Heidy Khlaaf: Pioneering AI Safety at Trail of Bits

Heidy Khlaaf: Pioneering AI Safety and Security

Early Fascination with Robotics and AI

Heidy Khlaaf, an engineering director at Trail of Bits, was captivated by the potential of robotics and AI from a young age. She began programming at 15, driven by the idea of automating workloads in manufacturing and assisting the elderly. Khlaaf pursued a strong theoretical foundation in computer science, earning her Ph.D. from University College London and her BS from Florida State University, to make informed decisions about AI’s suitability and pitfalls.

Bridging the Gap in AI Safety

Khlaaf takes pride in applying her expertise in safety engineering and safety-critical systems to provide critical context to the emerging field of AI “safety.” She has deconstructed false narratives and offered concrete steps to bridge the safety gap in AI through her work, such as “Toward Comprehensive Risk Assessments and Assurance of AI-Based Systems” and “A Hazard Analysis Framework for Code Synthesis Large Language Models.”

Navigating Challenges in a Male-Dominated Industry

Despite the lack of significant progress in the tech industry’s gender balance, Khlaaf emphasizes the importance of acknowledging the current state and building a strong personal community for support. She notes that relying on DEI initiatives alone may not be sufficient, given the persistent bias and skepticism towards technical women in tech.

Advice for Women Entering AI

Khlaaf encourages women to trust their own scientific integrity and challenge claims that exaggerate AI’s capabilities, even if they contradict popular narratives. She stresses the importance of vocally expressing skepticism against unsubstantiated claims, despite the influence of imposter syndrome among junior women in the field.

Pressing Issues in AI Evolution

As AI advances, Khlaaf cautions against viewing it as a singular solution to all problems. She highlights the trend of forcefully integrating AI into every system, disregarding its limitations and potential for harm. Khlaaf emphasizes that AI should augment human capabilities rather than replace them, citing the example of ShotSpotter, an AI system that recently led to an officer firing at a child.

Awareness of AI’s Unreliability

Khlaaf underscores the inherent flaws and high error rates of AI algorithms, particularly in applications requiring precision, accuracy, and safety-criticality. She explains that AI systems embed human bias and discrimination in their outputs due to their reliance on statistical inferences and correlations from historical data, rather than reasoning or factual evidence.

Building AI Responsibly

To ensure the responsible development of AI, Khlaaf advocates for the creation of verifiable claims that protect people’s rights and safety while holding AI developers accountable. These claims should be scoped to specific applications and be falsifiable. She emphasizes the importance of independent regulators assessing AI systems, as is standard practice in other industries, to ensure public and consumer protection.

The Role of Investors in Responsible AI

Khlaaf encourages investors to engage with and fund organizations that establish and advance auditing practices for AI. She argues that relying solely on AI labs’ safety teams is insufficient and that independent auditors and regulators are crucial for maintaining public trust in the accuracy and integrity of assessments and regulatory outcomes.

Acknowledgment of how little the status quo has changed is not something we discuss often, but I believe is actually important for myself and other technical women to understand our position within the industry and hold a realistic view on the changes required.