Last Updated on July 25, 2023 by Editorial Team
Author(s): Ivan Markočič
Originally published on Towards AI.
How empathy can help AI agents predict and understand human behavior
Artificial Intelligence (AI) has made it possible to simulate human-like cognitive abilities such as empathy, emotional recognition, and experience-based decision-making. This article will explore the concepts of empathy, emotions, and experience in AI and how they can be measured and used to improve AI’s capabilities.
- The Theory of Mind (ToM)
- What is Empathy?
- How to Measure Emotions?
- How to Measure Motivational Parameters?
- What is an Experience?
- How to Get Experiences?
- How Does Experiencing Events Work?
- How Does Empathy Work When Observing an Individual?
- Limitations and challenges
- Key takeaway
The Theory of Mind (ToM)
Research on ToM and empathy in AI is still in its early stages, but there are several studies and papers that provide valuable insights into the topic. For example, a recent paper by researchers at the University of Cambridge explored the potential for ToM in AI to improve social interactions and communication between humans and machines . Another study by researchers at the University of Southern California looked at how machine learning algorithms can be trained to recognize and respond to human emotions .
What is Empathy?
Empathy is the ability to understand and connect with the emotions of others . It’s a powerful tool that allows us to form deeper relationships with people and plays a crucial role in human interactions. With the advancements in Artificial Intelligence, machines are now capable of demonstrating empathy by predicting the emotional state and future actions of individuals.
By empathizing with others and putting ourselves in their shoes, we can imagine how we would react in similar situations. For example, if we were an antelope encountering a hungry and strong lion, we would instinctively feel the need to run away. On the other hand, if we were the lion, we would see the antelope as easy prey and choose to attack. Empathy helps us understand the actions of others and how they affect our own emotional and motivational state.
Furthermore, empathy provides us with significant information. It gives us a sense of how others might feel about an unfamiliar object even before we interact with it. If we see others expressing fear towards a snake, we would probably avoid playing with it.
How to Measure Emotions?
One’s emotions are greatly affected by the prevailing factor in their surroundings that catches their attention. Observing the emotions of those in our vicinity allows us to understand their innermost sentiments. As our actions are directly linked to our emotions, it’s crucial to remain mindful of them.
We are aware that our emotions have an impact on our behavior. However, the reverse is also true. By observing someone’s behavior, we can perceive their emotions.
By observing an individual’s body language and facial expressions , we can gain insight into their emotions. A person who smiles with an open body posture is likely to be feeling happy. Conversely, if someone is frowning with closed body language, they are likely feeling angry or sad.
How a person walks can often be a revealing indicator of their emotional state. Walking speed, posture, and facial expressions are common cues that convey feelings. For example, if someone is walking briskly with a frown, they may be angry or upset. On the other hand, if someone is strolling leisurely with a smile, they are likely feeling content or happy.
Researchers have developed a skeleton recognition program  to track the angular velocity of limb movements, providing a way to determine a person’s emotional state more accurately.
The tone of voice and the words used can also provide clues about a speaker’s emotional state. The speed of speech can indicate excitement, while slower speech may suggest sadness. Analytical techniques, such as FFT analysis and sentiment analysis, along with machine learning algorithms, can be used to identify emotions based on spoken words.
More about artificial emotions can be found in the following articles:
Theory of Mind AI: How Does Emotion Recognition Work?
How does measuring emotions enable personalized behavioral adaptation?
Theory of Mind AI: Are Goals and Actions Driven by Emotions?
How can emotions be harnessed to enhance AI-driven behaviors and outcomes?
How to Measure Motivational Parameters?
We can assess a person’s motivation by examining their actions and objectives, as these reflect their underlying motivational factors.
AI agents can interpret the outcome of actions in a variety of ways.
Event recognition involves analyzing the movement of an object using a skeleton recognition program. By analyzing the speed and direction of the subject’s limbs, we can draw conclusions about what they are doing. For example, if someone is turning their head, they may be searching for something. Alternatively, if someone is quickly approaching an object, it could be interpreted as an attack. We can then determine the level of motivation for each individual involved.
Intention recognition is the process of identifying what someone wants or is attempting to do, particularly in situations where communication is ineffective. It can be challenging to determine someone’s intentions, but by observing their actions and taking the context into account, we can make informed assumptions. If someone is holding a present, they are probably intending to give it to someone. Understanding others’ intentions is critical in developing effective communication and building strong relationships.
For more information, check out the article:
Theory of Mind AI: What is the next frontier in artificial intelligence?
Is the Numerical Representation of Human Needs, Desires, and Emotions a Revolutionary Approach to Understanding…
What is an Experience?
Our experiences  encompass a collection of events that have impacted our lives, both positively and negatively. These events vary greatly and can include interactions with others, personal achievements, and even challenging circumstances we have faced. It’s important to note that events are not random occurrences but rather the result of actions taken by either ourselves or others.
By taking note of the events that have transpired and understanding the sequence of actions that led to them, we gain valuable experience. Analyzing the patterns and consequences of past events can provide valuable insights into how to approach future situations and make more informed decisions.
It’s no surprise that experience is often regarded as a valuable asset, as it can inform and improve our future actions and outcomes. At its core, the experience can be viewed as a list of actions and their causes and consequences, providing us with a roadmap for navigating future challenges.
