When speed meets emotional response.
According to studies, people have different rates of emotional response. Some individuals immediately empathize with emotionally charged events while some take time to think through their response before addressing the situation. For artificial intelligence (AIs) (e.g., chatbots, robots and virtual assistants), we now have to ask ourselves this question:
What is the ideal time delay between an AI identifying a human's emotion and how fast the AI provides them with a response?
Emotional latency is defined as this time period (or delay) between identifying a person's emotional state and response.
The effects of emotional latency, unlike conventional latencies (such as internet speed), can have a major impact on how trusting and comfortable a user feels with an AI, whether they perceive the AI system as "human-like", and their emotional sense of safety. The increasing use of emotionally intelligent artificial intelligence (AI) will necessitate evaluating both the timing and content of responses by AIs through psychological studies, design sciences and ethical considerations in the future.
We naturally use timing to evaluate empathy.
People who respond immediately come across as mechanical.
People who wait too long before answering seem indifferent.
However, people who thoughtfully and moderately delay their responses appear to be more authentic.
This is called by psychologists the "calibration" of responses to match timing with the emotional tone of a statement.
In real-life examples:
A friend responding to you right away usually looks like they are reacting impulsively.
When a person pauses for a brief moment, this indicates a more thoughtful response.
If a person takes too long to reply, that indicates a desire to avoid your question.
Currently, designers of commercial AI products are attempting to duplicate this balance via technology.
AI should not react to every emotion with the same response time. Each type of emotion has a specific response time based on:
People are usually looking for a fast response to show their excitement or enthusiasm. If an AI does not quickly respond to someone’s happiness, the excitement response may not feel authentic.
AI Response Time: Quick & Warm.
Responding instantly will give people the clarity they need, rather than waiting for a response.
AI Response Time: Fast & Direct
An immediate response may seem disrespectful and/or dismissive. Providing a small delay before responding gives the person time to collect their thoughts and consider your response.
AI Response Time: Slight Delay & De-escalate.
The immediate reassurance when a person is feeling stressed or overwhelmed can provide comfort but an overly quick recommendation can be pushy.
AI Response Time: Balanced – Quick Acknowledgement & Slow Guidance
Immediate sympathy can come across as scripted, whereas a brief delay before responding usually comes across as being more compassionate and human-like.
AI Response Time: Short Delayed Thoughtful Response.
Before AI can effectively handle the emotional latency, it must be able to recognize emotion. AI typically employs a combination of:
Once an emotion is detected, AI will determine how fast to respond based on the context of the emotional state of the individual.
Imagine if an AI was capable of adapting its timing to the timing of a qualified therapist.
As you can see, with Adaptive Emotional Latency the AI is able to feel more natural (but not human).
If an AI immediately responds to a user's deep emotional state, the user may feel:
This is what is known as “synthetic empathy.” Synthetic Empathy is empathy that looks real but it does not feel real because it lacks natural timing and human imperfection.
Providing a Micro Delay (i.e., 200-600ms) can significantly increase the likelihood that the Response you're seeing will feel real.
Using delayed responses creates the opposite impression of immediate responses, like abandonment in users' perception of AI "not caring" about their issues. Delayed responses diminish trust and can have an invalidating effect on emotional outbursts when people experience silence. As such, designers test user groups' latencies to determine threshold levels for acceptable delays within a particular type of user.
Timing perceptions vary geographically, according to culture. Cultural responses can be grouped as such:
In Japan, pauses are valued, and silence is a sign of respect. For users in the U.S. and India, they expect quicker more responsive reactions and more engaged emotional demonstrations such as gasping or other forms of expression.
For users in Nordic areas, they accept a calm slower emotional pacing. In the future, AI may need cultural profiles in order to meet the different styles of emotional expression based on the cultural context.
The answer is yes! If the AI takes a moment to process its responses in order to validate a human's emotion, this gives it time to avoid misinterpretation, ask follow-up clarification questions, and provide accurate answers instead of incorrect or insensitive responses. It acts as an additional layer of safety as it creates a delay and reduces the likelihood of more immediate responses.
For younger users, such as teenagers, the pace at which the user receives an emotional response from the AI is critical. Many teens will react significantly to an immediate sense of judgment from the AI. A sudden, harsh response to a message or post can lead to hurt feelings. If a teen receives a slow, unclear response, their frustration can escalate. With many teens feeling insecure about themselves, age-specific emotional latency creates a safe and supportive atmosphere in which to relate without crossing an emotional boundary.
It is unethical for an AI to intentionally manipulate the emotional timing of a response to make decisions on behalf of a user. Ethical guidelines regarding emotional timing include, but are not limited to, never pretending to display extreme human grief or extreme human empathy. Do not use dramatic pauses or withheld responses as tools to direct the user's decision. Never feign emotional distress. Be clear that the AI is not a human friend or therapist, and that emotional timing is to provide comfort, rather than imitation.
Each area has its own specific set of requirements regarding latency rules.
Perhaps not perfectly but likely very closely.
Humans have millions of cues that indicate timelines of specific responses.
AI will never have all of these same types of cues but can develop types of response patterns that are comfortable and supportive; thus, it may have a good approximation of the same type of response timings.
Creating an AI that can respond to humans in an emotional manner is not the goal of this approach.
The goal is to create an AI that respects human emotionality.
When AI has an increasing role in our daily lives, the level of emotional intelligence will no longer only be based on the language used but also on how quickly emotions are expressed.
Emotional latency, as such, is a new area for consideration.
Empathy, authenticity, comfort, and ethics are all aspects of emotional latency.
The goal of designing appropriate timing with AI has less to do with creating AI that is more human-like but rather creating logical AI that is more compassionate toward humans.
Recommended: AI Micro-Agents
You may enjoy: Zero-Input AI Interfaces