How to achieve responsible and effective generative AI solutions?

If you are not yet adopting generative AI (GenAI) in production applications, you have some catching up to do. The evolution of Artificial Intelligence (AI) is transforming industries, pushing the boundaries of innovation and creating opportunities once considered unimaginable. It’s already proven that GenAI engines are enhancing human productivity in many important tasks. Areas such as radiology, diagnosis, content creation, chatbots, image analysis, and many other areas are clear indications of a future where AI agents are effective companions for humans, making many roles more effective and cutting the time to complete complex tasks.

But as AI becomes more pervasive, we must also acknowledge and confront its hidden challenges and risks—particularly around unintended bias, regulatory non-compliance, and the escalating costs of implementing and maintaining AI solutions.

Addressing AI Bias: A Crucial First Step

One of the most significant pitfalls in AI development is bias. AI systems are only as good as the data they are trained on, and when that data contains historical or societal biases, AI inadvertently perpetuates them. This issue has garnered increasing attention globally, including in India, where AI's influence continues to expand across sectors.

Bias in AI manifests in numerous ways, from gender to ethnic and socio-economic bias. In one notable example, an AI-powered recruitment tool was found to favour male candidates over equally qualified female applicants. Additionally, language bias has been observed in AI systems, particularly concerning Indian languages, where models trained predominantly on English data often struggle with tokenization and representation of regional languages.

This results in operational challenges and reinforces existing biases, highlighting the need for more inclusive and balanced training data to ensure equitable language processing across diverse linguistic communities.

To combat data bias, companies must adopt a multi-layered approach. First, it's essential to carefully curate datasets, ensuring they represent a broad spectrum of demographics, experiences, and viewpoints. While this is difficult to do manually, some organizations (including our own) have created a data curation engine that ensures a well-curated and balanced input set. Second, organizations must deploy bias detection tools to help spot and mitigate potential bias during the development phase. Finally, human oversight remains critical.

AI models should not operate in a black box; instead, there must be mechanisms to audit and correct any bias that might emerge as systems interact with real-world data. Every organization should establish an AI governance process that monitors use and results so they can react in real time to any potential issues.

Reducing bias is not just a technological challenge but also an ethical imperative.

Mitigating AI Compliance Risks for Safer Adoption

Compliance risks in AI are also becoming more of a focus as companies use this technology more broadly and for critical tasks. As AI technologies evolve, they are being leveraged in increasingly sophisticated ways, which also opens the door for malicious uses. The rise of AI-powered cyber threats is particularly concerning, with AI-enhanced phishing and deepfake scams gaining traction.

“Concern about artificial intelligence (AI)-enhanced malicious attacks ascended to the top of the Gartner emerging risk rankings in the first quarter of 2024, according to Gartner  80% of respondents expressed concerns about these threats. AI-driven threats, which range from data breaches to AI-manipulated misinformation, pose significant risks to a company’s security and integrity.

In addition, regulatory frameworks around AI are tightening. The General Data Protection Regulation (GDPR) in Europe and the California Consumer Privacy Act (CCPA) in the United States have set new benchmarks for data privacy and the ethical use of AI. India is also in the process of enacting its data protection laws, which will further raise the compliance stakes for businesses operating in or with India. The penalties for non-compliance are severe, not just financially but also in terms of reputational damage.

The path to compliance involves establishing robust governance frameworks around AI, ensuring transparency in data collection, storage, and processing. Companies should be proactive, embedding privacy and security considerations into their AI systems from the ground up. Another key aspect is fostering a culture of transparency, where end-users are aware of how their data is used and have control over it.

Navigating the Budgetary Complexities of AI

Beyond bias and compliance, organizations often underestimate the total cost of ownership for AI projects. While the initial cost of AI may seem high, the long-term operational expenses pose a more substantial challenge. “At least 30% of generative AI (GenAI) projects will be abandoned after proof of concept by the end of 2025, due to poor data quality, inadequate risk controls, escalating costs or unclear business value, according to Gartner.”

The costs of AI go beyond software and hardware. Training complex AI models, particularly in domains like natural language processing or computer vision, can require significant computational power, resulting in high infrastructure costs. Additionally, maintaining these systems—through constant updates, retraining, and troubleshooting—demands ongoing investment.

To mitigate these costs, companies should focus on designing AI systems that are flexible, scalable, and cost-effective, such as cloud-based AI services. It is crucial to select AI projects strategically, focusing on those that align with business goals and offer a clear return on investment.

Ultimately, as AI continues to evolve, businesses must navigate these hidden challenges with as much rigor and foresight as they apply to harnessing AI’s transformative power to safeguard their operations, reputations, and bottom lines in an increasingly AI-driven world. Companies that get this right will absolutely demonstrate higher return on investment and increased shareholder value. All it takes is a strong desire and focus on creating value while managing risks.

By Neil Fox, Senior Vice President, Engineering – AI, Persistent Systems