In today’s column, I explore the human-to-human psychological phenomenon known as mirroring and reveal that modern-day generative AI can also partake in mirroring behaviors too. This is part of my ongoing Forbes.com column coverage on the ways in which AI can impact human cognition and mental health (see my comprehensive overview at the link here).
Let’s get underway on this intriguing topic.
Humans Mirror Each Other
I’m sure you’ve seen or experienced mirroring from time to time.
The idea of mirroring is that a person imitates another person while they are interacting with each other. One person raises their eyebrows, the other person might do the same. If one of the two gets louder and raises their hands, the other might do the same. And so on.
The most obvious mirroring involves gestures, facial expressions, and other physical motions and movements. Those you can readily see. Another means of mirroring involves what people say and how they say what they say.
It goes like this. Suppose you are normally a calm person and only speak sparingly. You meet another person who is very animated and talks incessantly. During your interaction, you may start to become more animated and begin talking with a greater frequency than normal. You are mirroring the other person.
Of course, they potentially can do the same about you. If they sense that you are quiet and reserved, there is a chance they might begin to mirror those same attributes. That being said, our natural inclination and how long we’ve been acting a certain way is bound to limit the amount of mirroring that takes place. You can’t suddenly turn an orange into an apple.
Mirroring Includes Words And Language Usage
The vocabulary that someone uses can be mirrored. Someone that you are conversing with is using slang from their region of the country. You might find yourself surprisingly adopting that slang and employing it during the conversation. Later, when you are conversing with someone else, you might have completely dropped the slang. The mirroring happened only when you were with the person who was making use of their slang.
There are heated debates about whether we do mirroring via a subconscious means or instead do so when aware and consciously intending to do so. I vote that we do both. Not necessarily at the same time.
The temporary adoption of someone’s slang words might happen without you fully being cognizant that you are doing so. It would be a shocking realization if you watched a video of your interaction and observed that you started to use the slang. In that instance, you were presumably subconsciously doing mirroring.
In contrast, mirroring can be done with your conscious awareness and purposely undertaken. A person that you look up to has agreed to meet with you. Your admiration for the person has led you to wish that you could be like that person. During your interaction, you watch intently and try to mirror how they speak, what they say, etc. This is something you overtly are aiming to achieve. Sometimes this is referred to as a chameleon effect.
In a quick recap, there are two major modes of mirroring:
- (1) Physical mirroring. You imitate the physical actions of the other person such as their gestures, facial expressions, limb and body motions, and the like.
- (2) Speech and wording mirroring. You imitate how the other person speaks such as their tone and do so with the actual words they use too.
When you do mirroring, the intention can be:
- (i) Subconsciously undertaken. You are doing mirroring without necessarily realizing you are doing so, and it is seemingly a kind of spontaneous reaction.
- (ii) Consciously performed. You intend to do mirroring and are purposely trying to imitate the other person.
Now that we’ve covered the basics, we are ready for further unpacking.
Research On Human Mirroring
The field of psychology has carefully studied how humans mirror each other. Children when growing up frequently mirror those around them. It is considered an essential facet of learning about how to behave. Parents discover quickly that if they say a foul word or make an obscene gesture, the next thing they know their child is doing the same.
The youngster might not have any clue of the meaning. They could be simply doing a physical mirroring such as holding up a middle finger in the air. Worse still, if they are old enough to understand what is going on, they might be doing a speaking form of mirroring. Trying to get that to subside or be eradicated can be extremely challenging for parents.
An article entitled “The Power Of Mirroring” by Marty Nemko, Psychology Today, October 31, 2021, made these key points about human mirroring (excerpts):
- “We all like people who are like us.”
- “So, it’s no surprise that mirroring a conversation partner’s behavior can make them like us more and be open to our ideas.”
- “Indeed, we all want to persuade others, whether it’s to watch the movie we want, close a sale, or get that no-longer-special someone out of our lives.”
The gist is that part of the reason we do mirroring is to align with someone else and potentially create a stronger bond. This can serve too as a persuasion technique. If they see themselves in your actions and words, doing so can make them more open to being convinced of whatever points you are making.
AI Enters Into The Mirroring Realm
AI developers and AI researchers often lean into mirroring when devising AI-based systems such as robots.
The assumption is that if a robot imitates you, there is a heightened chance you will feel more comfortable interacting with the robot. In addition, when trying to train a robot to perform tasks, sometimes the easiest way to conduct the training is by having a robot imitate what you are doing. For example, rather than laboriously programming a robot to assemble a car door, it is feasible to have the robot observe a human performing the same task and then mimic or imitate similar moves.
A recent research paper highlighted the value of human-robot interaction and the role of mirroring. The article entitled “Review Paper: Body Movement Mirroring And Synchrony In Human-Robot Interaction” by Darja Stoeva, Andreas Kriegler, and Margrit Gelautz, ACM Transactions on Human-Robot Interaction, 2024, made these salient points (excerpts):
- “In human social interactions, people tend to mirror and synchronize to the body movement of their interaction partner as a way to coordinate and adapt to each other.”
