How social and philosophical assumptions shaped artificial intelligence explanation expectations

As a result of cogito, ergo sum. I believe. As a result, I am. Doomsayers have used this line to warn about the day when evolved AI will crush mankind under its enlightened and powerful feet. There is a heated debate concerning artificial intelligence (AI) and its intellectual potential, as well as how it may eventually overtake humans. The beliefs are based on numerous philosophical systems and reasoning. While the possibility of humanoids enslaving humanity appears remote (at least for the time being), let me walk you through the concepts that underpin its explainability. I previously discussed how relations and functions can be used to explain a trained model, and how this is applicable to anybody who has dealt with an AI agent.

Academic Master is a US based writing company that provides thousands of free essays to the students all over the World. If you want your essay written by a highly professional writers, then you are in a right place. We have hundreds of highly skilled writers working 24/7 to provide quality  essay writing services  to the students all over the World.

Explainable AI (XAI) encompasses more than just one agent revealing its underlying causes or another agent’s judgments. XAI solutions are more of a problem of human-agent interaction. The influence of social and philosophical dimensions on the AI solution, as well as human relationship with it, may be seen in the Venn diagram.

The author redrawn the image.

XAI aims to explain why a particular observation or event occurred. Users will be unable to comprehend the model’s observed behaviour or decisions unless they are aware of the decision’s rationale. Assume a trained model that correctly labels arthropod images based on criteria such as the number of eyes, legs, wings, and the existence of compound eyes and stingers.

Author’s redesigned image | Freepik.com’s insect icons

Now, as a user, you’d like to learn more about the labelling behaviour using an explanation agent, which we’ll call EA. Consider the following exchange between you and your EA:

You may wonder why the test image IMG 1.jpg is labelled as a Spider rather than a Beetle.

EA: Because the arthropod in IMG 1.jpg has eight legs, it belongs in the Spider group.

You: Why did you deduce that the arthropod in IMG 1.jpg had eight rather than six legs?

EA: As shown in the figure, the model tallied the number of legs. (Click here to see how discriminative photos are highlighted.)

You: Where did the model get the information that spiders have eight legs?

EA: Because practically all critters with eight legs were categorised as Spiders in the training set.

You: An octopus, on the other hand, can have eight legs. Why didn’t IMG 1.jpg get labelled as an octopus?

EA: Because the model’s sole purpose is to categorise arthropods.

As a result, explainable AI is more than just a list of connections and causes. It’s motivated by a desire for context based on the information the explainer has chosen to emphasise. The trained model had concentrated on the number of legs and other supporting characteristics in this case. However, the output was influenced by the fact that it was used to classify arthropods rather than all forms of multi-legged animals. Why did Andrew break the glass dish, to put it another way? The fact that he was a Jewish groom could be a favourable factor in his response. It’s also possible that Andrew worked as a bad dishwasher in a restaurant. As a result, the context is an important aspect of XAI.

offline data entry services are the secret sauce many organizations have used to improve customer experiences, innovate products, and disrupt entire industries. At CloudFactory, we’ve been providing data entry services for more than a decade for more than 360 organizations. We’ve developed the people, processes, and technology it takes to scale data entry without compromising quality

The relevance of context leads to the concept of contrastive explanation, which is a fascinating topic. The explanation is “interest relative,” as we saw in Andrew’s instance. We ask why questions and receive answers in order to have a deeper understanding. As a result, a contrastive study of the event can better highlight the relativity. A contrastive analysis comprises of the intended event and a foil, which is a contrasting event that did not occur. Andrew destroying the plate was a desirable outcome, however the foil may involve Andrew smashing a cup instead of the plate. A foil can be thought of as a way to focus the explanation by giving the why question structure. A fact alone is rarely sufficient; we must additionally define a foil. Furthermore, the factors that explain a phenomenon with respect to one foil will not always explain it with respect to another foil. This limits the number of possible explanations. Why do leaves turn yellow in November but not in January, for example, will evoke a different response from the foil than why do leaves turn yellow instead of blue in November. Foils are other (real or synthetic) examples in the dataset that are related to the example being explained in XAI; remember the octopus?

So now we may make our first remark about explainability philosophy: the why questions are contrastive. They are sought in response to specific hypothetical scenarios known as foils. That is, individuals do not inquire as to why event P occurred rather than why event P occurred instead of some other event Q. For XAI, this has significant social and computational implications.

When it comes to the reasoning for selecting an explanation, it is possible that there are as many causes of observation x as there are explanations for x. Consider a person who has an unlucky mishap due to the presence of a tall shrub at a curve. The physician would write down the victim’s cause of death as “many haemorrhages.” It will be referred to as “negligence on the part of the maintenance department” by the lawyer representing the victim’s family, and as “a flaw in the brake lock construction” by the other party involved in the accident, the carriage-builder. Although all of these statements are correct, the context of the question makes some of them more relevant than others. As a result, we arrive at the second statement concerning explainability philosophy: Explanations are chosen based on a bias.

In an explanation, there are three sorts of cognitive processes:

The causal connection is a technique for determining the origins of occurrences.

People utilise the explanation selection procedure to choose a limited subset of the identified causes as the explanation.

The explanation evaluation, which is the process of how a user judges the quality of an explanation.

The majority of studies reveals that people have cognitive biases that are created by their beliefs or roles that they employ to generate, choose, and evaluate explanations.

The explanation process is viewed as an interaction between two roles: explainer and explainee (occasionally the same person/agent plays both roles), with certain rules governing this interaction. To present an explanation in an understandable manner, it should have the following characteristics:

The data should be of excellent quality. A good explanation avoids saying things that are believed to be erroneous or for which there is insufficient evidence.

An explanation should contain the appropriate amount of information. That is, make its contribution as informative as necessary, but not more informative than necessary. Don’t we all despise the overwhelming results that recommenders provide?

An explanation must be pertinent.

Avoid uncertainty, chaos, and obfuscation.

The third assertion of explainability philosophy summarises these properties: Causal explanation is first and primarily a form of social interaction, and as such is bound by conversational conventions.