Ms. Brainard offers an example of how credit rejections may be explained, provided that the regulators are willing to accept the explanation. A bank may input data that is similar to, but not identical to, the data of a customer whose application has been rejected, and see if a change in the data results in a change in the credit determination. For that to be an acceptable means for a bank to explain a credit rejection, the regulators must be willing to accept the possibility that it is not always obvious why a black box may view a particular bit of data as being negative or positive, or why that data may be viewed as positive or negative in combination with other data specific to that customer.