Land the Plane!

When the pilot is unavailable, the copilot takes over and assumes responsibility. It is a safety measure, a fallback, a replacement in precarious situations, and a convenient load balancing on longer flights.

To this end, the copilot must be able to "safely land the plane" or in other words: he/she must be able to perform all the essential tasks of a pilot.

AI Copilots

The Microsoft Copilot system uses the term to suggest that the human "pilot" is still in control while the AI system only assists. But it also suggests that it might be a replacement. Otherwise, they would have called it "tool" or "assistant".

We covered in AI does not Work how AI systems cannot be responsible for anything. But without responsibility, there is no copilot.

The whole topic goes far beyond a failed analogy, misleading marketing, and the pushback that Microsoft received for their AI product. It touches on what we generally believe these systems to be capable of.

False Assumptions

If a person can talk eloquently, give convincing answers, show general knowledge, and even has special abilities - e.g. landing a plane in fair weather like board computer - we immediately make a number of assumptions and conclusions:

• This person has an excellent education.
• This person will not just lie out of the blue.
• This person will not crash the plane on purpose or for lack of care.
• This person will to some extent be able to handle unforeseen circumstances.
• This person has a self-awareness of its own abilities and will not perform activities outside its realm of competence.
• This person will not "guess" in important matters.

We make these assumptions because this is how the overwhelming majority of people are. This is how we are. This is how our friends are. We assume competence, self-awareness, and professional behavior.

None of these assumptions hold for AI systems.

The AI does not care

When being caught in a lie, we feel ashamed. This feels bad. So we avoid that in the future. AI never feels ashamed. It just lies and there is no consequence. The AI does not care.

We depend on our social network and if we would carelessly lie to our friends we would soon be alone. That would be frightening. The AI does not care.

When we are unsure about an important task, we stop. We think. We ask. We reason. We have an innate knowledge of when to rely on our own experience, when it is ok to "just guess", when we better think thoroughly, and when to ask for expert help. The AI does not care.

The AI does not care, and this is why it is a tool and not a copilot. A tool does not care. People care.

These false assumptions are so easy to fall for. I hope that by being aware of this human weakness, you might responsibly decide which tasks can be delegated to the mighty machine and which can't.