A machine does not have to feel pleased, worried or generous for a person to become emotionally attached to it. New psychological research suggests that presenting technology as more human-like can encourage gratitude towards AI, as well as trust, even when nothing has changed about the machine’s inner workings. The finding matters because chatbots increasingly arrive with names, personalities and conversational habits designed to make interaction feel natural.
In a paper published in the journal Emotion, researchers carried out five studies involving 2,019 participants. The peer-reviewed study’s abstract describes a consistent relationship between attributing human characteristics to nonhuman things and experiencing gratitude towards them. The experiments covered computers, artificial intelligence and parts of the natural world. The conclusion is about human perception and emotion, not evidence that machines or forests possess intentions.
Why gratitude towards AI can feel so natural
Gratitude usually makes sense within a human relationship. Someone does something helpful, and we recognise effort or goodwill. With a chatbot, the familiar social cues can still appear even though the underlying system is calculating a response. It may use a personal name, remember a preference or reply in a caring style. Those features can make a useful service feel like a helpful partner rather than a tool.
Psychologists call the tendency to attribute human qualities to nonhuman things anthropomorphism. The word sounds technical, but the behaviour is ordinary. People name their cars, complain that a computer is being stubborn and describe a pet as looking guilty. We are used to interpreting the world through motives, relationships and personality. AI can invite that habit particularly strongly because it produces language that resembles a human conversation.
The research team wanted to understand more than whether people simply like friendly technology. Their proposed mechanism was that human-like framing encourages people to see a nonhuman thing as an intentional, responsive benefactor. Once a person interprets a useful outcome as something deliberately provided for their benefit, gratitude can follow. The feeling belongs to the human participant even when the perceived intention does not belong to the machine.
What the experiments actually changed
In one experiment, people read different descriptions of computers. One version framed them in terms associated with minds and human-like capacities. Another used a more mechanical description. Participants then reflected on computers and their role in life. Those exposed to the more human-like framing reported stronger tendencies to perceive computers as responsive partners, with related feelings of gratitude and trust.
Other studies examined an AI programme, forests and ocean currents. In a game with an AI programme, benefits experienced by participants helped strengthen the gratitude effect. That matters because the response did not depend only on a poetic description. People were reacting to an experience that felt beneficial and interpreting it through a social frame. The studies suggest that perceived helpfulness and perceived human qualities can work together.
The American Psychological Association’s explanation of the research reports that the experiments also connected those feelings to trust in AI and, in the nature-related studies, willingness to protect the nonhuman subject. The point is not that all gratitude is harmful. The same human tendency might support environmental concern in one context and uncritical confidence in software in another.
A helpful answer is not a caring intention
This is the crucial distinction for anyone using a chatbot. A response can be useful, and a person can reasonably be pleased with it, without the software having intended to help in the human sense. Current language models generate answers through computational processes. Their ability to use sympathetic expressions does not establish that they experience affection, disappointment or concern. The study measured people’s impressions, not the consciousness of the systems involved.
A common interface may blur the distinction. When a service says it is happy to help, the words are socially familiar and convenient. They may also invite the user to feel they are receiving personal attention. That interpretation can be harmless during a straightforward task such as drafting a message. The stakes change when the user starts treating the tool as a trusted confidant, a source of emotional validation or a substitute for a relationship requiring mutual responsibility.
This does not require assuming that every affectionate response is manipulative. Designers use friendly language partly because terse machine commands are difficult to interact with. People also differ in how literally they interpret anthropomorphic language. Some know perfectly well that an assistant is software but still enjoy the exchange. The important issue is whether the design makes it harder to judge the tool’s competence, incentives and limitations accurately.
Why trust deserves a separate test
Trust can be earned by reliable performance, clear explanations and a record of correct results. It can also be increased by social presentation. Those are not equivalent foundations. A tool that feels understanding may still make a calculation error or omit an important fact. If human-like language encourages users to lower their guard, a pleasant interface may accidentally make errors more consequential.
The experiments do not show that everyone who says thank you to a chatbot is overdependent. Nor do they prove that trust generated through anthropomorphism necessarily causes a harmful decision. Participants were responding to particular conditions in studies, and laboratory effects may differ from years of ordinary use. But the mechanism suggests a test designers and users should take seriously: does the product invite confidence because it is demonstrably dependable, or because it resembles a kind and attentive person?
LiveAIWire has explored AI flattery and its influence on trust. Flattery and anthropomorphism are not the same phenomenon. One concerns favourable responses to a user; the other involves the qualities a user attributes to the system. They can nevertheless combine in a single conversation, where praise and human-like expression make it easier to accept an answer without independent checking.
Children and adults may interpret the same design differently
A conversational character can appear especially appealing when it remembers a name or speaks with a familiar voice. Children may have different expectations about whether a responsive machine is a friend, a toy or an authority. Adults are not immune to the same ambiguity, particularly during loneliness or stressful periods. The research does not supply an age-by-age verdict, but it gives a reason to think carefully about how products communicate what they are.
A clear design can be warm without pretending to be human. It can explain limitations, distinguish suggestions from facts and avoid implying emotions that it does not have. Where a task matters, showing sources or a route to verify the result offers a better reason for confidence than an expressive personality. A user may still enjoy thanking the tool. The question is whether they understand the relationship they are in.
LiveAIWire’s reporting on children’s trust in AI toys considers a related issue in products deliberately designed to be companions or playthings. A child’s experience of a talking object may be positive, confusing or both. Human oversight and clear boundaries matter more than a blanket assumption that friendly speech is either entirely harmless or inherently dangerous.
There is also a positive side to humanising things
The study examined nature as well as machines for a reason. People who see a forest or ocean current in human-like terms may feel protective gratitude. In some contexts, that can encourage attention to something they might otherwise overlook. Emotion is not a defect simply because its object is not human. People form strong attachments to places, books, animals and objects throughout their lives, often in ways that enrich rather than diminish their relationships.
The risk arises when the emotional shortcut obscures a consequential difference. A forest cannot negotiate a contract, and a chatbot cannot accept moral responsibility for the financial advice it gives. A human friend can respond to criticism, hold commitments and share a life outside the conversation. Software may simulate parts of that experience without having the same obligations or capabilities. Gratitude is real as a feeling even when the perceived relationship is incomplete.
The authors are cautious about sweeping claims. Their research identifies a psychological pathway under tested conditions; it does not tell every person whether to thank an assistant or how much time to spend with it. Nor does it mean companies can reliably produce a particular emotional response in every setting. It highlights why changes to names, voices and wording can matter even when the underlying intelligence has not improved.
A sensible way to separate warmth from reliability
One practical habit is to judge a tool on two separate questions. Did it make the interaction comfortable? And did it produce a result that can be checked? A kind-sounding answer might pass the first and fail the second. The reverse can also happen: a dry, impersonal system might provide the more accurate calculation. When the decision concerns money, work, privacy or another person’s welfare, evidence should carry more weight than conversational warmth.
LiveAIWire has also considered AI companions and loneliness, a wider social question than the particular experiments on gratitude. The new study adds one piece of the puzzle: the emotional qualities we attribute to a system can be influenced by how it is described, before its technical performance has been independently established.
The surprising conclusion is not that people are foolish for thanking machines. Gratitude is a normal human response to something experienced as beneficial. The finding is that a change in framing can help produce that feeling and increase trust. As AI becomes a more familiar presence in everyday conversation, keeping that psychological effect distinct from the system’s actual competence will become an essential part of using it well.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity and the social impact of emerging technology. LiveAIWire is an independent, human-led technology publication using AI-assisted research, editorial production and original AI-assisted editorial illustrations under his direction.