scalability issue. Furthermore, with the increasing adoption of blockchain technology in various industries, the demand for scalability solutions will only continue to grow. Without addressing this issue, the full potential of blockchain technology cannot be realized and its widespread adoption may be hindered. Today’s solutions for scalability in blockchain technology are still struggling to keep up with the rapid pace of growth. From these solutions, we can mention the pruning technique based on artificial intelligence for blocks with transactions that will not be used in future blocks. Future works should combine or add other techniques to the existing ones in order to achieve high scalability that can meet the needs of blockchain and that can keep up with the rapid growth of blockchain technology.

AI can help intelligently to deal with open source and open data for decision-making for changes made to it by checking or aggregating data from external sources to check their coherence and validity. Also, the widespread of using open data in the main existing systems and solutions around the world, set a new challenge of having intelligent open data. As blockchain-based data resources become more widely available, both businesses and consumers will require assistance with data accessibility, exploitation, and interpretation. AI’s capabilities make it ideal for such jobs.