He Who Doesn't Know the Falcon Roasts It: Becoming a Falconer of AI

In 2021 falconry was inscribed on UNESCO's Representative List of the Intangible Cultural Heritage of Humanity, in a joint file involving 24 countries, Saudi Arabia among them. UNESCO describes what falconers do: breeding, training and caring for the birds and their equipment, and "the bonds between the falconer and the bird". In this heritage the falcon is a partnership built on knowledge and patience, not a bird of prey one simply owns.

People also repeat a saying: "He who doesn't know the falcon roasts it." We have not found a documented origin for it, but its meaning is plain: whoever misjudges the value of something ruins it while believing he is doing well. It describes many people's relationship with AI. Some fear it and leave it alone; others use it without understanding and spoil their own work.

First: know the limits of your bird

A falconer does not fly his falcon at every quarry. Intelligent tools have limits too, but they cannot be seen with the eye.

Researchers from Harvard and other institutions, working with Boston Consulting Group, ran a field experiment with 758 consultants. When participants used a language model on tasks within its capability, they completed 12.2% more tasks, 25.1% more quickly, and at higher quality. On a complex managerial task outside that capability, they were 19 percentage points less likely to reach the correct solution than colleagues who worked without the model. The researchers called these limits the "jagged technological frontier".

The lesson is that the danger lies neither in the tool nor in abandoning it, but in trusting an elegantly worded answer that lies beyond what the tool does well. Responsibility stays with the person using it: polished phrasing does not excuse him from checking. For why such errors occur, see the article "How Do Large Language Models (LLMs) Like ChatGPT Work?", and for the most common misconceptions, "AI Myths".

Second: train your hand before you release

Skill comes with practice. In a study by researchers at Stanford University and the Massachusetts Institute of Technology of 5,179 customer-support agents, a generative assistant raised productivity, measured as issues resolved per hour, by about 14% on average. The effect was about 34% for the least experienced and minimal for the most experienced.

So the benefit is not evenly spread, and no one knows where it falls for them until they try. A useful trial is small: one task, a clear measure of success, and a human review of the result. It is also wise to note what worked and what failed, so that your experience becomes a record you can revisit rather than an impression that fades.

Third: learn with your community

A falconer learns from falconers who came before him, and societies learn the same way. States have begun to treat AI literacy as a duty, not a luxury. Article 4 of the European Union's AI Act, in force since 2 February 2025, requires providers and deployers of AI systems to take measures to support the AI literacy of their staff, without requiring any specific level. In Saudi Arabia, the Ministry of Education reported in 2025 that the SAMAI initiative, launched by the Saudi Data and Artificial Intelligence Authority (SDAIA), had trained more than one million citizens.

Three exercises for this week

1. Ask the tool to summarise a text you know well, then compare to see where it went wrong.
2. Ask it about a subject whose answer you know, and notice what confidence looks like when it is mistaken.
3. Write yourself a list of what you will not hand to any tool: a financial decision, a medical question, and a religious ruling, on which one turns to scholars.

Whoever relies on God also takes the means: the glove, the training, the patience. We should treat this tool the same way, learning well and keeping responsibility in human hands.