Predict & Influence Muses - Series 009: How to make AI more useful?

Following last week (Series 008) thoughts about ChatGPT taking on the world, it is timely to discuss the topics about the usefulness of AI, or rather, how to use AI. It is all about user experience!?

Using the current hot debate about ChatGPT vs. Google Search, it seems that ChatGPT has adopted better user experience than Google Search. While predicted results from Google are great and we have been using it for more than 20 years to find answers to our questions, the way we as users use it already familiar with its AI output – the search results pages that mixed with some search ads and all the most relevant (based on Google algos – we may not necessarily agree) links and we as users will need to investigate the results ourselves and then decide whether to check out the resultant links predicted by Google Search AI. There are extra steps that we need to take though we will have more choices to decide on the answers. In recent years, with the Google Search Engine infuses with knowledge engineering / knowledge graph, an important area of AI, when we search important/ famous entity/ person, the results of knowledge graph, with the relationship and links with different information about the entity and person displayed on the right side of the search results, represents a better user experience. You can try Google Brad Pitt or Microsoft to get what I meant.

ChatGPT, on the other hand, assume that the user wants a direct answer, rather than a bunch of links as results. Therefore, for every question/ search, it will return the best possible answer (but sometimes wrong factually) from the knowledge base that it is trained about (the latest data was from the year 2021). It is a new user experience to the world that familiar with Google Search results as outputs. We now have direct answers to our questions! There is no need for us to investigate the search results and explore.

?Now, how’s AI be useful for the corporate setting? Let’s use real life examples. ?Now, we use AI for fraud detection and risk engine purpose. Let’s first find out who are the users. While there are decision makers like CEOs, CFOs or CIOs who may have to collectively make the decision to make the purchase, the actual users would typically be the investigation officers (IOs) for law enforcement agencies of government, or compliance, anti-money laundering (AML) analysts for the banks, for example. For the purpose of this writing, let’s use IOs (for law enforcement agencies) as the target users.

?We need to know the day-to-day tasks for the IOs, and what are their common challenges. In this case, through interactions and workshops, we would know that they are mostly hit with false alarms as their current methods are typically either manual or rule based (or both). The majority of their time will be spent to close off the false alarm alerts. Some IOs may have to rely on manual ways to piece different data points and information together to find and trace data patterns, which is a big pain. Next, they also want to have more accurate indicators on potential entities, persons, transactions, events that may have happened that are having a higher chance of actual fraud/irregularities, rather than misses those. The cost (mainly reputation and financial) of missing these are big, typically. Lastly, they will need a set of tools that allow them to investigate high probability cases more effective, ie to know the why (ie why the risk score of certain entities are high), to draw linkages among entities, persons, events and connections, automatically. Thus, it would meet their 2 objectives: 1) leave no stone unturned for all the data and intelligence that are collected (ie more accurate), and 2) reduce the investigation time of cases. AI and data, therefore, would help with all the steps possible, from the backend (ie data processing and AI predicted results) and how the processed data and predicted results organised in the IOs’ user interface to facilitate their work.

That’s how to infuse AI into day-to-day works, and ensure better human-machine collaboration!

?

?

??