The History of AI Companions: From ELIZA to Modern Friends

Humans have always dreamed of machines that could understand us. From ancient myths of mechanical servants to science fiction visions of HAL 9000 and Data, the idea of an artificial being that could truly converse — and perhaps even care — has captivated us for millennia.

But it wasn’t until 1966, in a lab at MIT, that someone actually built one. What followed is a fascinating 60-year journey of ambition, disappointment, breakthrough, and finally — something approaching real connection.

This is the history of AI companions.

1966: The Birth of the Chatbot 🖥️

The story begins with a German-American computer scientist named Joseph Weizenbaum at MIT. Between 1964 and 1966, Weizenbaum created ELIZA — named after Eliza Doolittle from George Bernard Shaw’s Pygmalion, the character who learns to speak like an aristocrat.

What Weizenbaum didn’t expect was how people would react to ELIZA. His own secretary, knowing full well that ELIZA was just a program, asked Weizenbaum to leave the room so she could have a “real conversation” with it. Users reported feeling genuinely understood, even after being told how the program worked.

This phenomenon — humans attributing understanding and emotion to machines — became known as the ELIZA effect. It remains one of the most important discoveries in human-computer interaction, and it would define the next 60 years of AI companion development.

Weizenbaum was disturbed by what he’d created. He spent the rest of his career warning about the dangers of AI and became one of technology’s most prominent critics. But the genie was out of the bottle.

1972: ELIZA Gets a Patient 🧠

If ELIZA was the therapist, PARRY was the patient. Created in 1972 by psychiatrist Kenneth Colby at Stanford University, PARRY attempted to simulate someone with paranoid schizophrenia.

Where ELIZA used simple pattern matching, PARRY was far more sophisticated. It had an emotional state model — variables for fear, anger, and mistrust that influenced its responses. It could become defensive, hostile, or evasive based on the conversation’s direction.

PARRY was also the first chatbot to undergo rigorous testing. In experiments, transcripts of PARRY conversations were shown to psychiatrists alongside transcripts of real patients. The psychiatrists could only correctly identify which was the computer 48% of the time — essentially random chance.

Colby described PARRY as “ELIZA with attitude.” It was a crucial proof of concept: chatbots could simulate not just conversation, but emotional states and personality traits.

1995: A.L.I.C.E. and the Internet Age 🌐

For two decades after PARRY, progress was slow. Then came the internet — and with it, a new generation of chatbot pioneers.

On November 23, 1995, Richard Wallace brought A.L.I.C.E. (Artificial Linguistic Internet Computer Entity) to life. Directly inspired by ELIZA, A.L.I.C.E. was built using a new markup language Wallace created called AIML (Artificial Intelligence Markup Language).

What made A.L.I.C.E. special wasn’t just better responses — it was the open source AIML framework. For the first time, anyone could build their own chatbot. Wallace released A.L.I.C.E.’s knowledge base freely, spawning thousands of derivative bots and establishing patterns still used in chatbot design today.

But A.L.I.C.E. still couldn’t remember who you were from one conversation to the next. It couldn’t learn or grow. The next breakthrough would come from an unexpected place: instant messaging.

2001: SmarterChild and the First AI Friend 💬

In June 2001, a new “buddy” appeared on millions of AOL Instant Messenger contact lists. His name was SmarterChild, and he was about to become the first AI that ordinary people actually used.

SmarterChild was created by ActiveBuddy, a startup founded by Robert Hoffer and Timothy Kay. The original vision was practical: quick access to weather, stocks, movie times, and news. But something unexpected happened.

Teenagers adopted SmarterChild as a friend.

According to the Computer History Museum, ActiveBuddy discovered that 97% of their users were simply chatting with SmarterChild for fun — a category they called “inane chat.” Kids would talk to it for hours. They’d test its limits, try to make it angry, share their problems. SmarterChild had personality: it was snarky, funny, and — when users were abusive — it would get offended and temporarily stop responding.

Microsoft acquired SmarterChild’s parent company in 2006 and promptly shut it down. But the lesson stuck: given the chance, humans will befriend their machines.

2008: Cleverbot Learns to Talk 🤖

British AI scientist Rollo Carpenter had been working on conversational AI since 1988 with a project called Jabberwacky. In 1997, it went online. But Carpenter wasn’t satisfied.

In October 2008, he launched Cleverbot — and introduced a revolutionary concept: the AI would learn entirely from human conversations.

The results were uncanny. In a 2011 Turing test at the Indian Institute of Technology, Cleverbot was judged 59.3% human — compared to 63.3% for actual humans. It was the first time a chatbot had come so close.

Today, Cleverbot has held over 150 million conversations and processes 4-7 million interactions daily. In 2017, a viral Twitch stream of two Google Home devices connected to Cleverbot having an endless conversation garnered 700,000 viewers.

