Herman Cappelen - Making AI Intelligible: Philosophical Foundations

This is a book about some aspects of the philosophical foundations of Artificial Intelligence. Philosophy is relevant to many aspects of AI and we dont mean to cover all of them. Our focus is on one relatively underexplored question: Can philosophical theories of meaning, language, and content help us understand, explain, and maybe also improve AI systems? Our answer is Yes. To show this, we first articulate some pressing issues about how to interpret and explain the outputs we get from advanced AI systems. We then use philosophical theories to answer questions like the above.

Here is a brief story to illustrate how we use certain forms of artificial intelligence and how those uses raise pressing philosophical questions:

Lucie needs a mortgage to buy a new house. She logs onto her banks webpage, fills in a great deal of information about herself and her financial history, and also provides account names and passwords for all of her social media accounts. She submits this to the bank. In so doing, she gives the bank permission to access her credit score. Within a few minutes, she gets a message from her bank saying that her application has been declined. It has been declined because Lucies credit score is too low; its 550, which is considered very poor. No human beings were directly involved in this decision. The calculation of Lucies credit score was done by a very sophisticated form of artificial intelligence, called SmartCredit. A natural way to put it is that this AI system saysthat Lucie has a low credit score and on that basis, another part of the AI system decidesthat Lucie should not get a mortgage.

Its natural for Lucie to wonder where this number 550 came from. This is Lucies first question:

Lucies First Question. What does the output 550 that has been assigned to me mean?

The bank has a ready answer to that question: the number 550 is a credit score, which represents how credit-worthy Lucie is.