SIGQ, PLUS, and SAMURAI ARCHITECTS Co-Host 'AI Business Development Study Session', Advocating Reliability Design in the Era of Physical AI

On April 27, 2026, PLUS Corporation, SAMURAI ARCHITECTS Inc., and SIGQ Co., Ltd. co-hosted an 'AI Business Development Study Session'. At this session, our CEO, Takaaki Kanechiku, took the stage to discuss the current state and future possibilities of AI from the perspectives of business development and physical AI. This article provides an overview of the lecture, which closely relates to SIGQ's mission of 'designing reliability.'

Business Scale in the AI Era Rivaling National Budgets

Kanechiku began by explaining the technical foundations of AI.

When regularly interacting with tools like ChatGPT, Claude, or Gemini, it's easy to conflate AI with LLMs (Large Language Models), but LLMs are merely one type of AI. Within the broad concept of AI exist deep learning and machine learning, and LLMs have emerged as a particularly prominent presence within these fields in recent years.

Why can LLMs return natural responses to human questions? LLMs learn how words connect and patterns according to context from massive amounts of data existing in the world, such as books and texts on the web. When text is input, the model performs probability calculations among numerous candidates to determine which word naturally follows next, outputting the most 'likely' (highest probability) word. By stacking this process, a coherent answer is generated. Services utilizing this mechanism are already incorporated into many business operations, such as summarizing contracts, translating, drafting emails, and creating presentation materials. Modern AI is supported by massive data processing.

Next, Kanechiku introduced the scale of business growth in the AI era. In a quiz format, he asked the audience how long and how much it took for Uber to achieve operating profitability. The answer was 14 years and approximately 4.5 trillion yen. Nurturing a business to the point where it can be called social infrastructure required the resolve to endure that long period and massive deficits. On the other hand, Anthropic, which develops Claude—a presence rivaling ChatGPT—has achieved an annualized revenue of 4.5 trillion yen in just five years since its founding. Furthermore, right before the lecture, news broke that Google made an additional investment of 6.4 trillion yen in Anthropic. Now, an amount comparable to 70-80% of Japan's annual defense budget is being poured into a single startup. Kanechiku also shared the realities of talent in Silicon Valley, gained through extensive overseas experience and interactions with international clients and tech communities, conveying the impact of an era where the advent of AI is rewriting the very growth curves of businesses.

The Potential of Physical AI Expanding into the Real World

Moreover, AI's sphere of activity is not limited to the digital space but is expanding into the real world. Based on this reality, Kanechiku introduced how physical AI is improving in accuracy and practicality, using the latest examples.

The first example is 'AGIBOT G2', a robot for the manufacturing industry developed by a Shanghai-based company. It handles precision tasks such as assembling and repairing tablet devices, where even a slight misalignment leads to fatal failure, adjusting its position in millimeters while working. It also features the ability to navigate autonomously around the factory as needed.

Its greatest feature is that it learns from its failures. Sometimes it receives human guidance, and other times it analyzes the causes itself and makes corrections, accumulating these experiences to improve accuracy. Kanechiku noted that practical application is progressing even in the production sites of precision instruments, where AI introduction was previously considered difficult, and it is expected to serve as a countermeasure against labor shortages caused by population decline.

The second is autonomous driving. While autonomous driving has been a researched technology for 15 to 20 years, it is approaching the stage of social implementation overseas. In San Francisco, driverless taxis are already running on the streets, and if you specify a destination on an app, it will take you there directly. He added that while Japan has also begun collecting driving data for autonomous vehicles, many challenges remain in complex environments like Japan's narrow residential roads.

The third is the table tennis robot 'Ace', developed by Sony. It uses cameras to capture where and with what spin the ball is flying, calculates the angle and spin required to return it in less than a second, and hits it back. Kanechiku introduced an anecdote that Ace won every match against professional table tennis players, emphasizing its advanced technological prowess. Mastering a sport that requires continuous judgment and action in a short period, like table tennis, is an event that symbolizes the broad application range of physical AI.

The evolution of physical AI indicates that AI is no longer just an entity that generates text and images, but is becoming an entity that judges and acts in real space. This is why the question of what happens when its behavior goes wrong is becoming increasingly weighty for business and society.

Designing Reliability in a Society Permeated by AI

During the Q&A session of the lecture, a striking question was raised by a participant: Since AI judges based on probability, errors will never be zero. While humans can review and correct documents, how do we guarantee responsibility in physical AI, which can involve human lives?

Kanechiku's answer to this was to incorporate 'fail-safe' engineering principles into physical AI. Fail-safe is a design philosophy where, if an incident such as a breakdown or operational error occurs in a system, it is designed to always operate safely rather than continuing to operate in a dangerous direction. For example, machines on a manufacturing floor stop operating when they detect an anomaly. Similarly, Kanechiku argued that physical AI must be designed to transition to safe operation when an incident occurs. In fact, in the field of autonomous driving, controls that prioritize safety, such as applying brakes when danger is detected, are incorporated. To trust AI's judgments, it is important not to assume it will always be perfect, but to have mechanisms in place that ensure safety even in situations where proper judgment cannot be made.

Actually, the concept of fail-safe is deeply connected to the knowledge SIGQ has cultivated regarding incident response and the continuous improvement of reliability.

At the core of SIGQ's business is 'Reliability Design'. Kanechiku himself has built a career as an engineer centered on massive data processing and SRE (Site Reliability Engineering) for about 10 years, while also engaging in research on databases and large-scale distributed systems. Such knowledge is utilized in professional services that support incident response and the development of the autonomous AI agent 'Incident Lake'. The intersection of system reliability and the massive data essential for AI is exactly SIGQ's business domain.

Now that AI has advanced into the real world and involves human lives and massive businesses, reliability design is no longer just a technical issue, but a fundamental requirement for business continuity.