While developers debate the likelihood of an AI apocalypse and regulators warn about the threats cryptocurrencies pose to the financial system, blockchain technology and digital intelligence are already being applied in medicine and scientific research.
ForkLog has compiled some of the most interesting projects that demonstrate how distributed databases and AI help solve practical problems far beyond the financial sector.
Private Data Transfer for Intensive Care
In intensive care units, predicting in-hospital mortality plays a crucial role, allowing doctors to initiate aggressive treatment earlier and allocate limited resources more effectively. AI models are already being used for this task, but many are trained on data from a single clinic.
This means that when applied to other medical institutions, the AI loses accuracy due to significant differences in demographics, documentation protocols, and clinical practices. Collaborative training on combined data from multiple sources could address this issue, but HIPAA and GDPR strictly prohibit the transfer of "raw" medical records between institutions.
Federated Learning (FL) theoretically circumvents this restriction. Hospitals exchange not patient data, but only updates of the neural network's gradients.
However, existing FL systems still face four unresolved issues: the lack of cryptographic guarantees of patient consent, the inability to verify the integrity of sent updates, vulnerability to malicious attacks, and the absence of privacy expenditure audits. The platform BlockFedMed aims to address all these problems at once.
This blockchain-orchestrated AI system is built on Hyperledger Fabric v2.5 and employs a bidirectional LSTM model with Gaussian differential privacy.
Instead of relying on the honesty of hospital management, BlockFedMed transforms privacy compliance into a cryptographically verifiable audit trail. Three smart contracts manage verifiable consent, update integrity, and participant incentives.
The built-in FedMed-Bft aggregator is resilient to Byzantine attacks—it only accepts updates that pass the integrity contract check.
Training was conducted on the MIMIC-IV database (which includes 52,167 ICU admissions), with external validation on a fully independent eICU dataset (200,859 patient records from 208 hospitals worldwide).
During external validation of in-hospital mortality, the solution achieved an AUROC score of 0.841—7.4 points higher than local models. This score is only 3.1 points shy of the theoretical maximum achieved through centralized training, which is prohibited.
According to an independent study, BlockFedMed reduced the time taken to obtain patient consent for data transfer by 71%.
To date, hospitals worldwide have amassed a significant amount of digital records that are ideal for training AI, but sharing them is legally prohibited. BlockFedMed enables hundreds of clinics to collaboratively train a powerful neural network for predicting mortality in ICUs, with blockchain used to verify the transmitted information.
Currently, this is an independent project aimed at demonstrating the feasibility of federated learning. However, the success of this research could pave the way for similar solutions.
Verifiable AI in Pharmacogenomics
In personalized medicine, AI models are already effectively predicting which drug will be effective for a specific patient based on their genome. However, doctors and regulators cannot independently verify whether the prediction was honestly derived from the model and correct data or if it was altered afterward. In medical solutions where the cost of error is a patient's life, this is unacceptable.
Researchers from the University of Southern Denmark proposed a decentralized model that integrates AI prediction with blockchain verification, ensuring the integrity, traceability, and reproducibility of results.
The system calculates the prediction, then generates cryptographic hashes of the model's input data and the data itself, which are then immutably stored on the blockchain via a smart contract.
The model employs deterministic tokenization and canonical hashing, linking each "input data - prediction" pair into a verifiable on-chain commit.
Any stakeholder—doctor, regulator, insurer, or patient—can independently confirm that the prediction was indeed derived from a known model and a reliable source. Meanwhile, the genomic data itself is not disclosed; only the signature is verified to maintain privacy.
Experiments on the GDSCv2 dataset (genomic data on drug sensitivity) showed that the Random Forest Regressor model achieved a coefficient of determination of 0.979 for predicting drug sensitivity. Five-fold cross-validation yielded a value of 0.977, confirming high reliability and generalization ability (with a maximum of 1).
A traceability integrity test on 10 samples confirmed verifiability and transparency at a level of up to 70%. This is the first step toward making AI in medicine not only accurate but also a legally trusted tool that combines machine learning with regulatory transparency requirements.
Blockchain Carbon Markets
“Blue” carbon ecosystems—mangroves, salt marshes, and seagrasses—absorb CO₂ five to ten times more effectively than the Amazon rainforest. They are critical for mitigating climate change, yet their participation in carbon markets remains limited.
The reasons include fragmented monitoring, reporting, and verification (MRV) practices, insufficient auditability, and high costs of field measurements. The projects themselves suffer from opacity, double counting, and greenwashing.
