“Artificial Intelligence brings with it opportunities for innovation and significant responsibilities.” As Google integrates generative AI into more of its products and services, the tech giant is taking a responsible approach, developing anticipatory safeguards, and embarking on a bold and reliable journey together with users and stakeholders. This article delves into this accountable approach, outlining Google’s efforts in ensuring safety, privacy, and fairness in their AI systems.
Building Protections into AI Products from the Outset
Google’s priority is anticipating and testing for a broad spectrum of safety and security risks. They are embedding protections into their generative AI features by default, guided by their AI Principles. One of these principles is protecting against unfair bias. In their machine-learning models, Google has developed tools and datasets to identify and mitigate unfair discrimination. This is an active area of research for their teams, and they have published several key papers on the topic. They also regularly seek third-party input to help account for societal context and assess training datasets for potential sources of unfair bias.
Another essential part of Google’s approach is its red-teaming programs. These programs enlist in-house and external experts to test for a broad spectrum of vulnerabilities and potential areas of abuse. These dedicated adversarial testing efforts help identify current and emerging risks, behaviors, and policy violations, enabling the Google team to mitigate them proactively.
Moreover, Google has implemented generative AI prohibited use policies that outline the harmful, inappropriate, misleading, or illegal content they do not allow. Their extensive system of classifiers is used to detect, prevent, and remove content that violates these policies.
Providing Additional Context for Generative AI Outputs
Google is making strides to provide context about the information produced by their models. They are adding new tools to help people evaluate information produced by their models and have added the ‘About this result‘ feature to generative AI in Search. They have also introduced new ways to help people double-check the responses they see in their product, Bard.
Safeguarding User Information
Google’s approach extends to the protection of user privacy and personal data. They are building AI products and experiences that are private by design. Many privacy protections they’ve had for years also apply to their generative AI tools. They have implemented privacy safeguards tailored to their fertile AI products, ensuring user content is not seen by human reviewers, used to show ads, or used to train their model.
Collaborating with Stakeholders to Shape the Future of AI
Google acknowledges that AI raises complex questions that no company can answer alone. They are actively collaborating with other companies, academic researchers, civil society, governments, and other stakeholders to promote the responsible development of AI models. They have also published dozens of research papers to share their expertise with the industry and are transparent about their progress on their commitments.