SoK: A Systematic Bidirectional Literature Review of AI & DLT Convergence
This systematic bidirectional literature review (2020–2025) analyzes the convergence of AI and DLT across five architectural layers in both directions, revealing that current research is narrowly focused on specific layers, lacks production-scale validation, and requires cross-layer co-design to address fundamental challenges in scalability and interoperability.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine two powerful technologies trying to build a house together: Artificial Intelligence (AI) and Distributed Ledger Technology (DLT), which is the family of technologies behind blockchains.
- AI is like a brilliant, fast-thinking architect who can learn from blueprints and make decisions on the fly. However, this architect is often a "black box"—you don't always know how they made a decision, and they can be prone to mistakes or hidden biases.
- DLT is like an unbreakable, public notebook where every single brick laid is recorded forever. It's transparent and trustworthy, but it's slow, rigid, and bad at making complex decisions on its own.
This paper is a systematic review (a massive, organized search) of 53 research studies published between 2020 and 2025. The authors wanted to see how these two technologies are helping each other. They looked at the relationship in two directions: AI helping DLT and DLT helping AI.
Here is the breakdown of what they found, using simple analogies:
1. The Two-Way Street
The researchers organized their findings into two lanes of traffic:
Lane A: AI helping DLT (The Smart Architect fixes the Notebook)
In this direction, researchers are using AI to make blockchains faster, safer, and smarter.
- What they are doing: They are using AI to act like a security guard, a traffic cop, and a mechanic for the blockchain.
- Where it's happening: Most of the work is focused on the "Execution" layer (checking the rules of the smart contracts) and the "Consensus" layer (deciding who gets to write in the notebook).
- Analogy: Imagine AI scanning the public notebook to instantly spot if someone is trying to forge a page (security) or using a smart algorithm to decide the best order to write entries so the line moves faster (consensus).
- The Tools: They are using "Reinforcement Learning" (AI that learns by trial and error) to tune the system and "Large Language Models" (like the AI you chat with) to read and understand complex code.
- The Result: It works well in simulations, but no one has actually built a massive, real-world factory using this yet. It's mostly still in the "lab test" phase.
Lane B: DLT helping AI (The Notebook helps the Architect)
In this direction, researchers are using the unbreakable notebook to fix the problems of AI.
- What they are doing: They are using the notebook to prove where data came from, who trained the AI, and to make sure everyone playing together is being honest.
- Where it's happening: Most work is focused on the "Data" layer (making sure the ingredients for the AI are good) and the "Model" layer (managing how the AI learns).
- Analogy: Imagine a group of chefs (AI developers) trying to cook a meal together without a central boss. The notebook records exactly who added which ingredient and when. This stops anyone from sneaking in bad ingredients (data poisoning) and ensures everyone gets paid fairly for their contribution.
- The Tools: They are using "Smart Contracts" (automatic rules) to manage rewards and "Federated Learning" (training AI on many different computers without sharing the raw data).
- The Result: Again, this works in theory and small tests, but no one has deployed this at a massive, industrial scale yet.
2. The Big Imbalance (The "Skewed" Map)
The authors noticed that the research is very unbalanced, like a map where some cities are crowded and others are empty deserts.
- The Crowded Cities:
- For AI helping DLT: Everyone is obsessed with the "Consensus" (voting) and "Execution" (running the code) layers.
- For DLT helping AI: Everyone is obsessed with "Data" (ingredients) and "Model" (the recipe).
- The Empty Deserts:
- Very few people are looking at the "Network" (how computers talk to each other) or the "Infrastructure" (the actual hardware/power) layers.
- Analogy: It's like everyone is arguing about the best paint colors for the house (Application) and the best locks for the door (Security), but almost no one is checking if the foundation (Infrastructure) or the plumbing (Network) can actually support the weight of the house.
3. The Hard Truths (The Gaps)
Despite all the exciting ideas, the paper points out some serious problems:
- No Real-World Proof: Almost every study is a "prototype" or a "simulation." It's like building a perfect model car in a garage but never actually driving it on a highway. No one has proven this works at a massive, production scale.
- The "Black Box" Problem: AI is supposed to be smart, but if the AI makes a mistake, it's hard to prove why on a blockchain. The paper notes we still don't have a perfect way to make AI decisions "verifiable" (provable) on a blockchain.
- The Speed Bump: Blockchains are slow, and AI needs to be fast. Putting them together often makes things slower or more expensive.
- The Language Barrier: Different projects are using different rules and standards, making it hard for them to talk to each other (interoperability).
Summary
The paper concludes that while AI and DLT are a perfect match on paper (one is smart, the other is trustworthy), they are currently stuck in the "lab."
- AI is making blockchains smarter and safer.
- Blockchains are making AI more trustworthy and transparent.
However, we are still in the early days. We have great blueprints and small models, but we haven't built the skyscraper yet. The authors say we need to stop just talking about the theory and start testing these systems in the real world to see if they can handle the pressure.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.