Inside Rakuten AI: Ting Cai on AI innovation and responsibility
In this series, we sit down with Rakuten AI leaders for a deep dive into the stories behind this transformative technology and the inspiring individuals driving Rakuten’s vision of AI for all.
It is estimated that AI will contribute nearly 20 trillion US dollars to the global economy through 2030, accounting for 3.5% of GDP worldwide, according to a 2024 IDC report. As this exciting new technology makes its way into more and more sectors of society, promising game-changing benefits to both companies and end users alike, expectations around its potential are sky high. Yet this excitement also comes with some concerns.
In the Cisco 2024 Consumer Privacy Survey, while 63% of respondents agreed that AI can be useful in improving our lives, 78% believe that organizations have a responsibility to only use AI in an ethical manner.
As the architect of Rakuten’s AI strategy, development and global deployment, Rakuten Group Chief AI & Data Officer Ting Cai is on a mission to build foundational principles and practices that can guide society toward a better future – one where AI tools augment human capabilities and empower industries, while avoiding unintended consequences.
From vision to action in an AI-first landscape
“Generative AI is such a transformative technology that it will trigger profound changes, much like the Industrial Revolution, fundamentally changing the way we do things,” noted Cai. “We want to share the tremendous potential of AI with our customers and business partners, so they can reap the benefits of this technological revolution.”
“Integrity is the number one priority. Users must trust our services.”
Ting Cai, Rakuten Group Chief AI & Data Officer
With AI promising to unlock new value and opportunities, the need for companies to have a responsible AI strategy is more critical today than ever. A recent McKinsey survey of over 100 organizations with more than 50 million US dollars in annual revenue found that while 63% of respondents think implementation of generative AI is a high or very high priority, 91% don’t feel very prepared to do so in a responsible manner. Rakuten, however, is tackling both pieces with equal priority.
Applying “first principles” thinking to AI development
When crafting Rakuten’s AI strategy, Cai applies first principles thinking, a problem-solving approach that involves breaking down complex issues to their most fundamental components. The result is the Rakuten AI Foundation, a sequence of three building blocks upon which all AI development is based.
“Our AI journey really began with deep learning because it’s a fundamental and generic foundational technology applicable to various services. In our first wave, we focused on building a deep learning foundation and its applications such as search, because even large language models (LLMs) will need retrieval augmented generation (RAG) to connect with real time data such as pricing and inventory information,” he shared.
“We are transparent about what data we collect – and make it easy for users to control what they share and understand what benefits it brings to them.”
“Meanwhile, we created embeddings for people, places and things, such as for all Rakuten Ichiba products, and leverage those to connect products and services with users across the Rakuten Ecosystem in a more granular and meaningful way. With this strong foundation, we rapidly scaled up our services to now include semantic search in nine services and semantic recommendations in six applications, with plans to further integrate into our ads businesses,” he added.
Crafting the perfect flywheel for human + AI execution
“Our approach to execution is all about creating a dynamic flywheel that empowers AI and human intelligence to work synergistically.”
Cai describes the second wave of AI development as Rakuten AI for Business where Rakuten can leverage its domain expertise in many industries. “Our approach to execution is all about creating a dynamic flywheel that empowers AI and human intelligence to work synergistically. Each interaction in our ecosystem starts with AI and goes for review by domain experts who help train, evaluate and set critical guardrails for our AI models. Finally, we validate with the customer and get the necessary feedback to improve the entire experience. This feedback loop, enriched by trillions of customer interactions, empowers both our AI and our people on e-commerce, travel, finance and software engineering teams.”
This flywheel concept is both broad and highly applicable to specific domains. For instance, in customer support:
- AI chatbots handle inquiries with human monitoring
- When a human agent steps in, the AI assists by providing conversation history and searching the internet and datasets for relevant information, enabling the agent to continue the dialogue seamlessly
- This reduces time spent understanding previous interactions and enhances efficiency and UX
- Every interaction between AI and customers contributes to a service playbook, making future human interventions even more effective
Cai believes the third wave of AI-powered consumer applications will be a natural evolution of Rakuten’s unique ecosystem offering. “Our strategy is to embed the latest AI technology directly into existing apps, like Rakuten Link AI, so all users can benefit from it with a low barrier to entry and the highest level of safety.”
Establishing responsible AI governance and frameworks
Rakuten’s foundational strategy for responsible AI is to build ethics and safety into everything the company does, at every phase of design, development and deployment. This commitment comes to life in the AI Code of Ethics, the company’s playbook for building AI that works for everyone.
Rakuten’s approach to managing and supervising AI involves the following cross-functional programs:
- AI governance
- AI R&D
- AI security & safety
- Public relations & public policy
- Data utilization
To ensure clear oversight, the outcomes of AI-related discussions, including proposals and key decisions, are reported and discussed in the AI & Data Committee. Depending on the nature of the issue, further deliberation may be conducted at the Board of Directors and Corporate Management Meeting to ensure transparency and accountability in Rakuten’s AI governance.
In addition to internal governance, Rakuten is a member of the AI Governance Association, established in October 2024 by leading Japan-based companies with a mission to maximize the value of AI through appropriate risk management and establishment of AI governance norms.
“Rakuten’s Red Teams employ ‘ethical hacking,’ proactively simulating attacks and testing AI systems with challenging scenarios to uncover vulnerabilities, biases and ethical concerns. This rigorous testing and monitoring ensures AI safety 24/7.”
As a champion of responsible AI, Cai was a key architect and driver of Rakuten’s AI Code of Ethics released in July 2024 as a framework for building responsible AI. Key considerations include:
- How to ensure customer safety when using AI
- How to prevent harmful content from being generated by AI
- How to detect bias so AI recommendations are fair
“Integrity is the number one priority for us. Without trust, nobody will use our service,” Cai stressed. “We build trust through transparency, especially on how we collect and use customer data, making it clear that all data is used with consent to deliver more benefits back to users.”
Building and operating AI safely
Building and operating an emerging technology like AI also brings unique sets of challenges and novel threats that Cai’s teams must prevent or mitigate, including prompt injection, data drift and model drift.
“AI systems don’t just go through a one-time approval; we must continuously ensure that the ongoing usage is compliant with policies and standards.”
“I believe any business must take security seriously when launching products. We handle risk management and AI safety in a similar way to cybersecurity. In production, there is always a risk of security attacks when hackers try to steal data, algorithms or code. Some people may even try to trick AI models to produce harmful content. So, we must stay vigilant to detect any malicious threats in design and production,” he said.
Rakuten’s Red Teams employ “ethical hacking,” proactively simulating attacks and testing AI systems with challenging scenarios to uncover vulnerabilities, biases and ethical concerns. This rigorous testing and monitoring ensures AI safety 24/7.
Rolling out AI across an entire organization requires leadership to establish a roadmap and best practices. “First, it’s important to get current inventories for the entire company. We need visibility on what AI in-house and third-party tools are being used and any potential copyright or privacy risks,” Cai said.
Cai emphasizes that people are key to both building and implementation, “Beyond documented policies, we must have clear roles and responsibilities, safe access to data for all, and AI safety embedded into engineering and systems. AI systems don’t just go through a one-time approval; we must continuously ensure that the ongoing usage is compliant with policies and standards.”
Making AI a culture, not just a technology
Rakuten’s responsible AI is a commitment to building AI systems that uphold the company’s core values, protect human rights and safeguard user privacy. By staying true to Rakuten’s robust AI Code of Ethics, Cai and his teams are confidently blazing a trail while navigating risks and ensuring that every AI advancement propels the company towards a future where technology empowers, respects and elevates the user experience.