In an era where generative artificial intelligence (AI) is transforming industries and reshaping the way we interact with technology, businesses are facing a profound challenge. As they harness the power of AI, they must also grapple with the age-old dilemma of preventing technology from being misused for malicious purposes. This challenge closely mirrors the enduring struggle faced by social media platforms: how to maintain a safe and responsible digital environment.
In response to this pressing concern, forward-thinking organizations are adopting a multifaceted approach that combines software-based “guardrails” with the invaluable input of human moderators. This strategy aims to ensure that AI applications remain within prescribed bounds, striking a delicate balance between the capabilities of AI and the necessity for ethical and responsible content generation.
The Rise of Generative AI and Its Challenges
At the heart of this evolving landscape lies the emergence of generative AI models, such as OpenAI’s GPT-4. These powerful models are trained on vast and diverse datasets sourced from the internet. When provided with appropriate prompts, they exhibit the remarkable ability to generate content. However, this capability comes with a significant challenge: the potential to produce content that may delve into the darker and more objectionable corners of the web. As a result, the need for effective content moderation has become paramount.
Consider, for instance, the scenario where a large language model generates toxic or harmful content. This underscores the importance of proactive moderation, both during the AI model training phase and in real-time assessment of the content it produces.
Intuit’s Innovative Approach to AI Moderation
A notable exemplar of this evolving landscape is Intuit, a leading financial software company headquartered in Mountain View, California. Intuit recently introduced “Intuit Assist,” a generative AI-based assistant that provides customers with financial recommendations. This AI solution relies on large language models trained not only on internet data but also on Intuit’s proprietary data sources.
In recognition of the potential risks associated with AI-generated content, Intuit has embarked on an innovative and forward-thinking approach. The company is in the process of assembling a dedicated team of eight full-time human moderators responsible for reviewing the inputs and outputs of its AI-powered system.
Atticus Tysen, Chief Information Security Officer at Intuit, emphasizes the pivotal role of human oversight in the AI loop. Particularly as the company endeavors to provide precise financial answers to its customers, the presence of human moderators adds a layer of scrutiny and accountability. It ensures that objectionable or harmful content is promptly addressed.
Intuit’s approach incorporates a multi-tiered content moderation system. At its core, a separate large language model is employed to automatically flag content deemed objectionable. This includes identifying profanity and content unrelated to financial guidance. Additionally, the system is equipped to counteract malicious activities, such as prompt injection attacks, which aim to manipulate the AI into disclosing sensitive information. For example, an attacker might seek to trick a chatbot into revealing confidential customer data or the inner workings of the AI system.
The Synergy of Human Moderators and AI
Human moderators play a pivotal role in the content moderation process at Intuit. They are alerted to review flagged content, enhancing the system’s ability to detect and address harmful content effectively. This collaborative approach between human moderators and AI-driven systems represents a powerful example of how technology and human expertise can work in synergy to safeguard digital environments.
Furthermore, Intuit is committed to involving its customers in the content moderation process. Customers are encouraged to report instances where they believe their prompts were inaccurately flagged or if they perceive that the AI assistant has generated inappropriate content. This proactive engagement with users is a testament to Intuit’s commitment to transparency and user involvement.
AI Content Moderation Beyond Intuit
While there are no specialized firms dedicated exclusively to AI content moderation at present, Intuit is augmenting its workforce by enlisting contractors with expertise in moderating social media content. These AI content moderators represent a novel category of job opportunities arising from the expansion of AI technologies.
The integration of human moderators into AI-driven systems is not limited to Intuit. It mirrors strategies adopted by social media giants such as Meta, the parent company of Facebook and Instagram. These platforms have long relied on an outsourced team of human moderators to curate and filter out offensive and harmful content. They serve as both a reference point and a cautionary tale for AI-driven content moderation.
In recent years, AI-focused companies like OpenAI have recognized the need for human intervention. They have hired human workers to evaluate and categorize harmful text sourced from the internet and generated by AI itself. These categorized passages have played a pivotal role in developing AI safety filters, preventing users of AI systems from being exposed to content that violates ethical and safety standards.
OpenAI has also collaborated with Microsoft, its principal partner and major supporter, to create the Azure AI Content Safety service. This service deploys AI to automatically identify and flag “unsafe” images and text, including categories such as hate speech, violence, and self-harm content. Microsoft has adopted this safety service across its generative AI tools, including GitHub Copilot and Office applications.
Eric Boyd, Corporate Vice President of Microsoft’s AI platform, underscores the potential of AI systems to perform a wide array of tasks effectively, given the right instructions. This highlights the versatility of AI technologies, which extend beyond content moderation to various applications.
Responsible AI: A Growing Imperative
Across the technology sector, leaders are exploring the potential for human moderation or investing in third-party software solutions like Microsoft’s offering. Analysts and industry experts predict that content safety filters will soon become a prerequisite for enterprises seeking to leverage generative AI-based tools.
Larry Pickett, Chief Information and Digital Officer of Syneos Health, a biopharmaceutical services company based in Morrisville, North Carolina, acknowledges the merit of having monitoring and oversight in place. While the organization is considering hiring content moderators in the near future, its current approach involves reviewing AI model training data on a case-by-case basis, incorporating human feedback as deemed necessary.
The rise of “responsible AI” is gaining momentum. This movement aims to make AI algorithms more transparent, auditable, and capable of reducing unintended negative consequences. Brandon Purcell, a Forrester analyst specializing in responsible and ethical AI use, observes a growing interest in responsible AI practices across industries. This interest is fueled by the recognition that failing to implement AI responsibly can expose organizations to reputational risks, regulatory challenges, and potential revenue loss.
The Future of AI and Human Moderation
As we navigate the intersection of AI and human moderation, the symbiotic relationship between these two elements emerges as a defining feature of responsible AI deployment. In an ever-evolving landscape, the collaborative efforts of technology and human expertise remain crucial in ensuring the responsible and ethical use of AI-powered systems.
The future holds intriguing possibilities, with AI systems evolving to become more capable and sophisticated. While AI may gradually assume a more prominent role in content moderation, the irreplaceable value of human judgment, intuition, and ethical consideration will continue to play a central role.
As we explore the ever-expanding frontier of AI technologies, the integration of AI and human moderation serves as a compelling narrative of our capacity to harness technology for the betterment of society while upholding ethical standards and responsible practices.
