Be On The Watch For Human-AI Communication Ambiguities In Vital Circumstances

Ambiguity in Communications

The mid-air collision has brought human-to-human communication ambiguities into the spotlight, but similar issues can arise in interactions with AI systems. The National Transportation Safety Board (NTSB) is investigating the crash, and early analyses suggest that communication ambiguities may have played a role.

In a transcribed excerpt from the ATC audio, the controller asked the military helicopter (designated PAT-25) if it had a specific passenger plane (indicated by the acronym CRJ) in sight. However, CRJ refers to a family of jets, creating ambiguity. The controller then directed the helicopter to pass behind the CRJ without specifying which plane.

Interpreting the Interaction

The controller’s communication was ambiguous because CRJ could refer to several planes. Similarly, the pilot’s response that an aircraft was in sight did not specify which one. The potential for confusion is clear, as the pilot might have been referring to a different CRJ far enough away to be considered safe, while the controller had a specific plane in mind.

This incident underscores the inherent risk in ambiguous communication, especially in aviation where precision is paramount.

Controller-Pilot Ambiguities

Missed communications between controllers and pilots are not uncommon. The challenge lies in recognizing when ambiguity exists and addressing it promptly. In high-pressure situations, both controllers and pilots might make assumptions, leading to misunderstandings. The pilot or controller may need to seek clarification or provide specific designations to ensure mutual understanding.

Overworked controllers and pilots face a significant cognitive load, which can exacerbate communication ambiguities. It is crucial for both parties to remain vigilant and cautious at all times.

Human-To-Human Versus Human-To-AI

The mid-air collision example involved human-to-human communication, but similar issues can occur with human-to-AI interactions. In an increasingly AI-driven world, understanding these communication dynamics is essential.

Two key questions arise: Do people using AI recognize potential ambiguities? Are developers of AI systems designed to handle ambiguity and seek clarification when it exists?

Examples of Generative AI Usage Ambiguities

To illustrate these points, let’s examine four scenarios using generative AI like ChatGPT. These examples range from low-risk situations to potentially life-threatening ones.

Human Is Ambiguous, AI Seeks Clarification

Suppose a user asks for help with a report without specifying the type of assistance required. The AI could respond with a request for clarification, preventing misunderstandings.

• User: “Can you help me with my report?”
• AI: “Of course! Could you specify the kind of help you need?”
• User: “I need help summarizing key points.”

This interaction demonstrates the AI’s ability to identify ambiguity and request clarification, which helps prevent potential errors.

AI Is Ambiguous, Human Seeks Clarification

Imagine a scenario where the AI provides a vague recommendation, and the user asks for more details.

• User: “What’s the best investment for me?”
• AI: “Real estate could be a great option.”
• User: “What do you mean by ‘great option’?”
• AI: “I based it on general trends. Would you like me to consider your risk tolerance?”

In this exchange, the AI’s response was ambiguous, prompting the user to seek clarity, which led to a more tailored recommendation.

Human Is Ambiguous, AI Is Ambiguous

The most concerning scenario is when both the user and AI are ambiguous, leading to potential misunderstandings.

• User: “Tell me about the car I’m considering.”
• AI: “The car is a good choice.”
• User: “I’ll buy it.”
• AI: “Great, let me know if you need details.”

Here, the user might be thinking of one car model, while the AI is referring to another. Without clarification, both parties could proceed with incorrect assumptions, leading to mistakes.

Human Is Clearcut, AI Is Clearcut

This scenario illustrates effective communication where both parties are clear and specific.

• User: “Show me key points from my meeting notes titled ‘Marketing Strategy’ in my Shared AI folder.”
• AI: “Based on your notes, the key points are: (1) Define your marketing goals, (2) Specify tangible metrics.”
• User: “Expand on the first point.”
• AI: “Marketing professionals must specify goals in writing.”

In this interaction, clear communication prevails, reducing the risk of misunderstanding.

Insights on Human-AI Communication

Effective communication in AI systems hinges on recognizing and addressing ambiguity. Natural languages are inherently ambiguous, and misunderstandings can occur in both human-to-human and human-to-AI interactions.

To mitigate risks, developers should design AI systems to seek clarification when ambiguity exists. Users must also play a role in ensuring that their instructions are clear and specific.

Regulations and laws might also be necessary to enforce accountability in managing ambiguities in AI interactions.

Conclusion

As AI technology advances, it is crucial to remain vigilant about communication ambiguities. Whether in aviation or everyday interactions, clear and precise communication is essential. Always be aware of the potential for misunderstandings and work to eliminate them.

As Adam Smith once said, “On the road from the City of Skepticism, I had to pass through the Valley of Ambiguity.” We can all benefit from being more cautious and attentive in our interactions, whether with humans or AI.