How can companies make AI explainable?

explainable

There is a key topic that banks spanning the world are asking right now: how are they able to make AI explainable?

This was the opinion of Wolfgang Berner, the CTO of RegTech firm Hawk: AI, who recently presented a keynote speech on the above topic.

Berner remarked, “In heavily regulated areas such as combating money laundering, considerations as to how transparent and comprehensible the use of artificial intelligence is are entirely appropriate. Classic concerns about such a “black box AI” arise in particular when the decisions of the AI ​​are too disconnected from the original data and when there is no transparency about the way the algorithms work.”

Hawk AI sees the key to trust and acceptance in the compliance industry in the high level of transparency. For this reason, for the company, the need for AI explainability goes far beyond purely regulatory requirements.

With understandable AI, financial institutions have an overview and control, even over complex models such as neural networks. For Hawk AI, explainability is made up of two areas – what is the justification for an AI-driven, individual decision? And how were the algorithms that contribute to AI developed?

Hawk AI said, “For Hawk AI it is clear – only what can be explained technically will be accepted in the end. The exact criteria for a decision or the statistical probability of certain risks and the composition of the algorithms are just as important as complete documentation of the AI ​​decision-making process. It is also essential that all of this is expressed in clearly understandable language and not in purely technical jargon.”

The company believes every detail and every data source must be verifiable – for example, whether certain values are conspicuously high or low compared to a specific peer group. It must be transparent why the AI ​​assumes certain expected values ​​and how values ​​correlate with each other. The data situation must be so clear that a compliance officer would come to the same decision with the same data.

Furthermore, consistent feedback and validation processes help to continuously improve decisions – the AI, therefore, learns directly from the decisions of the compliance team and can support them better in future.

Hawk mentioned that the AI ​​does not only have to be transparent at the beginning of its application – since it improves independently through contact with new data, it is also necessary to be able to understand such optimizations. For this reason, the firm claims every change process of the AI ​​is also recorded in the software (on the so-called pipeline) and requires explicit approval. So the AI ​​never evolves without the compliance team being able to understand and control it.

Hawk AI concluded, “AI anti-money laundering is ready for action – with Hawk AI it is transparent and secure. For these reasons, Hawk AI speaks of a “white box AI” in connection with artificial intelligence, whose technology is completely understandable for the compliance team in contrast to the “black box AI”. Our software therefore offers complete control and therefore security. The application of artificial intelligence in the financial sector is revolutionizing the fight against financial crime.

“Technology-driven anti-money laundering not only clearly surpasses traditional systems in terms of efficiency and effectiveness, but is also particularly future-proof thanks to its ability to learn from criminal behavior patterns. The use of AI against financial crime will thus become the industry standard in the long term. The technology has been proven in practice for years. Even in very large financial institutions, it is already in use today or at least established in the first pilots.”

Hawk AI has partnered with Diebold Nixdorf, a leader in driving connected commerce for finance and retail, to expand the reach of the former’s AML solution.

Copyright © 2022 FinTech Global

Enjoying the stories?

Subscribe to our daily FinTech newsletter and get the latest industry news & research

Investors

The following investor(s) were tagged in this article.