Opening up AI's 'black box' with blockchain technology

The two most hyped technologies in recent years – blockchain and artificial intelligence – may solve for each other’s greatest drawbacks.

Blockchain technology, which in layman’s terms is a registry of data or transactions, could be the answer to how different industries, including home finance, track and keep records of its machine learning or AI decisioning. The possibilities of how the two technologies can mesh with one another is being actively discussed in technology circles, especially in the financial sector, industry stakeholders say. 

Some predict fairly soon technology companies using AI will look to the online ledger to outline their methodologies and data usage, creating “snapshots of data” visible to those with access. 

This pairing is now getting attention because “in the last two to three months, the whole discussion has shifted towards transparency, especially because of ChatGPT,” said Subodha Kumar, professor of data science at Temple University.

Grouping the technologies could answer the top-of-mind question of how to make the actions of AI more explainable by potentially opening up the “black box,” about which the Consumer Financial Protection Bureau has issued several warnings.

“Companies must take responsibility for the use of these tools,” CFPB director Rohit Chopra said in an interagency statement co-released with the Civil Rights Division of the Department of Justice, Federal Trade Commission and the Equal Employment Opportunity Commission in April. “Unchecked AI poses threats to fairness and our civil rights.”

But blockchain is not a panacea. Industry experts warn that the tech will not solve ethical concerns around machine learning and AI.

Technologists in the mortgage industry point out the inherent compatibility of blockchain and AI make them destined to come together, sooner rather than later.

Both have positives that can complement one another, said Prabhakar Bhogaraju [PB], the vice president of financial fitness app Finlocker.

“AI’s strength is finding abnormal patterns of what [may at first] look like normal data and it does this by recognition,” said PB. “Blockchain on the other hand has some strengths on data provenance, data lineage and absolute control on data, which itself is immutable.” 

He noted that “AI can help in optimizing how blockchain resources themselves are used [and] blockchain can help AI in providing more traceability of the data sets on the model iterations themselves.”

“But that’s still, in my opinion, very early stage matchmaking,” PB added.

Being able to view what is saved on blockchain, with the proper approval, can also be the key to battling fraud, industry stakeholders say.

“If it’s your IP nobody can touch it or change it and any variant of your IP will actually look like a variant 100% of the time,” said PB. “So blockchain can help fill one of the gaps that AI has, which is this deep-fake kind of problem.”

Blockchain could be effective for ensuring security in the property valuation space, Leah Price, co-chair of MISMO’s Emerging Technology community and independent advisor to fintechs, predicts.

“With the proliferation of deep fakes, how is anyone able to tell what is true?” Price said. “If you think of a property valuation and there are photos of a property, how exactly do you know that the photos haven’t been manipulated? It’s very easy to get a photo of a house and cover up potential problem areas.”

“But with an appraisal, you could put a hash and a timestamp on the photos, the data and documents to make sure that they’re authentic and haven’t been altered,” she added. “Blockchain has been designed in a way that you can always tell what’s true and authentic.”

The ledger technology may also have an impact on disclosures of what data AI systems are analyzing and relying on to make their lending decisions. 

Per Subodha, participants on the blockchain “can see how the data collection has been transformed.” 

“When a system is built out on the blockchain it allows different people to share information like who is putting something in and why and these types of things can be checked when somebody’s trying to mess with the system,” he said.

“People can see what the data [used to build a system] actually looked like and [from that] people can see the data itself was biased and some correction can be made based on that,” Subodha added.

The transparency provided by the blockchain can potentially allow permissioned industry participants to see what went into the underwriting of a loan, added Price.

“You could take all the inputs, you can have hashes on them and as a loan is originated, that data can be put on the chain,” Price said. “As the loan is sold into the secondary market, various participants would have access to that loan and see that a borrower’s credit report is authentic and that their income report is authentic, so that’s the kind of transparency that a blockchain would bring.”

But before blockchain and AI come together, some kinks need ironing out.

For example, when you create a see-through box, where all of your data is accessible to a third party, you also amplify the risk of fraud, technologists interviewed say.

“Blockchain helps you in creating more security, but at the same time, the blockchain also creates transparency, but with more transparency there is a chance people can misuse the data,” said Subodha. “So that’s going to be a very interesting challenge going forward for the financial industry…” 

Apart from potential fraud scenarios, being more open with data usage and how a company builds its internal code can also reveal a company’s well-guarded trade secrets, which could be a barrier for implementation in the financial sector, some warn.

Brett Brumley, CEO of Lender Toolkit, noted that if the two technologies come together “there needs to be a delicate balance between transparency and protecting a company’s proprietary information.” 

“Disclosing the exact coded variables could reveal too much about how the system works and remove competitive advantages,” said Brumley. “At the same time, too much information could actually put mortgages at risk. If you post too much information to a blockchain, you can actually make investors very nervous by sharing normal changes that are expected in the manufacturing process of a loan.”

Last but not least – though not necessarily a barrier – blockchain will not solve the problem of bias, which Rohit Chopra, director of the CFPB warns “poses threats to fairness and our civil rights.”

“Blockchain in itself is not going to make the data better or algorithms better,” said Subodha. “It cannot and does not have the capability to do that, but it creates more transparency, and more involvement of people can magnify the understanding of results.”

Going forward, the ledger technology will not be the only tool used by the financial sector to disclose their use of AI. It is believed regulators will roll out guidelines in the near future.

“[It is uncertain whether] every AI will require a blockchain kind of solution, or model provenance, but I do think blockchain definitely should be a part of the solution kit,” PB said. “One of the business constraints in these two technologies moving faster than they are is we don’t have the right regulatory governance yet.”