November 8, 2024
In our previous blog, we discussed how the advent of generative AI in our day-to-day lives has skyrocketed in the past few years, helping individuals and companies efficiently tackle tasks through automation and reduce the time taken to complete them.
Furthermore, new applications of gen AI for business solutions are being developed at a breakneck pace across industries such as Retail And Consumer Packaged Goods (RCPG) Retail and Consumer Packaged Goods, Banking And Financial Services, Healthcare and Life Sciences, and Human Resources , among others.
Additionally, companies are now expecting more tangible results from the implementation of gen AI to avoid losing market share. This is true for all the previously mentioned stakeholders: technology providers, service providers, and enterprises.
At the same time, these stakeholders must be mindful of their critical role in fulfilling the DEIB (Diversity, Equity, Inclusion, and Belonging) mandate, which includes promoting inclusive and equitable practices in gen AI development and deployment. The absence of comprehensive DEIB measures in gen AI models can have detrimental effects both internally and externally.
Furthermore, equitable artificial intelligence (AI) learning is essential. A survey conducted by a leading consulting firm, indicates that only 10-15% of businesses have established AI roles focused on fostering diverse perspectives within their teams.
Professionals’ lived experiences provide critical insights for mitigating bias—a truth that all stakeholders must embrace. Before exploring potential solutions, it’s important to investigate the root causes of bias, the different types of biases present, and their implications, as our analysts have done below.
Reach out to discuss this topic in depth.
The Case for DEIB in Gen AI:
While technology offers substantial benefits, a significant DEIB challenge persists within current gen AI frameworks, leading to adverse effects for individuals and organizations. AI algorithms – a host of which are trained on existing framework models, lack diverse perspectives, and can mirror the biases of their creators, perpetuating inequalities and harming marginalized communities.
Cultural and social biases often infiltrate these systems, resulting in flawed outputs that do not accurately reflect varied experiences and knowledge.
Some benefits of unbiased gen AI Models include:
At the same time, adopting unbiased gen AI models can significantly benefit organizations by:
- Enhancing Decision-Making: Eliminating biases allows for more accurate, objective insights, improving decision-making across scenarios
- Improving Customer Insights: Objective data analysis helps businesses better understand customer needs, facilitating targeted marketing
- Promoting Diversity in Hiring: Unbiased AI can eliminate discrimination in recruitment, supporting diverse candidates, including neurodivergent individuals
- Streamlining Operations: Reducing bias in automated processes optimizes operations, enhancing overall efficiency and productivity
- Fostering Innovation: Bias-free AI models yield more diverse and creative ideas, propelling innovation across sectors
- Improving Risk Management: Unbiased AI provides clearer, balanced assessments, aiding organizations in identifying and managing risks effectively
- Ensuring Compliance with Ethical Standards: Utilizing unbiased AI aligns with ethical norms and best practices, fostering trust and accountability
- Creating a More Equitable Workplace: By promoting fairness, unbiased AI contributes to a more inclusive environment, driving organizational growth
A deep dive into the causes and types of the bias in terms of DEIB?
Gen AI models are statistical by nature and prone to errors, especially when lacking domain expertise. Currently, a small, homogeneous group often determines the data used for training these models. Many models are built on foundational frameworks such as BERT or RooBERTa, which can carry inherent biases if not addressed from the outset.
Types of DEIB bias include:
The social and business cost for business by utilizing a biased gen AI model:
Addressing these challenges is paramount for companies when accounting for the vast use cases of this technology across sectors. For example, 19% of organizations are leveraging AI to develop new products and services across the RCPG space, according to an Everest Group insight.
Similarly, 40-45% of business leaders of mega enterprises (revenue exceeding US$ 1 billion) have reported successful implementation of gen AI across various operations in this Everest Group viewpoint. We expect this number to consistently increase in the coming years.
