During my recent AI PM interview, I was asked a question that seemed straightforward but was layered with complexity: "What's your initial step?" This question transported me back to my early days in AI, when amidst algorithms and data, I realized the real game-changer was something more fundamental—trust.
Trust isn't just a buzzword; it's the foundation upon which successful AI projects are built. It's about earning the confidence of your team, stakeholders, and users by being transparent, accountable, and consistent. In the realm of AI, where uncertainty and rapid change are the norms, trust is the anchor that keeps the ship steady.
To solidify trust from day one, consider these strategies:
1. Transparency: Emphasise clarity in explaining the rationale behind all AI decisions. While the intricacies of models may not always be a concern, illustrating the reasoning for choosing a particular approach and outlining both potential benefits and drawbacks is crucial for stakeholder understanding and buy-in.
2. Guardrail Metrics: Maintain the integrity of performance by ensuring that existing metrics are not adversely affected. These metrics act as a safety net, mitigating risks and outlining clear contingency plans to manage any perceived negative outcomes.
3. Human Oversight: Always let the human make the final decision, ensuring AI supports rather than dictates. Establish regular review sessions, 'brown bag' meetings, and demo days to address concerns and strategise on handling edge cases effectively.
4. Education (with empathy): AI isn't just a product; it's part of an experience. It's essential to onboard and educate users with a deep sense of empathy, ensuring they feel supported throughout their interaction with the technology.
5. Value Delivery: Clearly define success metrics and ensure AI's positive impact on user experiences. Given the network effect's significance, a phased rollout is crucial, starting with a smaller user group to validate success before expanding. Clarify the benchmarks for success at each PLC stage.
6. Accuracy: Make precision in AI outputs a top priority, as it lays the foundation for trust. Recognise the varying accuracy requirements across use cases, starting with conservative settings (i.e. lower temperature) and gradually adjusting to find an optimal balance the team can trust as a reliable benchmark.
7. Data Privacy and Safety: Implement robust data protection, like dynamic grounding and data masking, to safeguard user privacy.
8. Local LLM Development: Invest in developing localised large language models to cater to specific requirements, particularly when data privacy is a concern and budget allows.
Discussion about this post
No posts