The hidden bias in financial AI—can blockchain solve it?

Artificial intelligence has become a powerful force in the finance ecosystem, offering faster, data-driven insights that promise to improve investments, lending, and risk management. From AI advisors that personalize financial strategies for both companies and individuals to highly advanced trading systems that make data-driven decisions in microseconds, the financial AI sector has a lot of room to grow.

But theres one major problem: bias.

Despite offering speed, precision, and what seems like objectivity, financial AI systems carry the same bias that the industry has been trying to eliminate for decades. For example, according to Lehigh University, OpenAIs GPT-4 Turbo large language models—simulating an AI mortgage advisor or decision system—required certain demographics of applicants to have 120 credit points higher than white applicants to receive the same approval despite having the same income, credit history, and debt levels.

This bias doesnt just affect the traditional financial markets but also the decentralized finance and crypto ecosystems. Take AI-powered market forecasting platforms, for example. Since their data is based on price history, news sentiment, or social trends, these platforms could sometimes overreact to market anomalies—crypto is full of black swan events like the Terra collapse, FTX crash, or large penalties from the regulators.

Consequently, these prediction tools can become over-aggressive or even overweight social trends and chatter, leading to poor signals and predictions.

Blockchain, XAI to the rescue

The limitations and opaque nature of many AI systems prevent them from becoming fully transparent and accountable. Some even call them black boxes since AI models usually have little to no transparency.

Notably, the decisions made by AI tools within the crypto space are not usually explainable—this makes it hard for users to understand how decisions are made. The absence of standardized auditing protocols for AI systems would also result in inconsistent assessments and potential oversight of critical issues.

Integrating blockchain technology with Explainable AI, or XAI for short, can tackle this issue by providing the immutability and transparency that come with decentralized ledgers—potentially improving the auditing methods as well since the auditors will have complete access to the platforms data and underlying algorithms.

XAI models are already getting increased attention since they ensure the decision-making process is fair and ethical in addition to being efficient. Blockchain technology can complement XAIs fairness by creating immutable records of AI decision-making processes, ensuring that every action is traceable and verifiable. This will promote trust and accountability.

Blockchains operate in a trustless manner. This doesn‘t mean the technology cannot be trusted, but it suggests that third parties or central authorities won’t be needed to confirm any decisions. Decentralization removes the need for a centralized entity to oversee the processes, thanks to the smart contracts that function autonomously.

When a model changes or outputs a decision, the lack of logs and version control can cause trust issues with most of the AI platforms. Blockchain technology timestamps the records and data on an immutable ledger.

FICO, a credit scoring company, has used blockchain to log AI model decisions, so regulators can trace how decisions like credit approvals were made. The company received the “Tech of the Future—Blockchain and Tokenisation” award at the Banking Tech Awards in London last year.

From theory to practice

Blockchains and decentralized finance protocols have the opportunity to bake fairness, transparency, and accountability into AI models—something traditional financial companies have been struggling with.

Combining XAI with on-chain verification can transform how decisions are made and trusted in the web3 ecosystem. For example, using XAI to explain the voting of decentralized autonomous organizations could help users have a better understanding of the consequences of their choices. A more advanced utility would be using XAI for risk assessment in lending DeFi protocols.

Mixing XAI with blockchain technology could also be a powerful on-chain surveillance and manipulation detection tool. AI is good at analyzing patterns of sandwich attacks, MEV exploitation, or wash trading. This could help in finding market anomalies.

Some web3 projects are already trying to enhance AI transparency. SingularityNET, for instance, focuses on making AI processes auditable. Another platform called Ocean Protocol tracks the origins of the data, ensuring trustworthiness and traceability.

Conclusion

At this point, it‘s just the beginning of the integration of blockchain and AI. Researchers are now exploring hybrid models that combine blockchain’s integrity, XAIs clarity, and bias-detection tools into systems that can monitor and potentially correct themselves.

But technology alone won‘t fix this. It will also need attention from regulators, scrutiny from users, and humility from the developers building these systems. If the 2008 financial crisis taught us anything, it’s that blind trust in complex and centralized tools is dangerous.