AI Revolution: Overcoming Challenges in Regulated Industries

Highly regulated industries like healthcare and finance face unique challenges in AI implementation due to compliance concerns and opaque algorithms. Juan Jaysingh, president and CEO of Zingtree, explores how to overcome these hurdles through transparency, explainable AI and defined guardrails.

Everybody’s talking about AI. Tech companies are touting their latest innovations with clear AI labels, but AI will soon be implicitly woven into every platform. With three-quarters of companiesOpens a new window using or exploring this technology, there’s no doubt it will become a business imperative. 

Highly regulated industries, such as healthcare and finance, face significant hurdles in implementing AI. These organizations must find a way to leverage this technology while maintaining compliance. Successful AI implementation centers on privacy, transparency and trust. 

The 5 Challenges of AI Implementation

As organizations delve into the potential of AI in regulated industries, it has become essential to recognize the obstacles that arise. Here are 5 significant hurdles to AI implementation:

1. Risk aversion

Regulated industries are inherently risk-averse due to the high stakes involved in their operations. Patient safety and well-being are on the line in healthcare; finance and insurance companies play a major role in people’s livelihoods. Mistakes come at an enormous cost. 

2. Lack of transparency

In its current form, AI poses a variety of risks. Many AI algorithms, including large language models (LLMs), lack sufficient controllability and guardrails. Their opaque inner workings and decision-making processes make explaining how the system arrived at a particular output challenging. When it comes to people’s health and finances, this lack of transparency is untenable. 

3. Bias and fairness concerns

Black-box AI gives users no way to detect bias or flaws, meaning decision-makers can’t audit or defend the algorithm’s conclusion. For example, doctors might be unable to explain why an AI solution recommended a specific course of treatment for one patient while a similar patient with a similar diagnosis received a different recommendation. Are they both correct? What factors, if any, resulted in the divergent approach?

4. Regulatory compliance

Additionally, algorithms can inherit and amplify biases in training data, which is especially problematic for industries like healthcare and finance. Regulations exist to ensure fairness, but AI may inadvertently inject undetectable bias into loan decisions, treatment recommendations and many other critical functions. Organizations cannot adopt AI until they can understand — and control — the algorithms.

5. Resistance to change

Another hurdle is hesitancy to change. Healthcare and finance industries rely on legacy systems they’ve used for decades. The prospect of overhauling these systems to integrate AI can be daunting, as it requires a significant investment of time, resources and capital and carries a risk of substantial operational disruption. Any changes must comply with regulations, requiring extensive validation, testing and documentation.

But even the most regulated industries need to leverage AI. Organizations must find a way to overcome these challenges. 

How to Successfully Adopt AI in Regulated Industries

As an organizational leader, you should consider the following 5 actions in your implementation approach.

1. Set expectations

What is AI’s role in the organization? You must clearly answer this question from the beginning to gain organizational buy-in and ensure you meet the proper regulations. For example, an AI designed to gather patient paperwork must conform to rules different from one analyzing X-rays. Establishing the specific purpose and scope of AI implementation helps align stakeholders and provides a framework for measuring success. 

2. Provide transparency and observability

Regulated industries require AI tools that provide clear explanations for their conclusions. For example, the doctor needs to understand why a tool recommended a colonoscopy to properly analyze the results and decide on a course of action. Care providers also need context to explain how they arrived at a care decision. 

Answer the following questions to help establish transparency: 

• What training data did the model receive?
• How does the AI system use data?
• Can the data be used for outside use cases?
• How is the data secured?
• Is the system secure enough for the kind of data it’s using?

Transparency involves clear insight into the AI system’s decision-making process and the factors influencing its outputs. This level of openness allows organizations to audit the AI, ensuring it operates as intended and adheres to regulatory guidelines.

How do you achieve algorithm transparency? Employ explainable AI techniques, like decision trees, rule-based systems and attention mechanisms. Decision trees and rule-based systems provide a road map of an AI’s process, making outcomes easily auditable. Attention mechanisms operate in deep learning models to improve prediction accuracy and computational efficiency and illuminate what data the model is paying attention to and why. Seeking solutions with explainable AI will help organizations adhere to regulations.

SeeMore:4 Tips To Sell Your Generative AI Solution To Large Enterprises

3. Set guardrails 

Clear policies and procedures for both the algorithm and its users help ensure AI operates properly and adheres to regulations. 

Encoding guidelines directly into an AI platform lets you define AI’s operational parameters to control outputs. With predefined rules, AI is less likely to deviate from the intended path or violate policies. Embedded rules enhance transparency and auditability, allowing you to demonstrate compliance, correct issues, and maintain trust. 

You must also establish organizational guidelines and governance frameworks for the development, deployment and ongoing operation of AI systems. A well-designed policy should include these guidelines:

• Clear leadership and oversight.
• Defined roles and responsibilities.
• Explicitly defined use cases.
• Ethical guidelines.
• Risk management processes.
• Reporting protocols and continuous auditing. 

Humans are the most important element of an AI strategy. You can’t rely solely on technology to make decisions that significantly impact a person’s life. Human experts must interpret AI outputs and apply critical thinking to determine the best course of action. This oversight maintains accountability, verifies regulatory compliance, detects problems and addresses ethical considerations. 

4. Start small

Rome wasn’t built in a day. It’s a cliché, but it’s true. You should take the same perspective into your AI implementation. AI isn’t going to change business operations overnight, and large-scale adoption can lead to noncompliance. Start small and build up.

Consider the crawl-walk-run approach: 

• Crawl: Understand the basics of AI and its potential business applications to identify low-risk, high-impact functions that lend themselves to implementation.  
• Walk: Deploy AI for specific use cases or departments. You could start with a scheduling tool before expanding to an information intake system or adopt the technology in a single clinic or bank branch. 
• Run: Once you successfully test, refine and verify AI, you can begin integrating it into larger settings and for more complex use cases. 

5. Seek guidance

Employees in healthcare, finance and other regulated organizations are highly skilled in their domains and perform critical, specialized functions. They often lack the time and expertise to navigate AI’s complexities. Partnering with AI experts can help guide organizations in selecting the right AI solutions.

There’s a wide variety of available tools. Some are industry-specific. Others are an interface for a widely used general large language model. AI experts can look into each solution and outline its use cases and potential pitfalls to determine the best organizational fit. These partners can also assist in properly setting up new technology, providing training and establishing governance frameworks. 

Many startups are looking to shake up the healthcare and banking industry. Their products could bring significant improvements. For example, enabling pathologists to make more accurate diagnoses or helping bankers determine an applicant’s risk of loan default. 

Organizations in regulated industries cannot afford to ignore AI. The key to adoption is deliberate and cautious progress in implementation, paying particular attention to transparency and oversight. Implementation will be gradual, but with the right strategy and partnerships, even the most risk-averse industries can successfully navigate obstacles.

MORE ON AI IMPLEMENTATION

• AI Regulations Around the World: A Comprehensive Guide To Governing Artificial Intelligence
• AI Unleashed: A Guide to Responsible Implementation
• Top 10 AI Development and Implementation Challenges 
• Why GenAI Implementation Requires Collaboration
• Top Challenges of Implementing AI Models And How to Overcome Them