Navigating Ethical AI Development: Lessons from Everyday Scenarios

Mani Sitaraman

Senior Director - Data & AI Practice

October 16, 2024

Share this article

While “Ethics” may not typically evoke thoughts of technology, developing ethical AI systems requires a distinctly human touch. We, at Collabera Digital, work with our clients to help them build ethical AI applications.

Our clients leverage our “Creative Minds Generative Algorithms” framework to ensure that the tenets of ethics are embedded in each stage of the development of the AI application, i.e., designed, developed, tested, and productionized. In all instances, we keep out the technical jargon when conversing with our end users to ensure the core message resonates clearly.

Let’s explore this with our imaginary friend, ‘Cool Joe,’ a budding AI engineer who navigates through different situations to build an AI application and explains it to others. His experience offers a relatable ethical framework, helping us navigate and discuss these complex issues.

Do not question it, just do it

Joe typically doesn’t challenge the status quo – when given a set of requirements, he completes the task (we’ve all been there). While there is nothing wrong with doing what you are told, in the AI world, one needs to understand the implications of the past, present, and the future. It’s not just about developing the logic; it also involves the dataset, various permutations and combinations, and the algorithms used to generate the output.

The AI developers at Collabera Digital understand how sensitive, personal, and private data is used. We not only question the status quo but also suggest ways to ensure the AI application is ethical.

Fail in small steps

Developing a strong sense of ethics is a gradual journey, built over time. Joe now understands the need to reason and question, but his challenge is that he does not know whether he is building a spacecraft or a monument, given that he needs to code and deliver a module for a software application. In DevOps, Agile, and 2-week sprints, the AI logic and updates are constantly changing, and only a few have a clear end-to-end understanding. When Joe’s code does not work as expected, the question pops, “Hey Joe, how did this happen?”

Ethically, AI systems may fail when their initial intent is lost, and their logic becomes overwhelmed by exceptions, blurring the lines between rules and deviations.

We at Collabera Digital ensure that you succeed in small steps – steadily and quickly.

‘Lost in the group’: The need for a RACI matrix

Joe has now begun to think ahead and understands the AI framework before working on it. He also understands various permutations and combinations of the data and logic. However, his challenge remains to integrate things.

The output of every block of code that an AI engine produces needs a responsible person associated with it, and without a proper RACI matrix, AI is less likely to succeed.

Collabera Digital’s AI Governance framework brings all the stakeholders together and ensures accountability, transparency, and traceability of actions and decisions.

‘Competition has done it’

By now, our cool Joe has become quite an expert, but one of his friends tells him about their competitor’s advanced AI systems and how they have merged both public and private datasets to gain market advantage.

Joe reminds his friend that the competitor’s AI models and datasets may not have followed the laws and regulations, potentially leading to public backlash and legal scrutiny.

Collabera Digital’s AI developers have a sound understanding of the ethical perspective. We understand that often, one does not know the logic, algorithms, and models that the competition has used, and trying to imitate or expedite can give a different and undesirable output; hence, we ensure that this does not happen.

‘We won’t get caught’

Hmm… but you will; it’s only a matter of time. Now Joe’s friend toes a different line and says that when the users do not know the logic, then how can we be blamed?

Joe pauses for a moment and reminds his friend that with AI producing so many different outputs, it is only a matter of time before users begin to see patterns.

Collabera Digital’s priority is to ensure and safeguard our clients’ and users’ AI journey. Ethically speaking, we build, check, supervise, and hold the algorithms used in machine learning to the highest standard.

‘Doing is believing’

Now, the conversation gets interesting. Our cool Joe’s friend makes a realistic remark: “In the name of ethics, if we stop making changes, trying new things, and inventing new algorithms, then there will be no progress.”

Cool Joe smiles, acknowledges his friend’s point, and adds that customers, leaders, developers, and all stakeholders need a deeper understanding of the following:

1. Intent: It is important to verify whether the intent behind using AI has changed over time, because if it has, the machine learning logic, models, and algorithms need to be revisited.

2. Supervised and unsupervised models: The AI team needs to clearly know when, where, and how to cap the output of unsupervised models and have enough resources to check and re-calibrate the supervised models.

3. Test in the Development & UAT environments: It’s important to keep experimenting in a safe environment before going live and have the courage to take it down if the outcome is undesirable.

One needs to keep experimenting in a safe environment before going LIVE and have the courage to bring it down if it is undesirable.

While Joe is making progress, we at Collabera Digital remain committed to ensuring that all our AI and data applications meet customer needs and are fundamentally ethical.