International Airlines Group (IAG), parent company of British Airways, is taking a cautious approach to artificial intelligence (AI) automation. Ben Dias, IAG’s chief AI scientist, outlined a three-question test to assess the risks and benefits of AI implementation.
Speaking at the Skift Data + AI Summit in London, Dias emphasized the importance of balancing innovation with safety and regulatory compliance. IAG’s strategy involves a federal approach, allowing individual carriers to explore AI use cases while maintaining group-level oversight.
A Three-Pronged Approach to AI Risk Assessment
When considering AI automation, IAG asks: How critical is the function? What is the potential blast radius of failure? And what fallback options exist if the AI system fails? This framework helps the company evaluate the potential impact on revenue, safety, customer experience, and reputation.
Dias noted that quick recovery from AI failures is essential. If a system can recover rapidly, automation is more feasible. However, if recovery is slow or uncertain, a risk assessment is necessary to determine the viability of AI implementation.
AI Governance and Policy
IAG’s AI policy is built on 10 principles, including human oversight, bias prevention, and accuracy requirements. Each AI solution is mapped against these principles, and if it falls short, the company works to mitigate risks rather than rejecting the idea outright.
Centralized vs. Decentralized AI Development
Other companies, such as Accor and Strawberry, are also working through the challenges of AI implementation. Nicolas Maynard, SVP of data and AI at Accor, advocates for a centralized approach to AI development, citing the need to protect customer data and intellectual property.
In contrast, Kari Anna Fiskvik, chief digital and technology officer at Strawberry, emphasizes the importance of allowing employees to experiment with AI. By enabling staff to learn and experience AI firsthand, the company hopes to support innovation and uncover new ideas.
Centralization and Experimentation in AI Strategies
This approach reduces unauthorized AI use and ensures compliance with regulations like the EU Artificial Intelligence Act. The goal is to inform teams about potential risks without hindering progress.
Kari Anna Fiskvik of Strawberry takes a different tack, encouraging employees to experiment with AI to support learning and innovation. While acknowledging compliance and prompt-related challenges, the company allows controlled use to monitor costs and generate ideas. This hands-on approach aims to balance exploration with oversight.