Written by Patrick Reynolds*
Blockchain technology and artificial intelligence (AI) have both been major buzzwords over the past decade; whilst both are often discussed separately, few realise how effectively they complement each other. In classical programming, the programmer explicitly defines the rules (logic and algorithms) based on the data, and the program processes. But in AI, the paradigm is flipped: instead of coding rules, AI creates the rules. Alan Turing, a pioneering figure, significantly shaped the AI landscape with his 1950 paper, ‘Computing Machinery and Intelligence’, which posed the foundational question: “Can machines think?” Machines “think” by uncovering, by themselves, model patterns hidden in the data, enabling them to create their own rules to reach a proxy. However, there are growing issues with the security of the data and algorithms, with transparency in the decision-making processes and with the integrity of the data used — all of which can compromise effectiveness and trustworthiness. And where AI falls short, particularly in areas such as transparency, data integrity and security, blockchain excels by providing a decentralised, immutable and transparent framework that addresses these vulnerabilities effectively. It is why more research and startups exist now, both focusing on how blockchain technology can serve as the foundation for AI development in the next 10 years.
One of the main challenges in deploying AI systems is maintaining the integrity and reliability of the data used in the training and operation of AI. This challenge is particularly important because AI models rely heavily on the quality of input data; flawed data leads to flawed outputs, commonly referred to as the “garbage in, garbage out” problem. For example, Microsoft’s Tay AI chatbot, which was designed to learn from social media interactions, quickly began generating offensive content due to exposure to toxic data from users. One of the distinct features of blockchain technology is its immutable ledger, which prevents data from being altered or deleted once recorded. This characteristic could be essential for AI systems, as it helps maintain the consistency and reliability of training data, transactional data and other inputs. Blockchain also uses cryptographic hashing to further strengthen data integrity, which makes any attempts to modify the data detectable. Each block in a blockchain contains a cryptographic hash of the previous block, forming a chain that resists tampering and blockchains can use consensus protocols to verify the authenticity, accuracy and integrity of data before it’s used in AI models. This therefore reduces the likelihood of flawed data entering the system. In AI, consensus protocols ensure that data undergoes rigorous verification before being input into models, e.g. when new data is added to a blockchain, network nodes independently verify its integrity by checking cryptographic hashes against stored records. Only when a majority of nodes agree on the data’s validity is it accepted. This tamper-proof feature is certainly crucial for AI applications such as healthcare diagnostics where data integrity is essential for accurate results. For instance, blockchain-powered platforms can protect medical images from adversarial attacks, ensuring AI diagnostic tools provide accurate and trustworthy diagnoses. The integration of blockchain technology and AI also advances data from raw to structured information and actionable knowledge. Initially, accessing and indexing data across decentralised networks poses challenges, but companies such as the Graph have made blockchain data more searchable. Now, the focus is shifting to organising this data into coherent information for AI analysis. Tools, such as Nansen, bridge the gap between blockchain records and real-world identities, enabling useful information extraction. AI can automate the extraction and organisation of knowledge from large datasets, with projects such as OriginTrail developing decentralised knowledge graphs. However, integrating blockchain technology introduces computational overhead, slowing data processing and increasing costs and the immutable nature of a blockchain, whilst preventing tampering, can also make correcting errors difficult, potentially locking in flawed data — a challenge that must be addressed.
Auspiciously, blockchain technology could be the antithesis of Big Brother. The properties of a blockchain mean that AI models can use data and be deployed more securely. Firstly, the technique of cryptographic hashing in a blockchain protects against unauthorised access and tampering. Even the slightest modification to the data will result in a completely different hash, making it easy to detect any unauthorised changes. If someone tries to alter a block, the corresponding hash will no longer match and the network will reject the modified block so thwarting that particular cyber-attack. For example, during the deployment of AI models, hashing can be used to verify that the model’s code and data remain unchanged from their original state. If an attacker tries to modify the data or the model, the corresponding hash will no longer match, alerting the system to potential tampering. And as AI models are increasingly licensed or put in open source across platforms such as Hugging Face, then providing security is vital. Hashing guarantees that models haven’t been compromised during transmission or storage, preventing adversarial attacks where malicious actors might attempt to alter the model’s behaviour. Furthermore, public and private key cryptography secures transactions by controlling access. The private key, known only to the owner, is used to sign transactions so that only the legitimate owner can authorise transactions. In AI security, the private key, known only to the legitimate owner (such as the AI developer or data owner), is used to sign data, models or actions, ensuring that only authorised parties can make changes or access sensitive resources. This prevents unauthorised modifications and protects the integrity of AI models and data. Digital signatures add another layer to AI security by ensuring that transactions are both authentic and non-repudiable. As an important example, this can be used to track and control AIs such as lethal autonomous weapons systems, so that they operate only under authorised conditions. Smart contracts could also be applied to enforce strict operational protocols, reducing the risk of rogue deployments. This process ensures that the transaction is legitimate and prevents any unauthorised parties from forging transactions.
But, so what? Blockchain technologies provide solutions for data security and privacy. However, these solutions also introduce challenges that must be carefully navigated, particularly around scalability, security of smart contracts and balancing privacy with transparency. As the world becomes more digitalised, there is a growing need for enforcement of digital private property rights; we can establish clear ownership of various digital assets through non-fungible tokens (NFTs). NFTs can be used to represent ownership of anything from data inputs used to train AI models to the models themselves and even the outputs generated by those models. By tokenising these assets, creators and contributors can assert their rights over them, ensuring they are recognised and compensated appropriately. The immutable record-keeping of a blockchain further strengthens this by providing a verifiable trail of ownership, making it difficult for others to dispute or infringe upon these rights. Through the use of smart contracts, creators can automate and enforce access control mechanisms and these contracts can be programmed to grant or restrict access based on specific conditions, such as payment, geographical location or user credentials. This level of precision not only protects the creator’s intellectual property (IP) but also enables more flexible and efficient monetisation models, such as micropayments for individual uses or time-limited licenses. However, this approach requires the evolution of regulatory frameworks to formally recognise data ownership and establish mechanisms for effective monetisation.
Furthermore, blockchains can be used to track changes in AI models over time, including updates, retraining and adjustments. By recording every change made to a model on a blockchain, stakeholders can audit the model’s evolution, ensuring that modifications are properly documented and justified. This can be particularly useful for regulatory purposes, where it’s important to demonstrate that a model has been maintained and updated responsibly. But in terms of interpreting these models, it continues to be difficult. In sectors where AI decisions can significantly impact human lives, such as criminal justice or healthcare, the opacity of these models can lead to serious ethical concerns. For example, if an AI system used in a criminal justice setting makes decisions about sentencing or probation without clear reasoning, it can be difficult to determine whether the decisions are fair or biased. Indeed, publications such as the New Scientist are already posing the question: would an AI judge be able to efficiently dispense justice? However, the inability to provide explanations for AI-driven decisions can undermine trust in these systems and raise questions about who is responsible when things go wrong. Moreover, the lack of transparency complicates efforts to hold AI developers and operators accountable for the behaviour of their systems.
These problems are so pressing and clearly defined that Y Combinator, one of the world’s leading startup accelerators, is actively calling for more startups to focus on AI security, transparency and traceability. The need for innovation in these areas is undeniable, as the convergence of AI and blockchain technology presents both opportunities and challenges that require novel solutions. However, the path forward is not without difficulties. Developers ought to be encouraged to make security, traceability and transparency in the AI ecosystem foundational, not an add-on. And I think only blockchain can do that.
