Disclaimer: After a few prompts, I asked ChatGPT to write an appropriately titled blog post about Microsoft’s comment that AI can give “usefully wrong” answers, with examples of “usefully wrong” information that humans are used to in various fields, and what humans can do to check “usefully wrong” information, with examples. The cover image was generated by DALL-E. Was this useful?!!

Artificial Intelligence (AI) has become a buzzword in recent years, and many people are excited about its potential to transform various industries. However, AI is not perfect, and Microsoft’s recent comment that AI can give “usefully wrong” answers has sparked a lot of discussion. In this article, we’ll explore what it means for an answer to be “usefully wrong” and why humans might be more comfortable with this concept than we initially thought. We’ll also discuss how to identify “usefully wrong” information and what we can do to verify it.

What is a “Usefully Wrong” answer?

A “usefully wrong” answer is an answer that may not be entirely accurate but can still be useful in certain contexts. For instance, if you ask an AI to translate a sentence from one language to another, the answer may not be entirely accurate, but it can still be useful in conveying the general meaning of the sentence. Similarly, if you ask an AI to generate a summary of a long document, the summary may not be entirely accurate, but it can still provide a general understanding of the content.

Examples of “Usefully Wrong” Answers

Humans are also used to “usefully wrong” answers in various contexts. For example, when we tell a white lie to spare someone’s feelings, we know that the answer may not be entirely accurate, but it can still be useful in maintaining a relationship. Similarly, when we give simplified explanations of complex topics, we may not provide all the details, but we still convey the general idea.

Here are some examples of fields where “usefully wrong” concepts can be useful:

Art: In art, artists often use “usefully wrong” concepts to create visual interest and emotional impact. For example, impressionist painters intentionally used loose brushstrokes and blurry details to create an impression of a scene rather than a realistic representation. This technique allowed them to convey a mood or feeling that was more important than capturing the precise details of the scene.

Music: In music, dissonance is a “usefully wrong” concept that can be used to create tension and interest. When two notes clash, it creates a dissonant sound that can be unsettling, but when resolved, it can provide a sense of resolution and satisfaction.

Comedy: In comedy, exaggeration and absurdity are “usefully wrong” concepts that can be used to create humor. For example, a comedian might tell a wildly exaggerated story that is clearly not entirely accurate, but it still provides a humorous commentary on the topic.

Science: In science, scientists often use models and simplifications that are “usefully wrong” to understand complex systems. For example, models of the solar system that show the planets as perfect spheres moving in perfect circles are not entirely accurate, but they can still be useful in providing a general understanding of the system.

In each of these fields, the “usefully wrong” concept provides a useful tool for achieving a specific goal, whether it’s creating emotional impact, tension, humor, or understanding complex systems. By recognizing these concepts, we can appreciate the power of imperfection and embrace the idea that not everything needs to be entirely accurate to be useful.

How to Check “Usefully Wrong” Information

It’s crucial to identify “usefully wrong” information and distinguish it from entirely inaccurate information. Here are some steps that can help:

1. Consider the context: Think about the situation in which the information was provided. Was it a complex problem, or was it a simple one? Was the information given quickly or after careful consideration? Understanding the context can help you determine whether the information that is not entirely accurate can still be useful.
   
2. Evaluate the source: Consider the person or system that provided the information. Is the source reliable and trustworthy? Have they been accurate in the past? Understanding the reliability of the source can help you determine whether the information that may not be entirely accurate can still be useful.
   
3. Compare to other information: Look for other sources or information that can help you evaluate the answer. Does the information align with what other sources are saying? If there are discrepancies, can you identify why they might exist and which source is more reliable?
   
4. Assess the consequences: Consider the consequences of relying on the information. Will it cause harm or lead to negative outcomes if it is incorrect? If the information is “usefully wrong,” what benefits might it provide that outweigh the potential risks?

Examples of Checking “Usefully Wrong” Information

1. Consider the context: Suppose you’re working on a project with a tight deadline, and a coworker suggests a solution that might not be entirely accurate but could help you save time. In this context, the “usefully wrong” answer might be the best option because it can help you meet the deadline.
2. Evaluate the source: Suppose you’re researching a topic online and come across conflicting information. One source is a well-known and respected scientific journal, while the other is a blog written by an unknown author. In this case, you should give more weight to the scientific journal as a reliable source and be cautious about blindly trusting the blog.
3. Compare to other information: Suppose you’re trying to solve a math problem and get an answer that seems “usefully wrong.” You can check your work by comparing it to other sources, such as a textbook or an online calculator. If your answer is close enough to the other sources, it might be “usefully wrong” in that it gives you a good enough approximation of the correct answer.
4. Assess the consequences: Suppose you’re driving and come to a fork in the road. A stranger tells you to take the left fork, which might not be the most direct route but will take you through a scenic area. In this case, the “usefully wrong” answer might be beneficial because it provides an enjoyable drive.

Conclusion

In conclusion, “usefully wrong” answers and information are prevalent in our world, whether they come from AI or humans. By understanding what it means for an answer to be “usefully wrong” and how to identify it, we can make better-informed decisions and avoid blindly trusting unreliable information.