Photo by — @brunocervera

What is Explainable?

A short reflection on explanations and artificial intelligence

This may seem like an obvious question easy to answer. I am explaining what you already know. Explanations, after all, are something you are used to giving or being given. Still, let me ask you: what is an explanation?

An explanation is:

a statement or account that makes something clear.

Alternatively, or simultaneously:

a reason or justification given for an action or belief.

A statement that makes something clear.

Why did John walk across the street?

He was on his way to work.

This depends on what you want to know.

Did you want to know where John was heading?

Did you want to know why he crossed the street instead of using the pedestrian crossing?

An intended explanation requires, to some extent the right question, or phrasing a question correctly.

You could get a probable statement.

John was probably on his way to work.

Then again, if you have to guess and you do not know there could be a variety of possibilities, and these can be listed.

The explanation could describe why John decided to walk across the road at the given point of time given the situation or preconditions for crossing.

John walked across the street, because there were no cars on the road. It is probable that he felt safe given that the road is safe.

Given an interference or incident during the decision the explanation may differ.

John was hit by a car crossing the road. He was being careless not using a pedestrian crossing, it was dark and he wore nothing to distinguish himself from the darkness. The driver was drunk, but he stopped and called for an ambulance. John was driven to the hospital.

Explanation differs depending on situation. Still, one can attempt to predict situations.

Many discussions of these types revolve around responsibility.

Then, how do you explain love?

You do not have to answer, but think about it for a few seconds.

You might have seen an explanation.

Maybe a friend has tried to explain what love is.

Is it possible you might not have described it in the same way?

One explanation cannot apply unequivocally for all in some cases.

At least not in a social context.

Explainable AI refers to methods and techniques in the application of artificial intelligence technology such that the results of the solution can be understood by human experts.

In this manner it does not necessarily have to be a simple explanation. Understanding from a human expert may differ from someone who is not as familiar with, for example, AI.

Is it a clarifying statement or a justification?

There is certainly a need to justify a mistake in a given system.

At other times a statement could suffice.

An explanation has to consider the situational context of its implemented application.

Furthermore, AI applications must be able to explain how a decision was made and even why.

Yet as explained the perspectives on ‘why’ may very well differ.

Most humans are not able to understand exactly how they arrived at a decision.

Why did you grow a beard?

Why do you wear blue?

We could explain these easy with: I like beards, or I like the colour blue.

There could be deeper reasons why of course.

How did this automation process decide to deny you a mortgage based on the given input?

How does the YouTube recommendation system offering me conspiracy videos actually work?

Why did the automation in the financial system collapse?

An explanation can be a set of statements, but it can also establish or confirm rules.

An explanation is a set of statements usually constructed to describe a set of facts which clarifies the causes, context, and consequences of those facts. This description of the facts et cetera may establish rules or laws, and may clarify the existing rules or laws in relation to any objects, or phenomena examined

Explanations may aim to contribute understanding.

As such, XAI or explaining AI can be considered depending on the need for clarification, or laws for explanation.

This is #500daysofAI and you are reading article 392. I am writing one new article about or related to artificial intelligence every day for 500 days.

AI Policy and Ethics at Student at University of Copenhagen MSc in Social Data Science. All views are my own.