- “With the growing availability of social robots, interaction dynamics is becoming a topic of increasing interest, making the social interaction between humans and robots a research topic of high relevance.”
- “Since the adaptation that occurs during human-human interaction is often bidirectional, it is important to consider which human behaviors can be adapted or transformed into robot behaviors if robots are to interact with humans in a social manner.”
You might have astutely noted that the points above refer to interaction as often being bidirectional.
Here’s why that’s important.
Your first instinct might be that a robot will always and only be the one doing the imitations. The thing is, people will at times start to imitate a robot (or, more broadly, an AI system). Similar to how human-to-human interaction can occur bidirectionally when it comes to mirroring, the same can happen between a human and a robot.
More generally, let’s go with the overarching notion that this bidirectional mirroring can occur between a human and an AI system, such as a robot, but can also arise when interacting with other types of AI such as generative AI.
On a slightly different but related topic, I previously examined the latest in AI-powered smart mirrors that can be used in your bathroom or hallway. These state-of-the-art smart mirrors include an Internet connection to a generative AI system and interact with you while you are in front of the mirror, see the link here. That’s some eye-popping real-world mirroring. Expect this to be commonplace in a few years.
Generative AI And Mirroring Considerations
I logged into ChatGPT to come up with examples of how mirroring can occur between humans and generative AI.
You might know that ChatGPT is the widely popular generative AI app by AI maker OpenAI. The number of people using ChatGPT is impressive and estimated at around 200 million weekly active users. The examples I show here using ChatGPT can readily be done in other generative AI apps such as Anthropic’s Claude, Google’s Gemini, Meta’s Llama, and many others.
The best way to start a conversation with generative AI is to first gauge whether the AI has had any data training on the topic of interest to you. Most generative AI is data trained via scanning portions of data throughout the Internet. Not everything in the known world will necessarily be encompassed via this data training. For more about the scope of data training for generative AI, see my discussion at the link here.
I asked a basic question to see what ChatGPT might have to say on the mirroring topic.
- My entered prompt: “Can generative AI mirror the behavior of a user?”
- ChatGPT generated response: “Yes, generative AI can mirror the behavior of a user to some extent, particularly in conversational settings. This can happen through various techniques and capabilities built into the AI’s design. Here’s how AI can achieve a form of “mirroring” behavior. Generative AI can adapt its responses to mirror the user’s style, tone, or formality based on patterns it detects. In real-time chats, AI can mirror the pacing and flow of conversation. If a user asks rapid-fire questions, the AI can provide quick responses. Through ongoing interaction, AI systems can learn user preferences and adapt behavior to reflect the user’s interests and communication style more closely, creating a sense of personalization. This process is akin to mirroring.”
The response by ChatGPT seems reasonable and suggests that some semblance of data training associated with mirroring likely occurred. To clarify, the odds are that mirroring as a topic was probably only tangentially scanned while data training. The mirroring topic is not likely any form of devoted attention. There seems to be enough that I would judge that it is worthwhile to continue the conversation with ChatGPT.
I bring up this content fluency consideration because people often fall into a mental trap of thinking that generative AI knows all. It is an easy trick to have happen to you. You use generative AI, and it answers your questions. All seems good. After a while, you are lulled into thinking the AI knows all.
Nope.
Always make sure to double-check whatever response you get from generative AI. A response can be wrong, confounding, and possibly contain a so-called AI hallucination (see my explanation of AI hallucinations at the link here).
Getting Generative AI To Mirror Me
The first example showcases ChatGPT mirroring me.
I’ve instructed ChatGPT to focus on whatever I enter as prompts and then try to mirror whatever I bring up. See if you can spot the mirroring by AI.
Here we go.
- My entered prompt: “I’m so tired today. It feels like I’ve got zero energy left.”
- ChatGPT generated response: “Yeah, I get that. Some days, it feels like all your energy is just drained away.”
- My entered prompt: “Exactly. And no matter how much coffee I drink, it’s not helping.”
- ChatGPT generated response: “Right, when coffee doesn’t even work, you know it’s a tough day.”
- My entered prompt: “I just want to crawl back into bed and sleep forever.”
- ChatGPT generated response: “Totally! Crawling back into bed and sleeping forever sounds like the dream right now.”
Did you see the mirroring?
I said that I was low on energy, and right away ChatGPT expressed the same consideration. Next, I brought up coffee. Sure enough, ChatGPT replied by mentioning coffee. My final remark was about wanting to go to sleep. ChatGPT said that sleeping would be dreamy.
If you are curious how long this kept going, I continued the dialogue for a few minutes and gradually became weary of the imitations. ChatGPT would endlessly do the mirroring until I told it to stop doing so. Of course, I could have given a different set of base instructions. For example, I could have indicated that the mirroring should be subtle and only occur from time to time.
