Broadly speaking, AI systems are used to replicate or emulate certain aspects of cognition. We interact with AI almost every day in modern life from the use of online search engines to video games, to digital assistants commonly accessed through smart phones and smart devices, to automated cruise control functions in vehicles.

The FBI uses the definition established in the Fiscal Year 2019 National Defense Authorization Act (FY19 NDAA), which provides: 

• Any artificial system that performs tasks under varying and unpredictable circumstances without significant human oversight, or that can learn from experience and improve performance when exposed to data sets.
• An artificial system developed in computer software, physical hardware, or other context that solves tasks requiring human-like perception, cognition, planning, learning, communication, or physical action.
• An artificial system designed to think or act like a human, including cognitive architectures and neural networks. 
• A set of techniques, including machine learning (ML), that is designed to approximate a cognitive task. 
• An artificial system designed to act rationally, including an intelligent software agent or embodied robot that achieves goals using perception, planning, reasoning, learning, communicating, decision making, and acting.  

Is Artificial Intelligence good or bad? 

The FBI recognizes the value add of artificial intelligence and machine learning technologies. AI changes the threat landscape by automating tasks that previously required more time, effort, and labor. The availability, low cost, and short time it takes to train new AI that learn from existing models, combined with the availability of large data sets and lack of regulation, create opportunities for bad actors to acquire and use these tools against U.S. interests.

The use of AI is not a crime. Criminals or other malicious actors can, however, use AI in the furtherance of criminal acts or acts that threaten U.S. national security. As the public and private industry adoption of AI increases, the AI/ML attack surface will correspondingly increase along with the potential physical and economic costs of a successful attack.

Our approach to AI fits into three different buckets—identifying and tracking adversarial and criminal use of AI, protecting American innovation, and FBI AI governance and ethical use.

• The FBI is focused on anticipating and defending against threats from those who use AI and ML to power malicious cyber activity, conduct fraud, propagate violent crimes, and threaten our national security; we’re working to stop actors who attack or degrade AI/ML systems being used for legitimate, lawful purposes.
• The FBI defends the innovators who are building the next generation of technology here in the U.S. from those who would steal it. This effort is also related to defense against malicious cyber activities, since all-too-often our adversaries are stealing our trade secrets, including AI, to turn it against the U.S. and U.S. interests.
• The FBI is also looking at how AI can help us further exercise our authorities to protect the American people—for instance, by triaging and prioritizing the complex and voluminous data we collect in our investigations, making sure we’re using those tools responsibly and ethically, under human control, and consistent with law and policy.

Synthetic Content is commonly referred to as deepfakes; however, synthetic content can refer to anonymized, artificial data.

Several factors have decreased the resources, time, and effort required to create or use convincing synthetic content. Methods once limited to those with the necessary computing power and expertise can now be employed by a broader customer base via user-friendly applications. One consequence of this trend is that synthetic content creation has been essentially commoditized and scaled beyond once limited use cases.

Content consumers can look for visual distortions and warping in images and video such as unsettling silences and distorted decibels, video inconsistencies or unnatural movement, and poor video, lighting and audio quality. These inconsistencies may be indicators of synthetic images, particularly in social media profile avatars.