Sometimes we see artificial intelligence outputs that mimic reality
What percentage can be trusted based on artificial intelligence results? Saving Changes...
Depends on what the specific AI model has been trained to do. With a very focused application and a high quality data set the percentage of correct results when inputs are within the range of the training data might be close to 100% but as soon as either the training data quality suffers or the input data is outside of the training data range or the scope of the application is broad, the percentage will drop.
Kiron Saving Changes...
Sergio Luis ConteHelping to create solutions for everyone| Worldwide based OrganizationsBuenos Aires, Argentina
AI is a boarder term. Generative AI created an "explosion" mainly when new generative AI model was published by google in 2017. Returning to your point: none. That´s because human in the loop is the key ingredient for success. With that said, when you will start to create an AI entity there are statical analysis to be performed to understand the confidence rate from the very begining. For example, based on the data provided to the AI entity for doing it work. Saving Changes...
As a rule of thumb, the Central Limit Theorem tells us that 30 is the magic number in statistics. That is the sample size at which you have enough data points where a 95% confidence level may be determined. That means your question must be compared to 30 or more reference sources to determine whether your question actually fits those references used to provide an answer.
Your question might be an exact match for something unequivocally deterministic based on the laws of physics and 1 source will tell you the right answer. If not a perfect fit the algorithm may be looking at clouds and seeing the shape of animals. How confident you are in the correctness depends on the training data used to answer a question and whether it is actually relevant or just linguistically similar. Saving Changes...
