AI Hallucinations

No, these are not the hallucinations that one physically experiences in life…these are AI hallucinations… what are they? Let us see more in this post;

What are AI hallucinations?

AI hallucinations are misleading results generated by AI models. When AI perceives patterns that are not in existence and generates results in a totally nonsensical way for human observers, it is known as AI hallucination.

Why do AI hallucinations occur?

P.S: Is AI hallucinating in this picture?

It is obvious that these inaccurate results are generated because of misleading, biased and lopsided data that is fed to the AI engines. Since AI models learn from what is taught to it, if it is fed incorrect information, it will generate incorrect results. There are three types of AI hallucinations. They are – false positive(generating a positive when there is a negative), false negative(generating a negative when there is a positive) and incorrect results

How do we reduce AI hallucinations?

As a starting point, it is good to train the models with correct, diverse and unbiased data

Train the model with a template

We can query the AI model with the correct set of phrases…(will we get a AI query language in the future or is ‘prompt engineering’ already making inroads?)

It is good to verify all the outputs that AI generates for hallucinations.

References:

1. https://cloud.google.com/discover/what-are-ai-hallucinations#:~:text=AI%20hallucinations%20are%20incorrect%20or,used%20to%20train%20the%20model.
2. https://www.ibm.com/topics/ai-hallucinations

This post is for BlogchatterA2Z 2024!