Research and development into “XAI” (Explainable AI)—which aims to turn the AI ​​”black box” into a “white box”—is currently underway. XAI is a field where Japan’s research and development expertise can be a significant strength. It is a meaningful endeavor that promises not only to boost Japan’s international competitiveness but also to bring safety and peace of mind to all of humanity. (Kei Kitajima)

When an AI system operates as a “black box,” users can see the inputs and outputs but cannot understand the processes or judgments that generated those outputs.
This lies at the root of both the enthusiasm and the skepticism surrounding AI. Precisely because the inner workings are unknown, people harbor both high expectations and deep concerns.
The logical solution would be to transform the AI ​​black box into a white box, but this is easier said than done. Even if AI developers can observe what occurs at the input and output layers, they cannot fully grasp the events taking place within all the network layers situated between them.
These network layers within AI are known as “neural networks.” AI systems are structured in multiple layers, and neural networks serve as the core technology enabling them. Neural networks are a form of machine learning used to realize AI, designed to mimic the neural circuitry of the human brain.
However, much like the workings of the human brain itself, neural networks are said to be inherently opaque.

If this opacity can be resolved—making the system transparent—AI can be transformed into a white box. A white-box system allows users to understand how the AI ​​ingests data, processes it, and arrives at a conclusion.
This facilitates the verification and assessment of the reliability of AI-generated results, while also making it possible to adjust the model to refine its performance.
That said, turning AI into a white box is no simple task. Furthermore, it is argued that making AI transparent can lead to a decline in performance and a loss of flexibility. Nevertheless, research and development into making advanced AI explainable—known as XAI—is currently underway.
One example of this is LIME (Local Interpretable Model-agnostic Explanations). LIME analyzes the relationship between the inputs and outputs of “black-box” AI systems, aiming to identify the specific features that influence the output.

XAI is, in fact, an area where Japanese research expertise can demonstrate significant strength.
AI research and development is currently dominated by a fierce rivalry between the United States and China; in terms of the sheer number of research papers, institutions from these two nations account for the vast majority. Japan’s output is roughly one-tenth that of the US or China, meaning its presence appears minimal when judged solely by volume. However, in terms of research quality and impact, Japan is highly regarded and is said to maintain a standard on par with the US and China. Its uniqueness is particularly striking in the field of XAI, which seeks to resolve the “black-box” problem and visualize the rationale behind AI decisions.
For instance, NTT has developed “Rationale-Augmented Decoding” technology, which ensures consistency between an LLM’s reasoning rationale and its final output; in June 2026, two papers regarding this technology were accepted by CVPR (the IEEE/CVF Conference on Computer Vision and Pattern Recognition).
CVPR is a premier international conference in the field of computer vision.

The Japanese government has designated 17 strategic sectors and plans to drive a combined public- and private-sector investment exceeding 370 trillion yen by the 2040 fiscal year. AI is listed as one of these strategic sectors. Furthermore, based on tariff negotiations with the United States, Japan has pledged to provide a total of 550 billion dollars (approximately 80–90 trillion yen) in investments and loans to the US by 2029. If that is the case, it would be appropriate to make large-scale investments aimed at making AI “white-box”—transparent and interpretable—and to advance joint research and development projects between Japan and the United States.
Making advanced AI white-box is a meaningful endeavor that would not only boost Japan’s international competitiveness but also bring safety and security to all of humanity.
In fact, U.S. AI companies such as OpenAI and Anthropic have expressed concerns that AI spiraling out of control could destabilize social infrastructure, and they are calling for international regulation and governance. At the same time, both companies are focusing their efforts on research and development to make advanced AI white-box.
Now is the perfect opportunity to pursue the development of white-box AI.

※Translating Japanese articles into English with AI