Empathy has so far been largely regarded as a 'differentiator' between computer and human. But is that right? In this first article in a series on the meaning of AI for customer contact, we explore the concept of empathy and look at the "empathic ability" of chatbot technology.
Jump directly to:
- Empathy: competence or personality trait?
- How does empathy or emotion recognition work in AI?
- What can we expect from AI?
1. Empathy: competence or personality trait?
Empathy is the ability to recognize, understand and appreciate how others feel, also indicated by the ability to put yourself in the position of another. Scientists disagree on whether empathy is a personality trait or a skill.
Some see empathy as an innate personality trait or even an innate (universal) human trait, others see empathy as a skill that can be developed and improved through feedback and practice.
Some researchers distinguish between cognitive empathy (the ability to understand and/or respond to the emotions of others - often considered a skill) and affective empathy (the ability to feel and respond to the emotions of others - often considered a personality trait).
Our ability to communicate and act empathetically as human beings can be both strengthened and weakened. Reinforced by, for example, awareness and learning techniques; weakened by, for example, upbringing or by psychological disorders such as a narcissistic or antisocial personality disorder or psychopathy.
Showing empathy is different from being empathetic
In principle, showing empathy does not require you to have empathic abilities. You can give the impression that you are empathetic by giving certain responses. Also in text-based communication – letters, e-mails and even a chatbot – you can let empathy sound through as an organization. In the latter cases you can speak of simulated empathy, especially if written communication is generated automatically.
Empathy in customer contact
It depends on the situation what level of empathy is required in customer contact. Sometimes it is enough to let people know you are sympathetic: conversations between customer contact employees and customers are aimed at solving problems and answering questions. Sometimes a customer contact person will have to go further, for example because understanding the context and acknowledging the emotions are necessary to continue in the conversation. A conversation 113 of Victim Support Netherlands will generally be different from a conversation about a package from an e-commerce company.
When an empathetic response is expected from a customer contact agent, they must go through a number of different steps. Think of assessing the situation, the context and the emotions that play a role. But the factual information and what someone says and how someone says it also play a role: the verbal and non-verbal information.
Debby Mureau argues in her column 'What is the difference between an employee and a voicebot?' that customers with a problem or complaint only start listening after investing in their emotional bank account. There must be polite acknowledgment for what they feel; only then can the complaint be dealt with effectively. In a telephone conversation, the customer experiences this recognition when he or she (unconsciously) perceives this in the voice of the employee. This concerns variables such as rhythm, volume, pitch and intonation.
2. How does empathy or emotion recognition work in AI?
Systems don't have the same emotional intelligence as humans – simply because computers do calculations and humans feel. But to what extent can systems copy the art of humans when giving an empathetic response?
It is certain that AI systems can empathize simulate. In fact, in one recent study the Level of Emotional Awareness Scale (LEAS) was used to evaluate ChatGPT's responses to different scenarios and compare its performance to that of humans. The AI chatbot not only outperformed the human average, but also showed significant improvement over time, the researchers said.
The question is whether systems will improve in terms of empathic performance or whether they will continue to suffer from their limitations. Those limitations are serious: AI systems have difficulty recognizing and interpreting subtle signals (e.g. emotions such as sarcasm) and taking into account context and culture.
On the other hand, technology convergence can enhance the capabilities of AI. This is how chatbots can handle emotion detection in the spoken voice, in spoken language and in written language (as the mentioned study with ChatGPT shows). Facial expressions are also increasingly involved.
How does ChatGPT arrive at an 'empathic response'?
ChatGPT works primarily on word order probabilities, but other techniques and methods are also used in output generation, including contextual information, transformer architecture, conditional generation, beam search, and top-k sampling. These techniques contribute to a better quality of that word order. This also looks at relationships between words, the weight and importance of words within a sentence for other words in that sentence, the most important part of a sentence, and various possible sentences are placed in a ranking.
Techniques that ChatGPT uses
Specifically to assess the right sentiment and process it in an appropriate output, ChatGPT-4 uses the following techniques for text-based communication:
- Sentiment analysis: ChatGPT uses sentiment analysis to understand the tone of the input and determine whether the input is positive, negative or neutral. This involves analyzing words and sentences to determine whether they are positive or negative. For example, if someone says “I am very happy with this new job”, ChatGPT will recognize the word “happy” as a positive emotion.
- Specific Dictionaries: ChatGPT uses dictionaries that categorize words based on their emotional connotations. The dictionary ANEW (Affective Norms for English Words) categorizes words on a 9-point scale from negative to positive. This allows ChatGPT to better 'understand' the emotional charge of words and generate the right, appropriate responses.
- Context around emotion: ChatGPT can take into account the context of the input to understand the emotions. For example, if someone says "I'm in a bad mood", ChatGPT can understand the context and generate the appropriate empathetic response, such as "I understand you're not feeling well today".
- Tone of Voice: ChatGPT can also take into account the tone of the input, for example if capital letters, exclamation marks or question marks are used. A sentence with exclamation marks can indicate that the person is excited or excited, while a sentence with question marks can indicate that the person is unsure.
3. What can we expect from AI?
Getting to computers – or facets of them – become easy assigned human characteristics. When the consumer is presented with a question on the computer, people feel obliged to 'help' the computer. The type of voice and the type of communication influence the way consumers perceive a brand. There is even a theory based on the idea that computers are social actors. From this perspective, it is not inconceivable that computers (or software) can take over parts of the empathic human act.
The big question is whether convergence of different technologies – the combination of separate, well-functioning technologies, where the combination leads to new applications – will actually take place. And besides, the question is whether you should want that.
When this path is taken – some companies are already here quite a way along the way – then there is a good chance that the customer contact employee of the future will (also) be given a role as 'file reviewer'. For example, to substantively assess the work that an AI solution has done, to prevent bias from playing a role or ethically unacceptable decisions being made. (Ziptone/editors)
Featured, Human Resources, Knowledge base


