How do you manage AI in unpredictable environments?

Beyond defining the context (which is a critical and important aspect!) I would suggest a broader review of business strategy and a review of the value that applying AI will bring in various contexts (like an audit of sorts - consider the industry and then the organisation to seek competitive advantage leading from its application). With so many opportunities to apply AI (tactically) it is important to consider AI in the broader context of digital transformation as it is a catalyst to further action and needs a strong understanding of the industry & clarity on the value to be added (for product leadership; customer intimacy or operational efficiency? What’s the central focus of your business). Start with the big picture, then focus in!

In practice, automation can often pose the greatest risk to security, stability and safety . Understanding the capabilities of a device and enforcing limitations on what AI can do is critical. Generative AI, enjoys creating and generating things. This may sound beneficial, but it often isn't. Although AI seems to "think" like a computer, it processes tasks, often taking up to 10 to 15 tasks per input. Without specific objectives and limitations, AI tends to keep creating. This could yield positive outcomes or negative ones. As a result, I've allowed AI to completely destroy a few databases and cause a blue screen on Windows. At times, the AI was misrepresenting data, pulling information from other users and returning it to the wrong user.

For me I'd say human-in-the-loop. Keep a human operator in the decision-making process, especially for critical tasks. This way if the AI encounters an unpredictable scenario the human can intervene.

In addressing uncertainty, your choice of AI approach should align with the specific context and problem at hand. Rules-based ideal for well-defined and structured problems relies on predefined rules and logic to make decisions. It is suitable when the problem has clear and deterministic outcomes. When dealing with data-driven and adaptive challenges, machine learning excels. It enables AI systems to learn from data, recognise patterns, and make predictions. This approach is valuable in situations where outcomes may vary and evolve. Hybrid AI leverages the strengths of different techniques to tackle complex and multifaceted problems. The beauty of AI is it can be tailored to your specific needs, not a one size fits.

In the context of handling AI in unpredictable environments, a crucial step involves training and testing AI systems. It's essential for preparing AI to navigate unforeseen circumstances. The core of this procedure lies in using diverse, real-world data. It is imperative that this data accurately represents the complexities of the environment in which the AI operates, reducing the chances of biases and errors affecting AI decision-making. Ensuring that AI can apply its knowledge to new, unencountered scenarios is equally important. We want AI to be adaptive, responsive, and capable of effectively addressing novel challenges. This involves extensive testing of various scenarios, input cases, and data inputs to verify performance.

My experience taught me that in the world of finance, AI is an ever-evolving companion. Its continuous improvement is essential to stay ahead in the fast-paced and unpredictable market landscape. By updating and refining our AI, we can succeed in enhancing our financial strategies. The most used is robotic process automation which is a great tool for accountants to take out all the mundane and repetitive tasks we do daily. Companies are able to take 2 hours out of their team's day, leaving time for value-added tasks and relieving qualified accountants from robotic tasks. FP&A tools driven by AI that are available are making predictive analyses for whatever and what-if scenarios readily available. Cutting down time spent on spreadsheets.

Managing AI effectively isn't just about good communication and trust, although those are important. You also need top-notch data and automated feedback loops to keep improving the AI. Explainable AI is your friend here—it helps everyone understand why the AI makes certain decisions. Don't forget to build your system in a modular way, it'll make updates a breeze and help you scale up easily. Have backup plans and redundancies ready for when things don't go as planned. Keep an eye on performance with real-time dashboards and regular data audits. In complex setups, your AI may need to work with other AIs, so make sure they play nice together. And ethics? Non-negotiable, and always have a way for humans to step in when needed.

Structuring of your content also provides the power to ensure the AI is able to provide the best results. By clearly planning and organizing your content you have the options to segment your knowledge by audience or domain. This could enable the ability to provide more accurate and succinct responses both for internal use cases but also for external options if applicable.

Image recognition and OCR bring an entirely new multitude of considerations into play, particularly when identifying objects and marking threats. Depending on the software's controls, this could potentially lead to severe injury or worse.