Explainability matters: How symbolic AI can transform business intelligence
Interested in using AI tools but concerned about the reliability of its solutions? Podcast guest Casper Wilstrup describes an AI approach which puts transparency at the forefront.
One of the main reasons why businesses are reluctant to adopt AI tools is the lack of trust in the solutions they provide. Rightfully so: how can you entrust your business processes to a system which seems to just magically come up with an answer but no explanation for it?
Our interview with Abzu CEO Casper Wilstrup features an AI approach with transparency embedded into its processes. Called symbolic AI, it avoids the black box problem concerning explainability and interpretability that AI tools reliant on deep learning typically have.
What is Symbolic AI?
Symbolic AI aims to mimic rational human thinking. Unlike more popular approaches such as generative AI which arrive at answers using patterns from enormous amounts of data, symbolic AI follows rules like mathematics and logic.
This AI tool represents knowledge through symbols and uses logical reasoning to derive conclusions or make decisions. By interpreting symbols and their relationships, symbolic AI systems imitate human-like thinking processes, enabling problem-solving and decision-making in various domains.
A significant advantage symbolic AI has over others is explainability. It can clearly show how it arrived at its answers and decisions. This means you can always look into its reasoning process, unlike other AI approaches where nobody knows exactly how the system got its answer.
Apart from avoiding the black box problem altogether, symbolic AI does not need massive amounts of data before it can be implemented. It solves problems by following a logical and explainable procedure, which can then be tested using minimal data. Since you can trace how the system arrived at the answer, you can also refine it further to achieve better results.
Why trust and comprehension go hand in hand
“You have to trust the result when you want to use AI for any kind of business application. And in order to trust, we humans need to understand.”
One of the biggest barriers in using AI solutions is the uncertainty on how it arrives at its answers. More popular AI approaches seem to just give the answers without saying exactly how it got there.
Symbolic AI systems provide clear explanations for their decisions and actions. This is crucial for businesses to understand and trust AI-driven processes, especially in regulated industries where transparency is paramount. Some applications where symbolic AI can provide more reliable solutions are as follows:
Logistics sector: Using supply chain rules and constraints, symbolic AI can help businesses minimize costs, reduce lead times and improve overall efficiency;
Legal sector: Symbolic AI can facilitate the implementation of a fair legal system by giving lawyers, judges and other relevant parties a clear picture of how it arrived at its conclusions using clear and transparent explanations on its decisions;
Healthcare sector: Systems powered by Symbolic AI can assist healthcare professionals in giving accurate diagnosis, recommending correct personalized treatment options and predicting patient outcomes based on clinical guidelines & medical knowledge; and,
Finance sector: Symbolic AI systems can help financial institutions detect fraud, mitigate risks and prevent losses by giving transparent information on its financial decision-making processes.
Is symbolic AI the better approach?
“Just like the human brain, we can build AI that is good at both.”
It depends on what you need for your business.
There are applications where symbolic AI may not necessarily be the most efficient or effective approach, such as for natural language processing. Language learning models excel at natural language understanding and generation tasks, including language translation, text summarization and conversational agents.
As discussed earlier, there are business problems which demand greater interpretability and explainability. Symbolic AI is the better approach if you need a system which allows you to understand its reasoning processes from start to finish.
Ultimately, what could be ideal is a mix of both approaches. As Casper describes in the episode, it would be great to have an AI system that is great in communicating naturally with humans while passing on the task of problem solving to symbolic AI processes.
