MALVERN, Pa. — As people increasingly use artificial intelligence (AI) in various areas of life to either save time or improve performance, the influence of AI systems grows — as does the risk of untrustworthy results in high stakes decision-making, like those involving medical or financial choices. AI agents, such as autonomous trading bots or various diagnostic tools, are designed to learn, evolve and operate with little or no direct human involvement. So, what can technologists do to help ensure AI complies with ethical values for the public good?
A team, including researchers from Penn State, published a paper in the Journal of Cybersecurity and Privacy to explore ways that blockchain — the technology behind cryptocurrency — could help enforce ethical boundaries for AI systems.
In the Q&A below, study co-author and Penn State Great Valley Assistant Professor of Software Engineering Dusan Ramljak discussed the framework he and his team developed to investigate how to use blockchain for AI alignment with human ethics.
Q: Why is the behavior of AI such a pressing issue?
Ramljak: AI systems are trained to act in certain ways within specific parameters. If the parameters shift, such as when they are implemented in broader systems, they can act in unpredictable ways that contradict human values — for example, misrepresentation by mishandling sensitive health data or misinformation by recommending the wrong medical treatment. The consequences can be dangerous.
As we delegate more and more tasks to AI, we need to find ways to ensure that we maintain control over these systems and facilitate ethical conduct. AI can help us increase the efficiency of everyday life, but we need to use it responsibly by providing a proper testing environment, predicting its behavior beforehand and ensuring AI alignment with human values.
Q: People often think of blockchain in connection with cryptocurrency. How can blockchain help AI function ethically?
Ramljak: Blockchain creates secure, decentralized digital records that through cryptography, requiring transparent data lineage from inputs to outputs, with tamper resistant logs. Many AI systems are missing the kind of transparency, security and decentralized control that blockchain offers. By pairing AI with blockchain, we can build systems where decisions are consistent and traceable, and operational rules and data are harder to manipulate.
Blockchain-based consensus mechanisms allow a broad set of participants to vote on ethical rules, which are then stored as “smart contracts.” Smart contracts are machine-enforceable ethics, or rules that execute automatically. These mechanisms promote decentralized collaboration and democratization in the development of artificial intelligence systems and ensure that machines respect certain boundaries. Blockchain’s security features also make it harder for malicious actors to manipulate training data that could cause AI systems to give untrustworthy or harmful results.
Q: What is the framework your team developed to research blockchain’s uses for putting ethical guardrails on the behavior of AI systems?
Ramljak: Our research team — which included Penn State Abington Assistant Professor of Information Technology and Cybersecurity Analytics and Operations Maryam Roshanaei, master of data analytics student Jainil Kakka, Vienna University doctoral students Alexander Neulinger and Lukas Sparer, as well as University of Kragujevac Faculty of Science and Department of Mathematics and Informatics doctoral student Dragutin Ostojić — reviewed more than 7,000 studies and developed a framework to guide future research on how blockchain can facilitate ethical compliance in AI systems by supporting secure, transparent operations and providing accountability.
One study suggested blockchain could be used to verify AI input data and record AI outputs, which would help ensure traceable decision-making. Another study proposed that blockchain could support reputation systems for AI agents, offering incentives for ethical behavior and denial of resources for misleading or non-compliant actions.
We also noted some gaps in the literature, so we proposed a three-tiered analytical framework for further exploration of blockchain’s uses for AI alignment:
- Micro-level analysis: Technical and coding protocols to secure user data
- Meso-level analysis: Use cases for blockchain-based AI in various sectors, such as financial services and health care, with requirements specific to the unique needs of those sectors
- Macro-level analysis: Moral principles and international legal and cultural standards that govern ethical, sustainable uses of emerging technologies in the public interest. The European Union’s Artificial Intelligence Act is an example of a global regulation intended to support the trustworthy development and responsible use of AI systems through strong governance and transparency requirements.
Q: What are the downsides of using blockchain-based AI?
Ramljak: Large-scale blockchain and AI systems can be energy intensive. Also, privacy needs to be safeguarded so that making AI more transparent doesn’t mean exposing sensitive data. And different countries and cultural regions may prioritize values differently, from privacy to equity. So, alignment frameworks need to be flexible enough to work across cultures and legal systems.
Q: What future research is needed to explore blockchain-based AI alignment with ethics?
Ramljak: For future research directions, we need to develop multi-layered frameworks that incorporate micro-level security and privacy mechanisms, meso-level sectoral governance and macro-level legal constraints into a single, adaptable structure.
The convergence of blockchain and AI could give us not only enhanced reliability of technical systems but also new ways of human-centered governance, ensuring accountability at scale. Blockchain offers us tools to develop and use AI in ways that prioritize human values and societal well-being.