Consumers have grown increasingly dependent on the pervasiveness of AI in everyday life. For example, with a click of a button, consumers can make purchases and, based on user behavior, AI identifies patterns and generates helpful suggestions that organizations can then leverage. Yet, while day-to-day operations have simplified, a burning question remains — is this data being used and stored responsibly?
Universally, we’re prone to focus on the positives that smart technology and AI bring to our lives, and that’s why it is important for organizations today to constantly be aware of the implications of mishandling data while identifying new ways to successfully scale AI. In addition to initial awareness, organizations should be actively working to ensure that all teams are involved in Responsible AI practices and asking key questions about expansive AI integration — Are teams considering all aspects of how data can affect lives and livelihood? How can organizations mitigate the issue of data bias? Some of our guest speakers at Everyday AI New York helped us explore the important topics that these critical questions provoke; let’s see what they had to say...
Is Responsible AI Actually Something Organizations Can Achieve?
According to Nicole Alexander, Global Head of Marketing for Meta, and Noelle Silver, Executive AI & Analytics for IBM and CEO and founder of the AI Leadership Institute, Responsible AI is the need of the hour and absolutely necessary for a better and more secure future with data analytics and machine learning.
Defining Responsible AI
Better monitoring, better control, and better oversight are a few of many aspects that Responsible AI should ideally support. The term Responsible AI essentially means developing systems, frameworks, and processes that mitigate any harm to individuals and masses while developing a product that the society needs and values. Responsible AI is a practice that is tangible and something that needs to be constantly governed, because governance is the only opportunity for organizations to develop and build responsible processes. As a team working on a certain product, it is necessary to think about the repercussions of what is being built — the effect it will have on people.
It is not about what you meant, but about the outcome.”- Noelle Silver
The Importance of Inclusivity
In our mission of developing Responsible AI, the perceived outcome needs to be audited constantly by a multitude of individuals who can help developers notice what they’ve missed. Being a part of a project from ideation to execution can sometimes lead to blindspots that often reap unwanted consequences. By having a team of individuals from different walks of life, different professions, and different age groups, brainstorming becomes a very interesting process. Becoming familiar with different points of view, organizations can make inclusive decisions and in turn change how data is being used.
Noelle Silver gives a very interesting anecdote about how projects she and her team worked on in the past had unforeseen effects on users, from the impact of the name of the product to the use of the product. One good way to be able to anticipate impact is constant and thorough research and audit, so as to be at par with the ever-evolving needs of users and the condition of the perceived target market. To ensure the project delivers the desired results and more, it is beneficial to have a team that is not afraid to ask questions and share opinions. Historically, we know that asking the right questions has led to many discoveries and solutions to pain points unknown to us as individuals. Creating Responsible AI is about active listening, accepting various perspectives, checking the right boxes, and ensuring the changes are being implemented throughout the organization.
Building an inclusive team is as important as building an inclusive product!
The Spectrum of Bias In AI Across Organizations
Triveni Gandhi, Responsible AI Lead at Dataiku, was joined on stage by Dr. Brandeis Marshall, CEO of DataedX(a data education, equity, and ethics education cell) to discuss the growing importance of data ethics and data bias. Today, data is all about scalability. In order to ensure organizations scale business in the right direction, they must think about the context of data, be actionable, and practice conscious decision making. Reducing the risk of mishandling algorithms is key for mitigating data bias across organizations.
Understanding Bias Reverberation
Data is interconnected, and so is bias, in every form. We talk about how bias threads throughout the process, be it in the form of data bias or any other form of bias. Biases have always existed around us. Most data biases stem from historical practices of discrimination, which reverberate through the organization. Every decision made and policy passed has an effect that is felt within the digital infrastructure. When managing data, it is imperative for organizations to predefine processes and systems, and ensure they are constantly evaluated in order to reduce the risk of mishandling algorithms.
Today, every organization dealing with data has their own data management and data privacy policies, but these policies mainly focus on customers as opposed to internal processes. There has to be a provision for companies to assess how these policies are integrated within the products that are developed and distributed. It is essential for products to align with the data policies of the company to mitigate any form of data bias, and data governance and compliance teams play a big role here.
Who Has a Part to Play?
The governance team must ensure that the code does no harm by measuring risk and effect in advance, and be mindful of the effect the product can have on users. Organizations need to think of AI as a slate of viable options and after careful consideration pick the most appropriate one, keeping in mind the human aspect of algorithms. The entire organization needs to work together to build responsible systems and then implement these systems. Each member of the organization is responsible for making data more equitable. They are the stakeholders who decide how the product goes to market and ultimately what possible effect it can have on the users.
The solution to mitigate the risk of data bias lies within organizations. Investing in systems that will do more good than harm and developing responsible processes will result in reliable and Responsible AI.