Imagine a crucial decision that needs to be made. Who should be the one to make it: a human or an artificial intelligence (AI)? We’ve debated this topic before, acknowledging that humans outperform AI in certain tasks while AI excels in others. But when it comes to a single decision, the answer is a captivating fusion of holistic curves and human bias. Let’s delve into it together.
Contents
The Complex World of Fraud Detection
Consider a fraud detection system—a system that generates alerts for potentially fraudulent transactions. Financial analysts are responsible for reviewing each alert. With thousands of events generated daily, analysts find themselves overwhelmed, with 90 percent of those alerts turning out to be false positives. This is where AI can step in and alleviate the workload. But how do we determine which alerts should be handled by AI and which ones require the expertise of a skilled financial analyst?
Unveiling the Graph
To answer this question, let’s paint a visual representation. Picture a graph with an X and Y axis. The Y axis traces the success rate, measuring the accuracy of prediction. The X axis represents the confidence score—an indicator of how certain the system is about an alert being real or false.
For AI, the performance curve typically follows this pattern: it shows high success rates for alerts with high confidence scores, while low confidence scores indicate lower success rates. In other words, when AI is confident, it excels, but it struggles when it’s uncertain. Humans, on the other hand, have performance curves that are slightly different. Although not as precise as a confident AI, humans tend to make better decisions when faced with uncertainty. At a 50 percent confidence level, humans are likely to outperform AI.
The Art of Human Decision Making
Why is this the case? When an AI is certain, it performs exceptionally well, surpassing humans in consistency and focus. AI remains unfazed by distractions. However, when AI encounters complex or statistically rare cases, it struggles. Humans, on the other hand, have the ability to gather additional information and context. They can look things up or consult colleagues, enhancing their decision-making process. When faced with uncertain alerts, it is often the human’s ability to embrace the unknown that grants them an edge.
Enter Augmented Intelligence
But this doesn’t have to be a choice solely between AI and humans. There is another option: augmented intelligence. Augmented intelligence combines the decision-making prowess of humans with the assistance of AI. The performance curve of augmented intelligence falls in between that of AI and humans. For alerts with somewhat low and high confidence scores, which represent a significant portion of predictions, augmented intelligence yields the highest success rate.
The Power of Presentation and Minimizing Bias
However, to fully leverage the potential of augmented intelligence, we need to address the complexities of human cognitive bias. How we present information from AI to human decision-makers significantly influences how effectively that information is utilized. There are two approaches to consider: forced display and optional display.
Forced display involves simultaneous presentation of an AI recommendation alongside a decision case. While this can lead to automation bias—where humans favor AI suggestions—the decision-maker might ignore contradictory information and rely solely on the AI’s recommendation. On the other hand, optional display only reveals the AI recommendation when requested by the human decision-maker. This approach overcomes automation bias, allowing individuals to form their own impressions before consulting AI recommendations.
Trust is also a vital factor to consider. When an AI recommendation is accompanied by an accuracy percentage, indicating the likelihood of correctness, humans are less likely to fully incorporate the AI recommendation into their decision-making process. We tend to avoid recommendations that openly admit the possibility of being wrong.
The Perfect Combination
To summarize, the question of who should make a decision—a human, AI, or human assisted by AI recommendation—is one that can be answered by moving from subjective decisions to quantifiable measurements. We can determine the most effective decision-maker for a given scenario. When the most effective decision-maker is a combination of human and AI, we have augmented intelligence. However, to minimize human cognitive bias in the decision-making process, the presentation of augmentation requires careful consideration.
When humans and AI algorithms join forces, something remarkable happens—we enhance decision-making outcomes. The key lies in knowing when to seek the assistance of the other. If you’re thirsty for more insights like this, visit Techal for a wealth of knowledge.
Feel free to drop us a line if you have any questions. And if you want to stay updated with the latest videos, don’t forget to like and subscribe. Thanks for joining us on this journey of unraveling the secrets of decision-making.
