Author: Saima Saleem

From the last few years, AI became very popular in every field of life. It not only handles large amount of data which are not possibly managed and interpreted by human but also helps in decision making and complex computations. By using AI, we can save time and work force. AI returns high accuracy and precision using its advanced models like Deep Neural Networks. AI can process different types of data and returns the answer to the query without explanation. There are few things which are decided by human like which model is most suitable and which features are more relevant. The ML models are selected based on the types and behavior of data. For model selection and feature engineering the person should have complete internal knowledge of these complex ML models. Whereas, these models work like the black box model user only knows the input and output. Users don’t have the understanding why or how these results are computed. There was a need for such algorithms which can resolve these ambiguities and make these models transparent. So that users can understand the underlying decision-making processes of the systems and the bias of algorithm. These models will become more trustworthy and can be confidently used in various high-stakes situations like medical, defense and cybersecurity where we can’t bear even a small amount of risk or uncertainty. McKinsey found that improving the explainability of systems led to increased technology adoption. [1]

Explainable AI XAI is the solution to this problem. It is the process of explaining complex models and making them transparent besides the high performance (prediction and accuracy). Which increases the usage of complex ML models and makes them trustworthy. It will also improve the ML models‘ results due to a good understanding of features relation with the output and their biases. Thus, XAI is recognized as an essential element for the functional sending of AI models in frameworks and more importantly, for fulfilling the rights of AI clients connected with decision making. As shown in figure 1, in the first part learned function only the output “this is a cat” which is simple AI whereas in the second part the function also returned the reason” it has fur, whiskers and claws” along the output “this is a cat”.  Besides the benefits of XAI there are some significant challenges like the tradeoff between algorithm simplicity and model complexity [2]. How much explanation is required and how to explain is also a big challenge because XAI provide explanation in context of ml model which is not understandable for AI user. So, the explanations should be relevant to its application’s context. It is concluded that XAI addresses the advancement of AI, and offers valuable open doors for enterprises to make AI applications that are trusted, straightforward, unprejudiced, and advocated [3].  XAI is the major issue of information systems research, which opens a number of interesting but challenging questions to investigate.

References

[1]	V. TURRI, „What is Explainable AI?,“ Software Engineering Institute Blog, 17 JANUARY 2022. [Online]. Available: https://insights.sei.cmu.edu/blog/what-is-explainable-ai/. [Accessed 21 April 2022].
[2]	A. T. ,. M. Rafia Inam, „Explainable AI – how humans can trust AI,“ Ericsson White Paper, April 2021. [Online]. Available: https://www.ericsson.com/en/reports-and-papers/white-papers/explainable-ai–how-humans-can-trust-ai. [Accessed 21 April 2022].
[3]	S. Clark, „4 Reasons Why Explainable AI Is the Future of AI,“ cmswire, 27 Sep 2021. [Online]. Available: https://www.cmswire.com/digital-experience/4-reasons-why-explainable-ai-is-the-future-of-ai/. [Accessed 22 April 2022].
[4]	A. A.-R. Ser, „Explainable Artificial Intelligence (XAI): Concepts, taxonomies, opportunities and challenges toward responsible AI,“ Information Fusion, vol. 58, pp. 82-115, 2020.
[5]	„Introduction to Vertex Explainable AI,“ Google Cloud, [Online]. Available: https://cloud.google.com/vertex-ai/docs/explainable-ai/overview. [Accessed 20 April 2022].
[6]	D. M. Turek, „Explainable Artificial Intelligence (XAI),“ Defense Advanced Research Projects Agency , [Online]. Available: https://www.darpa.mil/program/explainable-artificial-intelligence. [Accessed 20 April 2022].
[7]	„Read how explainable AI benefits production AI,“ IBM, [Online]. Available: https://www.ibm.com/watson/explainable-ai. [Accessed 20 April 2022].