What if the cost of having machines that think, is having people that don’t? – George Dyson from the book Turing’s Cathedral
This is a relevant question from George Dyson’s master work: Turing’s Cathedral; which explores the early development of computers, nuclear weapons, AI, and more.
While this quote warns against relying too much on AI to do the thinking for us, it also brings to mind thoughtless AI implementations and the very real repercussions that can result. CIO.com recently featured an article highlighting some of these failures:
- In February 2024, Air Canada was ordered to pay damages to a passenger after its virtual assistant gave him incorrect information at a particularly difficult time.
- In March 2024, The Markup reported that Microsoft-powered chatbot MyCity was giving entrepreneurs incorrect information that would lead to them breaking the law.
- In August 2023, tutoring company iTutor Group agreed to pay $365,000 to settle a suit brought by the US Equal Employment Opportunity Commission (EEOC). The federal agency said the company, which provides remote tutoring services to students in China, used AI-powered recruiting software that automatically rejected female applicants ages 55 and older, and male applicants ages 60 and older.
- In November 2021, online real estate marketplace Zillow told shareholders it would wind down its Zillow Offers operations and cut 25% of the company’s workforce — about 2,000 employees — over the next several quarters. The home-flipping unit’s woes were the result of the error rate in the ML algorithm it used to predict home prices.
Continuing to think, while keeping a close eye on AI risk quantification, mitigation, and performance is foundational to successful AI implementations. The financial services industry has been doing it for decades. You can too.
In the rapidly evolving landscape of artificial intelligence (AI) and machine learning (ML), the governance of these models has become a critical aspect of ensuring their effectiveness, reliability, and ethical use. Proper AI builds begin with a robust definition of business requirements, recognizing that AI is a tool designed to solve specific problems. This article will explore the key steps involved in AI and ML model governance, from defining the problem to launching and monitoring the model in production.
1. Defining the Problem: The Foundation of Successful AI
The first step in any AI project is to clearly define the business problem that needs to be solved. The old adage, “a problem well defined is a problem half solved,” holds true in the realm of AI. Without a clear understanding of the business requirements, the AI model implementation is likely to be misaligned with the organization’s goals, leading to inefficiencies and wasted resources.
“A problem well stated is half solved”
Defining the problem involves several key steps:
- Identifying the Business Objective: What specific business challenge are you trying to address?
- Setting Clear Goals: What outcomes are you hoping to achieve with the AI model?
- Understanding the Data: What data is available, and how can it be used to inform the model?
By thoroughly defining the problem, you lay the groundwork for a successful AI project. This initial step ensures that the AI tool is being used to address a real, tangible issue rather than being implemented for its own sake.
2. Choosing the Right AI Tool
Once the problem is clearly defined, the next step is to select the AI tool that best suits the task. Every AI tool has its strengths and weaknesses, and choosing the right one is crucial for the success of the project.
- Supervised Learning: Ideal for problems where you have labeled data and need to make predictions.
- Unsupervised Learning: Useful for identifying patterns in unlabeled data.
- Reinforcement Learning: Suitable for problems where the AI needs to learn through trial and error.
Selecting the appropriate tool requires a deep understanding of the problem, the available data, and the strengths and weaknesses of various AI/ML tools. It is essential to consider the tool’s capabilities, the complexity of the problem, and the resources available for implementation.
3. Quantifying AI Risk
After defining the AI approach and selecting the appropriate tool, the next step is to quantify the risk associated with the AI model build. AI Risk Assessment is a multifaceted process that considers various factors, but much of it relates to the implications of incorrect or erroneous answers from the system.
- Data Quality: The quality and reliability of the data used to train the model.
- Model Accuracy: The likelihood of the model producing accurate predictions.
- Ethical Considerations: The potential for the model to produce biased or unfair outcomes.
- Other Factors
Risk assessment helps to identify potential issues early in the process, allowing for mitigation strategies to be put in place. This step is crucial for ensuring that the AI model is not only effective but also ethical and reliable.
4. Model Construction and Validation
With the problem defined, the tool selected, and the risks quantified, the next step is to construct and validate the model. This involves several key activities:
- Data Preparation: Cleaning and preprocessing the data to ensure it is suitable for training the model.
- Model Training: Using the prepared data to train the AI model.
- Validation: Testing the model with a separate dataset to ensure it performs as expected.
Validation is a critical step in the process, as it helps to identify any issues with the model before it is deployed. This ensures that the model is robust and reliable, capable of producing accurate and consistent results.
5. Launching and Monitoring the Model
The final step in the AI model governance process is to launch the model into production and monitor its performance. This involves several key activities:
- Deployment: Integrating the model into the organization’s systems and workflows.
- Monitoring: Continuously monitoring the model’s performance to ensure it is operating as expected.
- Maintenance: Regularly refitting and rebuilding the model to account for changes in the data or business requirements.
Monitoring is an ongoing process that ensures the model remains effective and reliable over time. It allows for the early detection of issues, enabling timely interventions to maintain the model’s performance.
Conclusion
AI and machine learning model governance is a complex but essential process that ensures the effective and ethical use of these powerful tools. By defining the problem, selecting the right tool, quantifying the risk, constructing and validating the model, and launching and monitoring it in production, organizations can harness the full potential of AI while mitigating the risks associated with its use. Proper governance is not just a best practice; it is a necessity for any organization looking to leverage AI to drive innovation and growth.
VentureArmor: Here to Help
VentureArmor’s AI/ML Risk Audit Services are here to help. Whether your company needs AI/ML risk assessment and mitigation capabilities built from the ground-up, or you need an independent audit of your existing capabilities, the industry experts at VentureArmor are here to help. With decades of experience implementing best-in-class compliant AI solutions in Financial Services, Supply Chain, and Healthcare, our expertise covers:
- AI Model Risk Assessment & Tiering Frameworks
- AI Model Build Standards
- AI Model Documentation Standards
- AI Governance Council Formation and Management
- AI Model-Ops Best Practices
- US and International Data and AI Compliance Consultation (US and EU)