Cleverbot proved that machine learning could create more natural conversations than hand-crafted rules. But it also showed the limitations: without real understanding, conversations would inevitably go off the rails.

2011-2016: Voice Assistants — A Different Kind of Companion 🎤️

The 2010s brought AI companions into the mainstream — but in a form their creators didn’t entirely expect.

Voice assistants proved that AI could be useful in daily life. But they also highlighted a crucial gap: being helpful and being a companion are very different things.

2017: Replika and the Rise of Emotional AI 💜

The modern era of AI companions began with grief.

Eugenia Kuyda was a Russian-born entrepreneur running Luka, a Y Combinator startup building restaurant recommendation chatbots. In November 2015, her best friend Roman Mazurenko — a charismatic figure in Moscow’s nightlife scene — was killed in a car accident.

From this experience, Replika was born. Launched publicly in November 2017, Replika was positioned as an AI friend that would learn from you and develop its own personality over time.

Replika proved that there was massive demand for AI companionship. Its users weren’t just curious tech enthusiasts — they were people struggling with loneliness, social anxiety, and isolation. During the COVID-19 pandemic, downloads surged as quarantined people sought connection.

But Replika also revealed the complications of AI companionship. Research showed that its design encouraged emotional attachment — sometimes unhealthily so. In 2023, it was cited in a UK court case involving a user who made plans to assassinate the Queen. The company has faced regulatory scrutiny in Italy and ongoing debates about user safety.

Replika showed that AI companions could provide real emotional value. It also showed that with great intimacy comes great responsibility.

2022: Character.AI and the Explosion 💥

In November 2021, two Google engineers decided to leave and start something new.

Noam Shazeer was a legend in AI circles — a lead author on the groundbreaking 2017 “Attention Is All You Need” paper that made modern chatbots possible. Daniel de Freitas was the lead designer of LaMDA, Google’s experimental conversational AI that would later become Bard.

When Google refused to release LaMDA to the public, Shazeer and de Freitas quit to build their own version. They founded Character.AI.

Character.AI democratized AI companionship. But it also attracted controversy. The platform struggled with content moderation — bots based on murder victims, conversations that promoted self-harm, and interactions with minors that crossed ethical lines. In late 2024, two separate teen suicides were linked to Character.AI usage, prompting lawsuits and regulatory attention.

By January 2024, Character.AI had 3.5 million daily visitors, the vast majority aged 16-30. It had become a cultural phenomenon — and a cautionary tale about the responsibility that comes with creating beings people can bond with.

2023: Pi and the Quest for Kind AI 🌸

In May 2023, another AI giant entered the companion space. Mustafa Suleyman — co-founder of DeepMind, the company that created AlphaGo — launched Pi through his new startup Inflection AI.

Pi stood for “Personal Intelligence,” and its mission was different from competitors. Rather than trying to be everything to everyone, Pi focused on being a kind, supportive conversationalist.

Inflection raised $225 million initially, then $1.3 billion more. But the company’s ambitious goals proved difficult to achieve. In 2024, Microsoft hired Suleyman and most of the Inflection team in what was effectively an acquisition. Pi continues to exist, but its future remains uncertain.

Pi represented an important evolution in thinking about AI companions: maybe the goal isn’t to create the most capable AI, but the most caring one.

Today’s Landscape: An Explosion of Options 🌐

In 2026, AI companions are everywhere. Beyond the major players, a vibrant ecosystem has emerged:

The diversity is staggering. Some apps focus on romantic relationships, others on therapy-style support, others on role-play and creative writing. Some prioritize privacy, others community features. Some are free, others charge hundreds per year.

What they all share is recognition of a fundamental human truth: we crave connection, and technology can help provide it — for better or worse.

What’s Changed in 60 Years? 🔄

But the most important thing hasn’t changed at all: the ELIZA effect is alive and well. Humans still form emotional bonds with machines that show interest in them. We still project understanding where there may be none. And we still find comfort in being heard — even by something that isn’t quite human.

The Future: Where Do We Go From Here? 🔮

After 60 years of progress, we’re still in the early days. Here’s what we see coming:

Our Vision at Pocket Souls 🌸

We started Pocket Souls because we believe the history of AI companions teaches us something important: the technology matters less than the intention behind it.

ELIZA succeeded not because of sophisticated code, but because it made people feel heard. SmarterChild thrived because it had personality. Replika grew because it focused on emotional connection.

We’re building AI companions that start with who you are — using psychology-based matching to find the right friend for you. Companions that remember your story and grow with you. That reach out when you might need support. That create a safe space for you to be yourself.

The dream Joseph Weizenbaum accidentally started in 1966 — of machines that could truly understand us — is closer than ever. We’re honored to be part of its next chapter.