AI challenges in fintech

Fintech will inevitably use AI extensively, but there are technological, regulatory, and educational concerns that need to be solved first. A number of factors will continue to drive up use while they are being remedied.

Financial businesses face particular hurdles as a result of society’s explosion in data generation, according to Shield VP of Data Science Shlomit Labin. Shield helps banks, trading groups, and other businesses keep an eye out for issues including employee behavior, market manipulation, and other compliance-related issues.

The increasing strain on compliance staff

According to Labin, financial services companies require technical support because the volume of their interactions exceeds what humans can handle. Recent changes in regulations make the issue worse. In the past, random sampling would have been sufficient, but not anymore.

“We need to put something in place, which presents more difficulties,” Labin stated. “Let’s assume I have to pick up one percent, or one-tenth of one percent, of the communications, so that something better than nothing is required. For any compliance team to analyze, I want to be confident that these are the excellent ones—the really high-risk ones.

“We observe directly and hear from our clients about the difficulties in handling and addressing these rapidly growing amounts of data,” stated Eric Robinson, vice president of KLDiscovery’s global advisory services and strategic client solutions. It’s not practicable or practical to use classic linear data management models anymore. Therefore, using AI in these processes—in whatever form—has become more essential than optional.

“It is no longer practical to try to do this linearly with manual document and data evaluation processes given the peculiarities of language and the sheer volumes of data.”

Robinson, a lawyer by profession, pointed to recent court rulings where judges chastised attorneys for utilizing AI in e-discovery and core litigation. Ignoring it is tantamount to malpractice since it exposes companies to fines for inadequate oversight, monitoring, or unsuitable procedures and frameworks.

AI is capable of addressing changing fraud trends

As technology advances, so too do measures taken to evade discovery, as noted by Robinson and Labin. Maybe a company should keep an eye on trader correspondence. One of the standard restrictions could be to not communicate on certain social networking sites. Taboo word and term lists are available for monitors to refer to.

Scammers may use code words and concealed phrases to confuse communications personnel. Compliance team warning fatigue results from combining that with older systems and larger data volumes.

That conclusion hasn’t, however, completely closed the door to technology. Since AI-based compliance solutions are still relatively new, skepticism extends beyond the judiciary. The press reports about judicial caution and artificial intelligence-generated case law are dubious.

It takes patience to watch AI technology develop

According to Labin and Robinson, societal attitudes and AI-based compliance solutions are always changing, just like any other technology. The caliber of the results increases. We are become increasingly acclimated to AI as it is used in more businesses.

According to Labin, “AI technology is becoming much more robust.” “I keep telling people that even though you don’t like AI, you look at your phone 100 times a day and expect it to open on its own because modern AI technologies are so advanced.”

“Compared to ten or fifteen years ago, the environment for technology acceptance is very different today,” Robinson continued. Predictive coding, latent semantic analysis, logistic regression, support vector machines, and other artificial intelligence components established the groundwork for numerous innovations employed by the legal sector from the very beginning of compliance.

Because of the quick advancements in technology, there has been a significant difference in the adoption rate. Things like natural language processing began to appear three or four years ago. These technologies are improved by them because they make use of context.

AI benefits and drawbacks from regulation

Pressures from regulations have proven to be both a blessing and a curse. It has become necessary for organizations, attorneys, and technologists to come up with solutions.

While the situation is changing, Robinson stated that outdated technology is insufficient. More is expected from regulators, which has paved the way for AI. The younger generation is more accustomed to it. It will assist when they take on positions of authority.

However, as AI is used in everything from big data analytics and discovery to contract lifecycle management, there are a lot of problems to be solved. Robinson mentioned three things: prejudice avoidance, confidentiality, and avoiding hallucinations (i.e., fabricated legal cases).

According to Robinson, compliance is crucial in this situation. When presented with proof that these AI tools are unreliable, some judges question how they can trust the information they are given. As generative AI becomes more integrated into these procedures, I believe that becomes a central discussion point.

How AI functions optimally

According to Labin, humanity can’t survive without AI anymore. It has made significant strides and continues to improve in areas like natural language comprehension.

However, it functions best when combined with human interaction and other technology. Even the most dubious cases can be handled by humans. One provider’s AI-based conclusions can be verified twice and three times using different tools.

Labin clarified, “You have to make sure that you use your AI in multiple ways to make it safer.” Additionally, while using several layers, there is no one approach provided to answer questions. You verify it against numerous models, systems, and checks to make sure you have covered everything and that you are not receiving rubbish.

Robinson went on, “One of the keys is that there’s no one technology.” “A combination of tools that enable us to perform the analysis, identification, and validation aspects is the effective solution. The question is how to combine these elements to produce a solution that is justifiable, practical, and efficient.

“It is best to monitor the model after the fact because it is already too big, too complex, and too advanced for me to ensure that it hasn’t picked up any bias,” Labin suggested.

Taking bias out of AI models

Eliminating bias (deliberate and unintentional) against low-income individuals and members of minority groups from systems is, according to Labin, a major problem. One cannot just enter raw data from previous judgments when there is evident bias against certain groups; doing so would only result in a more efficient discriminatory system.

Dedicatedly eliminate any data that can be used to easily identify categories that are at risk. The technology available today can already identify applicants based on their addresses and other data.

Is the answer a proprietary model designed for a single organization? Very unlikely. They took millions of dollars to build and require a lot of data to work well.

According to Labin, “you’re creating an inherent bias in the outcome if you don’t have a large enough data set because there isn’t enough information there.”

Promoting adherence

AI-based systems may make it impossible for compliance officers to comprehend the processes involved in evaluations and decision-making since they base their choices on intricate information patterns. Given the fragile regulatory faith in the technology, that raises legal and compliance concerns.

According to Labin, GenAI models can offer a “chain of thoughts” method in which a decision is asked to be broken down into manageable parts. Pose brief queries and analyze the answers to determine the general way of thinking.

According to Robinson, “validation and explainability are the core challenges.” “Once things are resolved, adoption will be much increased. Numerous AM Law 100 companies have fully embraced generative AI. Although they haven’t used it yet, they are stepping in to provide fixes.

“In the context of data and client information, a legal firm has serious concerns about privilege, confidentiality, and data protection. Until those issues are resolved in a way that allows for qualification and quantification… I believe adoption will get up steam once we’ve resolved the understanding, qualification, and quantification issues. It will also completely upend a lot of our customs.