Me Mirroring Generative AI
Two can play at this game. I decided to go ahead and mirror ChatGPT. Whatever ChatGPT brings up, I will try to mirror that same comment or connotation. This is intended to illustrate the bidirectional nature of mirroring.
This is how it went.
- ChatGPT generated question: “It’s wonderful to assist you today! How may I be of help?”
- My entered prompt: “Thank you! I truly appreciate your assistance. I’d love some help with learning how to write romance novels.”
- ChatGPT generated response: “I recommend starting with smaller steps when learning a new skill. Does that sound manageable?”
- My entered prompt: “That makes a lot of sense. Starting small is always a good approach. Can you begin with the basics?”
This was perhaps a bit more laid-back version of mirroring.
I was asked if I needed help. I could have just said that I wanted to learn how to write romance novels, but I decided to do a dose of mirroring by repeating back the comment of ChatGPT offering to be of help. The same occurred when ChatGPT indicated that small steps might be advisable. I could have simply said yes but decided to add that starting small is always a good approach.
If you are wondering what a non-mirroring approach would consist of, I did the same conversation, and this time tried to not mirror.
- ChatGPT generated question: “It’s wonderful to assist you today! How may I be of help?”
- My entered prompt: ” Teach me to write romance novels.”
- ChatGPT generated response: “I recommend starting with smaller steps when learning a new skill. Does that sound manageable?”
- My entered prompt: “Get underway.”
Notice that I merely provided my responses and didn’t attempt to mirror what ChatGPT was saying.
Mirroring Is Different From Parroting
You might be wondering if mirroring is the mindless act of repeating precisely what the other person said, or in this case what the AI had to say. I don’t think that’s a fair or reasonable form of mirroring. It can happen, let’s make that clear.
To me, I think a word-for-word echo is nothing more than a raw form of parroting and not what seems to be in the spirit of mirroring.
Here’s a parroting example (me trying to parrot the AI).
- ChatGPT generated question: “How can I assist you today?”
- My entered prompt: “How can I assist you today?”
- ChatGPT generated response: “I’m here to help with any questions you have.”
- My entered prompt: “I’m here to help with any questions you have.”
- ChatGPT generated response: “Great! What would you like to know more about?”
- My entered prompt: “Great! What would you like to know more about?”
That’s not going to accomplish much. If you’ve ever had a parroting discussion with a child, you know it becomes quickly exasperating. Kids sometimes get a chuckle out of it. Maybe that gives the parroting some redeeming value.
Introspection Arising From Mirroring
There is a potential bonus that comes from having generative AI mirror you.
After I had undertaken several conversations with ChatGPT, during which ChatGPT was mirroring me, I decided to ask ChatGPT to analyze my style or approach based on the mirroring that had been taking place.
Here’s what I got.
- My entered prompt: “You have been mirroring for numerous lengthy conversations. Summarize what I am like based on the mirroring that you’ve done of me.”
- ChatGPT generated response: “Based on our mirrored conversations, you come across as a thoughtful and reflective person who enjoys exploring ideas in a detailed, step-by-step manner. Your curiosity drives you to ask insightful questions, often breaking down complex topics into manageable parts, which reveals a strong analytical mindset. You’re open-minded and engaged, eager to consider different perspectives and reflect on your own behavior and that of others. Your attention to detail suggests that you value thoroughness and clarity, both in conversations and in your approach to challenges. There’s also a playful yet purposeful side to your explorations, as you enjoy hypothetical scenarios but always with a clear intention behind them, whether it’s to deepen understanding or just creatively engage with new concepts.”
Notice that ChatGPT was able to ascertain various elements of my writing style, along with making conjecture about my way of thinking.
Some sternly object to this kind of analysis by generative AI and decry that it is a bridge too far. The concern is that generative AI is acting as though it is a kind of mental therapist that can figure out your mind. I’ve covered extensively those concerns at the link here.
Mirroring Works For Humans And AI
You likely knew already that humans tend to mirror other humans. That’s nearly old hat.
Did you anticipate that AI can also do mirroring?
That might have been a twist on mirroring that you hadn’t previously considered. Now that the topic is out in the open, I’m sure you can readily see the logic of it.
With generative AI, you can activate mirroring if you want the AI to mirror you. That’s one path. You can also opt to mirror generative AI.
Why would you mirror generative AI?
Well, I’m glad you asked since it takes me to a notable mind-bender to conclude this discussion.
Some believe that with generative AI gradually and inevitably becoming ubiquitous, we might all as a society start to talk and interact like generative AI does. In other words, we will get so accustomed to using generative AI that we will either subconsciously or consciously start to speak and use words as does generative AI.
Scary or a good thing?
Go look in a conventional non-smart mirror and see if you like the idea that you will someday speak and act like generative AI. Exciting or disturbing? I trust that your decision about whether this is a good or bad outcome will come from that highly reflective and personal exercise.
