Psychological Perspectives on Trust in AI & Robots

If A believes that B will act in A’s best interest, and thus accepts vulnerability to B’s actions, then A trusts B

Two key aspects of trust are:

• anticipation or predictability

• vulnerability

reliance on or confidence in the dependability of someone or something

feeling that each party has when they can depend on the other party to do what they say they will

Trustor’s propensity to trust

Trust-relevant situation

Trustee’s trustworthiness

It is a stable indicidual trait

It has global effects on trust intentions and

The impact is most important at the beginning of interactions

Two scales to measure trust

• General trust scale –> questions which measure beliefs of honesty and trustworthiness

• Propensity to trust in technology scale –> questions which measure trustworthiness of technology

It is the trustors perception of the trustee

It could be described as the function of the interaction of the trustor and trustee

As these interactions mature the trustor relies and depends more and more on the behavior of the trustee

Situations which involve a certain vulnerability, meaning there is something at stake or risk

Without risk or being vulnarable there is no need to trust

AI-produced credit scoring may represent a risk to the loan officer

AI-supported diagnosis may inherent a risk for the patient to be misdiagnosed and not receive ideal form of therapy

Affective/social trust

• Integrity

• Benevolence

• The trustee is perceived as well-minded, warm-hearted and follows social norms

Cognitive trust

• Competence

• Understandablitity

• The trustee is perceived as competent, predictable and understanable

Cognitive trust heavily relies on the ability or competence of the trustee

It is formed by accumulated knowledge which makes it more easy and condident to make predictions whether the trustee can live up to its obligations

It is knowledge-driven, however it is important that there is some sort of lack of knowledge otherwise there would be no need for trust

Very important are familiriaty and propensity to trust which influence the three factors of perceived trustwortiness

Most of the connections between the different factors are emprical shown to be true

There are three factors of perceived trustworthiness which all result in trust in atomation

• Competence and reliability

• Understandablity and predictability

• intentions of the developers

  
  

The goal is that humans should approbiately trust technology and AI

There should be no over trust –> this leads to overestimation regarding the systems abilities

There should be no unapprobiate distrust –> this leads to reluctance or even fear, but of course there can be reasonable distrust

The subjectively perception should be aligned with the objectively trustworthiness of a system –> people have to know how to examine the trustworthiness of a system –> explainable AI

Research showed that unpredictable living beings, e.g. spiders lead to fear

To prevent this with machines or robots they should be predictable and understable

The intentions of a robot could be shown by auditive, visual or motor signals

In the scenario of a pedestrian crossing following questions are rising

• Has the (self-driving) car recognized me?

• Does it understand my intention to cross the street?

• Will it stop in time?

• Intention signaling needed!

Studies showed that the trust of humans rises if the intentions of a robot are clear and that humans could predict the motivations and intentions of a robot better if there have been signals which clarified the intentions of the robot

Trustworthiness is a often sited key desireable property of AI

There are ethical guidances regarding trustworthy AI, e.g. from the EU

The trustee must be perceived as well-minded and caring

It is based on the experience and feelings regarding the relationship to the trustee

The emotional-driven element makes it hard to objectively assess the risk

Since affective trust relies on emotions and integrity it is not really justifiable from an thical point of view to make AI systems more affective

Cognitive components are relevant in this matter

Users must be able to understand how competent an AI system is