Artificial intelligences are some of the most powerful tools created recently, and it's no wonder they are in the hype. There are many concrete applications; however, there are also many utopian promises. To know how to apply AI correctly, it is important to understand that they are just calculators that estimate the next word, image, or sound based on complex statistical analyses of millions or billions of parameters. In the current state of the art, AIs do not understand reality. Therefore, using them consciously is essential. Below, I present a guide on where it is safest and most reliable to apply AI and where it is more
🟢 Safe
When the activity involves manipulating information within a strongly pre-established context, AIs are a good choice, because, in these use cases, it is not necessary for the AIs to understand concepts of the real world or make judgments. All information and the relationships between them are contained in the initial prompt.
Examples:
✔ Summarize a text
✔ Classify text as positive or negative
✔ Highlight excerpts from the text
✔ Separate sections of a text, such as title, content, and footer
✔ etc
🟡 Apply with caution
When the activity involves manipulating information in contexts not fully described in the prompt, i.e., open questions with defined problems, attention is needed. AIs do not understand how the real world works and, therefore, are not capable of making deductions. If the database they consult does not specify the relationships established between objects in the real world, the AI will make mistakes. For this type of application, it is necessary to build specialized databases and perform extensive validations to ensure consistency in the results.
Examples:
✔ Search for and consolidate information
✔ Draw scenarios
✔ Suggest actions
✔ etc
🔴 Risky
When the activity involves seeking answers to undefined questions, it is basically impossible for the AI to provide accurate answers. Since it does not understand the real world and its interactions, it cannot act as a counselor.
