The introduction of artificial intelligence (AI) into the tech world brought endless innovation opportunities, and in 2021, it shows no signs of stopping. According to a McKinsey Global Survey titled, “The state of AI in 2020,” 50% of respondents report that their companies have adopted artificial intelligence (AI) in at least one business function. The adoption of AI systems continues to increase across all key industries.
But as AI’s pervasiveness grows, so do the concerns, specifically around the current black box approach being taken and lack of explainable AI (XAI). These concerns must be addressed, which means corporations will have to make a conscious shift to be more transparent about their AI systems’ decisions and explain the processes in simple terms for customers. This shift will allow AI technology to become more ingrained technology into key life-impacting processes.
Let’s take a moment to imagine AI systems as human interactions. If a bank director making critical loan decisions every day can’t justify his decisions when asked, he’s operating in a black box. Since his decisions are not visible to others, there’s no insight or accountability. While this isn’t a likely occurrence for humans, it is common in the world of AI.
Right now, experts across all industries are looking for ways to address the XAI dilemma. Companies can use Machine learning (ML) in two ways to address these challenges.
One option is to utilize a decision tree, an explainable ML algorithm that can be easily read. Decision trees can even imitate how humans make decisions by breaking the choice into many smaller sub-choices, making complicated processes more understandable.
A second option is to use the Shapley value. In game theory, the Shapley value assigns each feature a value that marks its importance in a specific ML prediction. In the ML setting, features often referred to as players each contribute a different amount to the final prediction. The goal is to compute the SHAP (SHapley Additive explanation) values for each prediction and see the contribution of each feature to understand why the system made a particular type of prediction.
When done right, these approaches provide transparency into outcomes generated by AI. As a result, various industries can use these approaches such as:
Banking and Financial Services
One way that AI can impact the banking and financial services industry is with fraud detection. AI/ML can digest historical fraud patterns, including highly sophisticated ones, and do so much faster than humans. With this historical knowledge, AI/ML can recognize and stop future cases from happening. The more data these systems take in, the more effective they become in protecting businesses and their customers.
Another area where AI can be impactful is in marketing. Using AI, teams can identify the right content for the right customers based on factors such as age, geographical location, and past actions. Through this detailed understanding, banks can replace SPAM communication with relevant and customized information. These insights can also anticipate a person’s needs to help boost cross-selling and up-selling opportunities. For example, they can identify which customers are parents and send them information about opening up a 529 account for their children’s future college expenses.
Healthcare & Life Sciences
When it comes to healthcare and life sciences, the potential use case for AI is incredible. Take diagnosis as an example. The National Cancer Institute states that AI technology in cancer care could improve the accuracy and speed of diagnosis, aid clinical decision-making, and lead to better health outcomes. What makes AI indispensable in this context is its ability to study patterns that are too subtle to the human eye and, when recognized, can guide doctors towards the correct life-saving treatment.
AI and specifically ML can also accelerate the drug discovery process. Take Google’s DeepMind, which introduced its AlphaFold program, that predicts protein structure and delivers computational forecasts reaching the quality standards of X-ray crystallography. By obtaining greater insights into the molecules that make up the human body, we can better understand the nature of diseases, which will allow us to develop new ways to treat them effectively.
Telecommunications
Telecommunication networks are becoming increasingly more complicated. That’s why, according to the McKinsey research referenced earlier, respondents from the telecom sector are among the most likely to report AI adoption.
The complexity is driven by an array of new services that telecoms are introducing. New services can fuel customer loyalty and open new revenue streams, but they can also increase the chances of costly downtimes and other performance issues. If there are too many service interruptions, the new offerings may lead to customers switching to the competition.
AI/ML can analyze these networks 24/7, looking for potential weaknesses. By identifying issues before they become problematic, AI/ML allow teams to act and avoid costly interruptions. Companies can also create self-optimizing networks to automatically optimize their network quality based on region and time zone traffic.
Transparency is key for the future of AI
There’s little debate about the impact that AI has and will continue to have on the world around us. To prepare for further AI/ML integration in the future, companies need to start moving away from the black box approach and implementing XAI. XAI will make the processes more transparent and consumers more willing and accepting of increased AI integration in everyday life. By allowing us to better understand the ‘why,’ we open the door to see how AI will continue to shape the world around us.