How to Get Experiences?
Artificial intelligence can gain experiences in two ways — either directly through its own sensors and processing or indirectly by observing the actions and data of humans or other machines, or by reviewing literature. Direct experience means processing and analyzing data first-hand through its sensors such as vision, audio, and touch. For instance, an AI designed to navigate through a maze will gain direct experience by using its sensors to understand the environment and move through it. On the other hand, indirect experience involves analyzing data collected from others such as studying the behavior of humans or other AI systems or reviewing data generated by them.
Experiences are essential for the development and growth of artificial intelligence. It is crucial to understand how experiences can be gained and to choose the most effective methods to improve AI’s capabilities. Developers use various techniques, such as supervised and unsupervised learning to train the AI system during its creation. Once the AI system is created, it starts gaining experiences through its operation, using various algorithms and techniques to process data and make decisions. As it performs tasks, it learns from the outcomes, and these experiences shape its future behavior.
For example, a chatbot used in customer service gains experience as it interacts with customers, answering their questions and resolving their issues. As it communicates with more customers, it learns how to respond to various inquiries and improves its ability to provide accurate and helpful responses. Experiences are crucial for AI systems to learn, adapt, and improve. The more experience an AI system gains, the better it becomes at performing tasks and making decisions.
LLMs are one of the possible ways to gather and process experiences. An LLM can act as a storyteller, creating stories from the sequence of events on the input. These stories can be later processed by other LLMs for different assessments and predictions.
How Does Experiencing Events Work?
Experiencing is essentially the evaluation of the impact of past, present, and future events on our own motivational parameters.
In the case of an AI agent, the interface for various sources of information can be streamlined so that the central processor does not differentiate between stories that come directly from the text or from the “storytelling” module that builds stories based on its own observations.
Past stories can be read from memory, and potential future stories can be obtained from the “generative module.”
To process the story, the “event module” is used first to dissect the story into events. The “evaluator” module then processes individual events, making an assessment of their impact on motivational parameters. The list of events is later sorted by their assessed value.
Based on these assessments, the best event to proceed with is selected from a range of possible future events. The worst possible future event is, of course, avoided.
To execute or prevent an event, a set of possible actions is read from memory. An action is chosen that has the optimal impact on motivational parameters.
Experiencing events is therefore, the processing of the stream of events, evaluating their impact on us (our motivational parameters), and seeking optimal actions to minimize or maximize the chances of a particular event occurring. Motivational parameters are, therefore, at the center of experiencing events. Their value depends on the events experienced and is assessed based on individual experiences. The actions that the AI agent will take depend on the value of these parameters.
How Does Empathy Work When Observing an Individual?
Empathy, at its core, involves experiencing what the observed individual is going through. When observing someone, we evaluate their motivational parameters and the emotions they display. Gathering information on the individual’s motivational parameters is a continuous process that leads to a better understanding of that person.
During the process of empathy, we temporarily set aside our own motivational parameters and emotions and load those of the observed individual into our memory. Based on these loaded parameters and emotions, along with our own experiences, we create a list of possible actions to take. We filter these actions based on their impact on the observed individual’s loaded motivators and choose the action with the optimal impact. This approach helps us determine the action that the observed individual is most likely to take at that moment, as we would if we were in their shoes.
After completing the empathy process, we reload our own motivational parameters and emotions regarding the observed individual. We evaluate the impact of the predicted action on our own motivational parameters. If the impact is positive, we will encourage the individual to take that action, but if it is negative, we will try to prevent them from doing so.
Using empathy, an AI agent can predict the desires and actions of users or third parties in their environment. This prediction is of great importance in various fields, such as robotics and automation, healthcare, marketing and advertising, education, and customer service.
Limitations and challenges
The Theory of Mind (ToM) and empathy are complex cognitive abilities that are challenging to incorporate into AI systems. Some of the limitations and challenges associated with integrating ToM and empathy into AI systems include:
- Lack of understanding of human behavior: There are still a lot of unknowns about human behavior. This makes it challenging to accurately model human behavior in AI systems.
- Interpretation of emotions: Emotions are complex and difficult to interpret. AI systems may struggle to accurately interpret human emotions and use them to guide decision-making.
- Contextual understanding: This can be challenging to achieve in AI systems, especially in complex and dynamic environments.
- Ethics and privacy concerns: Empathy is the collection of feelings and emotions of personal data, which raises ethical and privacy concerns. For example, AI systems that use empathy to predict human behavior could be used for nefarious purposes, such as manipulating people’s behavior.
The key takeaway from the text is that experiencing events involves evaluating their impact on our motivational parameters and seeking optimal actions to either minimize or maximize their occurrence. Empathy involves temporarily setting aside our own motivational parameters and emotions to evaluate those of the observed individual and make predictions about their actions. An AI agent can use empathy to predict the desires and actions of users or third parties, which is important in various fields such as robotics, healthcare, marketing, education, and customer service.
What is your opinion on the use of empathy by AI agents to predict the desires and actions of users or third parties in various fields such as robotics and automation, healthcare, marketing and advertising, education, and customer service?
Let’s start a discussion in the comments below!
Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.
Published via Towards AI