In the financial industry, responsible AI implementation is essential for moral behavior, equity, and openness. Prioritizing data protection, addressing biases, ensuring explainability, and engaging in continuous monitoring are all imperative for financial organizations. By doing this, they promote sustainable growth, reduce risks, and establish trust.
I wish it were that simple. Many businesses are unprepared to emphasize ethical issues in their AI adoption activities, as demonstrated by a recent FICO research on the state of ethical AI practices in financial services.
Specifically, a study of more than 100 executives in the research revealed shocking results: 43% of firms find it difficult to ensure that their AI governance structures and procedures comply with regulatory standards, even though 52% of respondents say they are a higher priority than they were a year ago.
AI systems rely significantly on human-generated or gathered data. This data may come from online platforms, user-generated material, or information gathered by sensors and other devices. Intentional or inadvertent human biases might infiltrate these AI systems through the data they use, leading to unforeseen and catastrophic outcomes under unhygienic data governance procedures.
Robust data protection protocols, objective decision-making, transparent justifications, frequent audits, and adherence to ethical norms are all components of responsible AI. It is difficult, though, and necessitates a thorough foundation and careful adherence to rules.
In the examination of their talk that follows, we look at two important findings:
Important factors for ethical AI: Taking into account the ethics, explainability, robustness, and audibility of AI systems throughout the early stages of adoption.
How to use AI in an ethical manner: Implementing ethical AI involves surveying the use of AI, creating criteria for specific approaches, and adhering to a model development governance standard.
Below, you can hear the entire episode:
Important Things To Think About In Ethical AI
Only 8% of businesses have reached AI maturity, according to a poll cited by Scott in the recent FICO report “State of Responsible AI in Financial Services.” He proposes that there may be two reasons behind this:
Absence of a unified, standardized strategy for ethical AI in the financial services industry.
The need that a single standard be used by the entire company.
The constant barrage of AI jargon takes focus away from developing reliable, ethical, auditable, and explainable models.
Scott highlights the necessity of unifying AI governance under a single, unambiguous framework:
Not many playbooks exist for this. I believe that financial services firms understand some aspects, but they really need to unite around a standard that would be used by all of the financial services organizations, which requires effort. Numerous models are being developed, and there are numerous differing viewpoints. Therefore, the organization will adopt and build their AI around this one model development governance standard, which is quite challenging. It has a tremendously transforming effect.
— Scott Zoldi, FICO’s Chief Analytics Officer
Scott adds that he has noticed that a lot of financial institutions claim that the board and C-level executives must comprehend the significance of responsible AI. He believes that these leaders want to invest in deep learning and try new things, but doing so calls for ethical consideration of responsible AI.
According to the report, 44% of the CEOs said that board-level definitions of responsible AI initiatives were still necessary. Former IBM Global Chief AI Officer and Trustwise AI CEO and Board Member Dr. Seth Dorbin can personally vouch for the report’s conclusions. He informs Emerj that he won’t stop there:
I concur, and I would even go so far as to say that very few are set up at the board level, and that only a small number recognize the significance or consequences of their ignorance. For the majority of boards, it is a major blind spot. Having said that, I know of several private equity firms who are pressuring their portfolio companies to have this discussion at the board level.
— Former IBM Global Chief AI Officer Seth Dobrin, CEO and Board Member of Trustwise AI
Zoldi takes advantage of the occasion to discuss the components of a responsible AI strategy that is established by the board. He lists the four essential components for creating morally and responsibly designed AI models:
Model Robustness: In a production setting where data may fluctuate and behaviors may alter, models need to be resilient and able to operate. To guarantee stability and dependability while creating strong models, extensive testing and validation are necessary.
Scott notes that some people employ automatic machine learning (ML) or run some data through an open-source trainer, and they label what they’ve created a model overnight. Rather, even for the teams creating them, it requires a lot more effort to truly comprehend models that affect people:
Model Explainability: To ensure that the decision-making process is clear and transparent, models need to be comprehensible and interpretable. Understanding what the model is learning and what influences those decisions is crucial.
Explainable AI algorithms, according to Scott, are good, awful, and ugly since each model yields a unique set of results. However, it is crucial that the data scientist, a governance team, and a regulator examine the factors that influence the model and inquire about the decision’s dependability.
Ethical Models: Models must treat people equally while taking into account how their actions may affect various customer groups.
According to Scott, the model’s choice will affect some customer types more than others when it comes to the overall amount owed on the mortgage or the number of late payments. He thinks the model should be aware of this ethical dilemma.
Models must have a transparent method for responsible development and monitoring in a production setting, and they must be auditable. According to him, developing morally and responsibly in laboratories is only important once it begins to affect people.
He therefore recommends putting in place a monitoring system so that we can identify instances in which a model that passed all development checks performs differently than we had expected.
How To Put Ethical AI Into Practice?
Businesses must establish the ethics of their organization and its solutions before using ethical AI.
Beena Ammanath, Executive Director of the Global Deloitte AI Institute, stated in another edition of the AI in Business podcast that there has been a belief that AI ethics is all about being transparent, eliminating bias, and making it more equitable. She believes those are just attention-grabbing headlines:
Fairness, prejudice, and openness are essential, based on my experience working in a variety of businesses. However, there are more considerations. Fairness isn’t really relevant, for instance, if you have an algorithm that predicts a manufacturing machine failure. However, safety and security are equally important concerns.
— Beena Ammanath, Executive Director of the Global Deloitte AI Institute
Zoldi suggests the following three-step structure for Emerj’s executive audience to follow in order to establish internal AI capabilities within an organization:
Examine the methods used now.
Either an ethics officer or a chief analytics officer should create a set of rules that highlight some of the strategies the company will use.
Aligning with a model development governance standard that all upcoming AI and machine learning activities will adhere to will help to mature present processes.
The company can more effectively concentrate its efforts and direct the attention of its data science teams to a single issue by adopting a single strategy like this.
He goes on to discuss the pandemic’s effects on financial services institutions and the growing significance of regulating AI. Many customers have switched to digital interactions with financial services companies during the epidemic, and it is crucial to be able to tailor those interactions according to each person’s financial background. But as AI becomes more widely used, regulations are also becoming more necessary, and the US needs to catch up to Europe’s GDPR and high-risk AI criteria.
Recent instances of heightened attention to AI policy include the US AI Bill of Rights and the executive order on improving racial justice. Customers are growing more aware of how AI affects their choices and are growing more concerned about how AI should be used, including data privacy and anti-discrimination measures.
Businesses that set themselves apart by utilizing AI effectively and addressing customer worries about algorithm protections and data privacy will have greater success.
According to Scott, raising businesses’ awareness of the possible risks and hazards associated with AI is the first step in teaching them about its significance. Numerous sophisticated organizations have already made grave errors in the design of their systems, leading to the reporting of inaccurate, offensive, prejudiced, and inappropriate information. Companies must consider if they want to be in these situations and understand the possible repercussions of acting irresponsibly.
Reminding businesses that AI models are merely tools and that they have the opportunity to select more socially conscious solutions that put safety, justice, and ethics ahead of predictive value alone is also crucial. When implementing AI, an organization should make ethics, interpretability, and explainability its top priorities.