Explainable Artificial Intelligence (XAI) is a developmental area of Artificial Intelligence (AI). The Defense Advanced Research Projects Agency (DARPA) is heavily invested in this project. AI and machine learning have come to the forefront of technology, andXAI is about explaining decisions in ways that you can more easily understand.
What is Explainable AI
Machine learning advanced AI to the forefront of technology. It made real-world applications more practical. DARPA believes the future of autonomous systems is profound. Machines will not only be capable of learning but of perception to decide and act on their own. A major limitation DARPA is working to overcome is the inability of AIs to explain their actions; explainable AI. Once humans can understand the rationale behind machine learning, the next evolution in AI is likely to take place.
Insight into the Explanations
Cognilytica Research stated that without explanations, AI would only be viable for trivial enterprises. As AI permeates more aspects of daily life, users will want to understand more about the ‘black box’. There are a few general areas that AI developers foresee users will want explained. The most basic question is why a choice was made over other options. Looking a bit deeper into the decision, users will likely want to know how a conclusion was made. Trust and error mitigation are other inquiries where users will want explanations.
Processes Used in Development
Around the world, researchers and scientists are collaborating to develop XAI. It is interesting that explanation techniques will play a role in producing machines more capable of explaining models and results. The machine-human relationship is being critically examined. A variety of prototypes are being explored. Some specific processesthese minds highlight for the development of XAI are:
- design data
- loss functions
- architectural layers
- model induction
- optimization technique
Classification and reinforcement learning are two areas in machine learning that need improvement. The Department of Defense looks to DARPA to advance intelligence and autonomous systems. The DARPA team plans to develop XAI within this paradigm. It is addressing two main challenges of machine learning problems. The first pertains to heterogeneous, multimedia data. The second looks at the autonomous system policies. Commercial applications are already being discussed.
Trust and Transparency Now
In a survey article, Holzinger, A., Biemann, C., Pattichis, C.S., and Kell, D.B. explored the implementation of AI for medical purposes. One pertinent expectation is the benefit to transparency and trust. Both of these concepts have become modern concerns. Pursuing the integration of these aspects early in the development of AI is important. Machine learning has grown by leaps and bounds. It is accelerating the advancements in AI, andit’s better to have transparency and trust built in on a fundamental level now. This is much preferred to needing to go back into the programming of an elaborate and ingrained system in the near future.
The minds developing XAI believe they will succeed in the near term. Medical and defense industries are already taking notice. The real-world applications are far-reaching and you can expect XAI to help realize ‘thinking’ machines.