Column: Being aware of AI limitations

This column, written by Sandra van den Poll, partnership manager at JADS, was published before in the Food+Agribusiness magazine.

Prompt engineering is key to harnessing the power of AI. However, as with any powerful tool, there are risks. When I asked AI to come up with a summer recipe, it produced surprising flavor combinations and cooking times that were not very realistic.

Interacting with AI models may have far-reaching consequences. Data security and privacy are major risks. Prompts containing sensitive business data, such as profit margins or livestock health data, are sent to third-party servers. Not every AI system guarantees that this data will not be stored or used for training purposes. Another significant danger is the possibility of unreliable output. AI models may sound convincing, but generate factually incorrect information. Blindly accepting advice about a crop disease can lead to a failed harvest. In addition, there is the risk of bias. Because AI models are trained on data that may contain human biases, they may unintentionally perpetuate discriminatory patterns. If my AI had been trained this summer on data from recipes from Asia, for example, the suggested recipe may turn out to be much spicier than I like.

Dealing with these risks is not a luxury, but a necessity. So how can you ensure safety? First, choose reliable tools that explicitly state that they will not use your data for training purposes, and read the terms of use carefully. Anonymize sensitive data in your prompts: remove all personal and company information.

Never consider the output as the final product, but as a first draft. Always check the information against your own knowledge. And remain realistic; AI may analyze data and contribute ideas, but the final decision is yours. Understand the limitations of the technology and be aware that AI has no intuition, experience, or moral compass.

By being aware of these limitations and acting responsibly, you combine the best of both worlds: the efficiency and innovation of AI with the reliability and ethical responsibility of your own expertise. That is the best recipe for AI.