If the models used for these products or services produce biased results or incorrect outcomes (an important component of ‘hallucinations’), it could negatively impact the companies’ reputations and their bottom lines. Thus, there are both direct and indirect costs associated with leveraging these models. The two key types of costs that businesses would suffer from are the following:
Business Cost: The direct financial expenses incurred by a business, including production costs, operating expenses, and the costs of complying with regulations. These costs can be both internal and external to the business
Social Cost: The total economic cost to society, including both direct costs borne by individuals and businesses, as well as indirect costs such as environmental damage, decreased quality of life, and social inequality
While unbiased AI models are essential, their development and deployment can be costly. Collecting high-quality data for model training, designing and customizing AI models from scratch, and employing sophisticated techniques and specialized talent all contribute to the complexity.
Additionally, scaling these models across large organizations or multiple geographies can introduce new biases due to variations in cultural, linguistic, and socioeconomic factors. Therefore, companies must be deliberate in identifying which products, services, or functions truly require such AI models.
In response, some organizations have appointed Chief Diversity, Equity and Inclusion (DE&I) Officers, but this approach may be limited, as these officers typically focus on talent acquisition and retention.
Effectively addressing AI’s DEIB impact requires input from multiple leaders, including the Chief Information Officer (CIO)/ Chief Technology Officer (CTO), Chief Product Procurement Officer (CPO), Chief DE&I Officer, and Chief Sustainability Officer, making it both resource- and cost-intensive. Furthermore, while algorithmic impact assessments are well-intentioned, they often fall short in fully capturing the broader social implications of AI models.
To address this challenge, Everest Group has developed a framework that stakeholders can use to navigate these complexities effectively, with the overarching principle of the “Comprehensive Inclusion Framework” viewed from both an internal and external perspective. This principle is broken down into four key areas:
- Inclusiveness emphasizes broad representation in the entire AI development lifecycle. It ensures that diverse perspectives, experiences, and needs are considered when designing, developing, and deploying AI systems
- Impartiality ensures that AI decision-making processes are neutral, objective, and free from bias or unfair influence by continuously assessing the outputs of the model and checking for impartiality. Thus, blending in objective data driven insights
- Equity, in the context of AI ensures that all user groups experience fair and just outcomes from AI systems, regardless of their background, demographics, or identity
- Accessibility, focuses on making sure that AI technologies are usable and beneficial to all individuals, regardless of their socioeconomic status, disabilities, education, or geographic location
The framework provides a comprehensive approach to integrating gen AI and DEIB policies within organizations across vertical and horizontal processes. It categorizes various policy combinations based on the level of emphasis placed on AI and DEIB and offers recommendations to achieve optimal alignment. The categories include:
- Low DEIB Impact: DEIB efforts are not prioritized due to the lack of strong business or social cases
- Medium DEIB Impact: DEIB efforts are focused on business and social benefits, with AI considered a tool to enhance these case
- High DEIB impact: DEIB values are deeply integrated into organizational culture, using AI to drive inclusivity and equity throughout the business
The current state of the market in terms of DEIB embodiment by stakeholders:
As mentioned in our last blog post, across stakeholders, the current level of DEIB integration according to our ROLE framework is as follows:
As gen AI increasingly influences business operations, stakeholders must prioritize DEIB in their AI development and deployment efforts.
Tackling inherent biases and fostering fairness will not only mitigate risks but also enhance decision-making, customer insights, innovation, and workplace equity. By adopting frameworks such as Everest Group’s “Comprehensive Inclusion Framework”, organizations can effectively align their AI and DEIB strategies, ensuring long-term success and ethical compliance.
We are actively tracking the evolution of artificial intelligence and its impact on the future of all sectors. To discuss the latest trends and their implications for brands, technology vendors, and service providers alike, feel free to reach out to Kanishka Chakraborty ([email protected]), Meenakshi Narayanan ([email protected]), Abhishek Sengupta ([email protected]), Abhishek Biswas ([email protected]), Rita Soni ([email protected]) and Cecilia Van Cauwenberghe ([email protected]).
If you found this blog interesting, check out our blog focusing on Building Purpose-Driven Generative AI (gen AI) – Why We All Have A Role To Play In The Future Success Of The Gen AI Ecosystem | Blog – Everest Group, which delves deeper into the subject of gen AI.
This is the first of a new series of blogs, with plenty more to come in 2024 and 2025!
