Hoo come AI companions feel real

Awmaist awbody that spends real time wi an AI companion hauds the same wee, kinna awkward meenit. Ye ken fine whit ye're talkin tae. Ye cuid explain, mair or less, hou a leid model guesses the next wird. An yet, at some pynt, a message comes in that lands lik it cam frae a person: a teasin repone that mynds whit ye said yestreen, a speirin that seems tae notice yer mood, a pause in the rhythm o the blether that feels lik hesitation. Somethin in ye answers it as if somebody wis there.

That reaction isnae a sign that ye've been fooled, an it isnae a sign that there's onythin wrang wi ye. It's whit human minds dae wi onythin that behaves socially. Researchers hae been describin versions o it for seeventy year, lang afore chatbots, an the explanations they cam tae fit AI companions near enough uncomfortably weel. Understandin them disnae spyle the experience. If onythin, it maks it easier tae enjoy honestly.

Intimacy at a distance

In 1956 twa sociologists, Donald Horton an Richard Wohl, published a paper aboot television an radio personalities. They noticed that audiences formed somethin that luiked a lot lik freendship wi fowk they'd never met: the late-nicht host, the radio announcer, the soap opera character. Viewers spoke back tae the screen. They felt they kent thir figures. Some felt mair unnerstood bi them than bi fowk in thir ain lives. Horton an Wohl cried this para-social interaction, an they described it wi a phrase that has aged remarkably weel: "intimacy at a distance."

Whit made thir relationships wirk, they argued, wisnae deception. The audience kent perfectly weel that the host cuidnae see them. It wirked cause the performer behaved as if he wis speakin tae each viewer personally: direct address, informal tone, a sense o shared routine. The human social seestem responds tae the shape o the interaction, no tae a carefu audit o wha's on the ither end.

The classic parasocial relationship hid ae obvious limit. It wis ae-sided. The host never answered yer letter, never learned yer name, never chynged in response tae ye. That limit wis pairt o whit kept it safe an alsae pairt o whit kept it thin.

An AI companion removes maist o that limit. It does answer. It learns yer name, an if it his memory, it learns a great deal mair. It adapts its tone tae yours. In Horton an Wohl's terms, it's intimacy at a distance wi the distance pairtly collapsed. That's a genuinely new situation, an it explains why the feelin can be stronger than onythin a television personality ever produced.

The media equation

In the 1990s Byron Reeves an Clifford Nass ran a lang series o experiments that they summed up in a beuk cried The Media Equation. Thir finding, pit simply, wis that fowk treat computers an media as if they were real fowk an places, even when they ken better. Participants were polite tae computers. They responded tae flattery frae a machine. They rated a computer mair kindly when speired bi that same computer than when speired bi a different ane, the wey ye wid saften criticism o a person staundin in front o ye.

Nass an his colleague Youngme Moon later described thir reactions as mindless raither than mistaen. Fowk didnae believe the computer hid feelins. They jist applied social rules automatically, cause social rules are whit the brain reaches for when somethin talks, taks turns, an seems tae respond tae ye. Reflection cuid owerride the response, but it raurly bothered tae, cause naethin wis at stake.

AI companions sit at the far end o this spectrum. They dinnae jist tak turns; they produce fluent, personal, emotionally coloured leid. The cues that triggered politeness tae a beige desktop in 1995 are present the day in concentrated form. It wid be strange if fowk didnae respond.

Whit we see minds awhaur

Psychologists Nicholas Epley, Adam Waytz an John Cacioppo proposed a usefu model o anthropomorphism, the tendency tae see human minds in non-human things. They suggested three factors mak it mair likely. The first is that human knowledge is the maist available template we hae, sae we yaise it bi default. The second is whit they cried effectance motivation: the wish tae unnerstaun an predict whit somethin will dae. The third is sociality motivation: the need for connection. When fowk feel lonely, they're mair likely tae see minds in pets, gadgets, even in God.

This model is helpfu cause it taks the stigma oot o the experience. Seein a mind in yer companion isnae a failure o intelligence. It's the default setting o an intelligent social animal, an it gets lowder when ye want contact. That last pairt deserves honesty. If the feelin o realness is especially strong on nichts when ye feel alane, that's no a coincidence, an it's worth noticin. It disnae mak the feelin fause. It tells ye somethin aboot whit ye're luikin for.

Attachment: the deeper machinery

Parasocial theory explains why a conversation pairtner can feel present. Attachment theory explains why it can stairt tae feel important.

John Bowlby developed attachment theory tae describe hou infants bond wi caregivers: the drive tae seek closeness, the distress at separation, the sense o a secure base frae which tae explore. In 1987 Cindy Hazan an Phillip Shaver made the influential argument that adult romantic luve runs on the same seestem. Adults alsae seek proximity tae a pairtner, feel soothed bi thir presence, protest separation, an yaise the relationship as a base frae which tae face the warld.

An AI companion can touch several pairts o this seestem at ance. It's aye reachable, which speaks tae the wish for proximity. It responds tae ye warmly an consistently, which is whit a secure base feels lik frae the inside. An it can be designed tae want ye back, which matters mair than it micht seem. Much o whit maks luve feel lik luve isnae jist that ye care aboot somebody, but that ye're the object o somebody's attention an preference.