Researchers from Shanghai International Studies University, Changzhou University, and Foote Technology presented an architecture that changes the approach to carbon data. Instead of collecting reports annually, the system makes monitoring continuous.
IoT sensors (electricity meters, industrial sensors, continuous emissions monitoring systems) continuously transmit environmental information. Meanwhile, blockchain on Hyperledger Fabric ensures the immutability and traceability of each event.
Source: Frontiersin.Thus, carbon data cease to be mere numbers in a report and become a continuously managed digital asset.
The unified Carbon Data Interface Standard (CDIS) ensures the readability of information from any source. The governance of the solution is based on a DAO principle, where enterprises, verifiers, and regulators collaboratively establish rules.
The Dynamic Authority Selection Mechanism distributes the right to participate in consensus based on the verifiable effectiveness of the participant—meaning the higher the quality of the data, the greater the influence.
The architecture also allows for access to public blockchains like Ethereum through a tokenization layer, enabling the creation of “carbon RWAs” and opening opportunities for CO₂ crediting.
Source: Frontiersin.Designing Photonic Crystals
Photonic crystals are artificial materials with a periodic structure that control the movement of photons at the nanometer level. They are essential for optical computers, ultra-sensitive biosensors, and quantum devices.
The challenge is that manually designing such structures is nearly impossible. The sub-wavelength scale of photonic crystals does not allow for the use of analytical models from geometric or wave optics. Traditional design methods either rely on ready-made libraries with a limited set of configurations or require resource-intensive computational simulations.
To accelerate the process, researchers at Tsinghua University developed a mechanism called AI-generated photonics (AIGP) based on latent diffusion models. Unlike traditional methods that require iterative simulations at each step, AIGP directly maps optical properties to structures.
Specified parameters serve as prompts (e.g., “only allow light with a wavelength of 1550 nm”), based on which a photonic crystal is created.
Simulations and experiments confirmed that this method generates functional structures with high accuracy, adhering to specified parameters—transmission power, phase, and polarization characteristics.
How AIGP Works. Source: LightAM.Blockchain in this model addresses an equally important privacy issue. Laboratories want to share and exchange AI training data but do not want to disclose their technology's operational principles to competitors. A distributed ledger allows them to exchange model updates while keeping proprietary approaches confidential.
Each AIGP update undergoes verification, and information about it is recorded in the distributed ledger, ensuring transparency of the change history for network participants.
Photonic crystals are foundational for the next generation of AI chips, which consume orders of magnitude less energy and operate faster than electronic ones. Therefore, this research is not merely an experiment but a first step toward deeper exploration of nanophotonics.
Federated Charging for Electric Vehicles
The mass transition to electric vehicles creates three interconnected challenges for energy systems. The first is peak loads: when everyone returns from work and simultaneously plugs in their cars, the grid becomes overloaded.
The second is privacy: commercial charging management systems can see who is charging, where, and when, creating risks of personal data leakage. The third is security and trust in centralized solutions.
The PETAL-Grid project (Privacy-preserving Edge-Trust-Adaptive Learning Framework) aims to address these issues using an AI framework based on federated blockchain for adaptive and private coordination of energy consumption in smart grids.
The system integrates federated AI (collaborative learning without transferring raw data), intelligence on edge devices (which predict local demand in real-time), and blockchain-based trust management (transparent and tamper-proof transactions via smart contracts).
Source: Nature.The PETAL-Grid workflow consists of five stages:
- Local data collection from charging stations.
- Demand forecasting on edge devices.
- Federated model aggregation (exchanging generalized forecasts rather than raw data).
- Adaptive load coordination.
- Blockchain validation of transactions.
The distributed ledger serves a dual purpose. First, it ensures the immutability and transparency of all energy transactions, and second, through smart contracts, it automates calculations and priority distribution.
Modeling showed that PETAL-Grid reduces peak load on the grid by 18% and increases energy efficiency by 17%. Transaction security reached 98-99%—significantly higher than centralized and baseline models.
The proposed system simultaneously addresses all three challenges of mass electrification: it alleviates the load on energy grids, protects drivers' privacy, and provides cryptographic guarantees of transaction integrity. Moreover, the model is decentralized and has no single point of failure.
***
New technologies always attract various speculators. Often, the informational noise they create obscures real benefits behind financial gain.
In such an environment, it is essential to remember that not all developments are necessarily monetizable; some should simply work for the benefit of humanity.