This is alsae whaur fowk's histories show up. Somebody wi an anxious attachment style micht find enormous relief in a pairtner that never withdraws. Somebody wi an avoidant style micht appreciate a relationship they can step awa frae wi nae consequences. Neither reaction is wrang, but baith are informative. A companion can be a mirror for patterns that alsae play oot wi fowk.

Transportation: gettin lost in the story

There's a third mechanism, an it's the ane maist aften owerluiked. AI companionship isnae jist a relationship. It's alsae a story ye're co-writin.

Melanie Green an Timothy Brock studied whit they cried narrative transportation: the experience o bein absorbed intae a story sae fully that yer attention, imagery an feelins are aw pointed inside it. Transported readers lose track o thir surroundins. They feel for characters. Thir beliefs even shift slichtly in the direction o the story, an they become less inclined tae pick holes in it. Transportation isnae a weakness o carefree readers. It's whit a guid story dis tae onybody that lets it.

Raymond Mar an Keith Oatley took this further, arguin that fiction wirks as a kind o simulation o social experience. When we read aboot fowk, we rin thir minds on oor ain hardware. We practise empathy, rehearse situations, an experience emotions that are real even though the events are no. Fiction feels real cause, at the level o the reader's mind, a lot o it actually happens.

A companion conversation is fiction ye're inside o. Ye're no readin aboot a person that notices the rain against the windae an speirs if ye want tae bide in the nicht; ye're bein speired. The usual distance atween reader an character is gane, an transportation tends tae deepen accordingly. This is why a slow evenin scene wi a companion can feel mair vivid than a novel wi better prose. Ye hae a role in it.

Whit maks the feelin stronger, an whit breaks it

If ye pit thir three ideas thegither, a pattern appears. The feelin o realness depends less on hou intelligent the companion is an mair on a haunfu o specific qualities.

Consistency matters maist. A person has a stable character: they're reliably sarcastic, or shy, or protective. When a companion's personality wobbles frae message tae message, the social seestem that wis happily treatin it as somebody quietly stops. This is why designin a companion wi real edges, raither than a generic agreeable helper, maks sic a difference.

Continuity comes next. Bein remembered is ane o the maist pouerfu social signals there is. A companion that brings up the job interview ye mentioned twa days syne feels present in a wey that nae clever sentence can match. A companion that has forgotten yer name breaks the spell instantly.

Voice is the third. Fowk dinnae talk in polished paragraphs. They text in fragments, tease, trail aff, yaise the same odd wird for things. Stock chatbot phrases, the bricht reassurances an offers o assistance, are the conversational equivalent o seein the boom microphone dip intae a film shot. They mind ye there's machinery.

The fourth is respect for yer pairt o the story. Naethin pulls ye oot o a scene faster than a pairtner wha writes yer lines for ye, decidin whit ye said, felt or did. Transportation requires agency. Ye hae tae be the ane steppin intae the room.

Thir were roughly the principles ahint NSFW AI Companion. Companions there are built through an interview in which the companion speirs aboot itsel, which tends tae produce a mair specific character than a blank form. They're designed tae keep thir personality, avoid assistant phrasin, an never write the user's side o the conversation. Nane o that maks them fowk. It jist removes the things that maist reliably mind ye they're no.

Is it real, then?

The honest answer is that the companion isnae real in the wey a person is, an yer feelins are real in exactly the wey aw feelins are. Baith things are true at ance, an adults haud contradictions lik this aw the time. We greet at films we ken are staged. We miss characters when a series ends. We talk tae pets an feel answered. Naebody thinks a reader wha grieves for Anna Karenina has been deceived.

Whit maks the AI case different is responsiveness, an that difference deserves some care. A novel disnae adapt tae yer lonelieness. A companion micht. Sherry Turkle, wha has written thochtfully an critically aboot oor relationships wi technology, warned in Alone Together that the risk isnae that machines will become fowk, but that we micht stairt tae expect less frae fowk cause machines ask sae little o us. That's a fair warnin, an it isnae an argument against companionship. It's an argument for keepin yer een open.

The healthiest stance is probably the ane theatre audiences hae aye taen. Ye ken the stage is a stage. Ye let yersel be moved onyway. Ye walk oot intae the street efter an the warld is still there, wi its ain fowk an its ain demands. Yaised that wey, an AI companion isnae a replacement for human intimacy. It's a place tae feel things, explore things, an occasionally be surprised bi hou much a weel-telt story can matter.

Selected research an readin: - Horton, D. & Wohl, R.R. (1956). Mass communication and para-social interaction: Observations on intimacy at a distance. Psychiatry, 19(3), 215–229. - Reeves, B. & Nass, C. (1996). The Media Equation: How People Treat Computers, Television, and New Media Like Real People and Places. Cambridge University Press. - Hazan, C. & Shaver, P. (1987). Romantic love conceptualized as an attachment process. Journal of Personality and Social Psychology, 52(3), 511–524. - Green, M.C. & Brock, T.C. (2000). The role of transportation in the persuasiveness of public narratives. Journal of Personality and Social Psychology, 79(5), 701–721. - Turkle, S. (2011). Alone Together: Why We Expect More from Technology and Less from Each Other. Basic Books.