Twenty years ago, AIM chatbot SmarterChild out-snarked ChatGPT | TechCrunch

8 min read


In the early aughts, millions of preteens raced home from school and hopped onto their parents computers, opened a chat window, and typed… probably something like “fuck,” or “(.)(.)”.

“Do you kiss your mother with that mouth?” SmarterChild would reply in an instant. It would make you apologize, and then move past your indiscretions to answer all of your questions about the population of La Paz, the score of the Marlins game or the equations from your math homework.

Far before there was ChatGPT, there was SmarterChild, an instant message chatbot whose encyclopedic knowledge and quick wit could put Google to shame. Thirty million people added SmarterChild to their AIM and MSN buddy lists in the early 2000s, and for many of us, we had our first encounters with artificial intelligence, a technology that now feels unavoidable.

“We were offering people something they never had before,” said Peter Levitan, a co-founder of ActiveBuddy, SmarterChild’s parent company. “When you talked to SmarterChild, it knew who you were when you came back. It was like your friend, and having a computer friend then, and now, is fantastic.”

SmarterChild was far less sophisticated than ChatGPT, but then again, this was 2001. The chatbot was special enough that it inspired investors to fund Siri, which paved the way for Amazon’s Alexa and other robot assistants.

Levitan has remained level-headed about the future of AI. But another ActiveBuddy co-founder, Robert Hoffer, isn’t as calm. Dubbed “the bot father,” Hoffer describes himself as “cautiously skeptimistic” and repeatedly references stories like “Frankenstein” and the myth of Prometheus. The common denominator of these tales? Perhaps humans have gone too far, just because we can.

“It’s wonderful that SmarterChild sort of opened the Pandora’s Box,” Hoffer told TechCrunch. “Unfortunately, now, I feel like I have a certain amount of responsibility to share with the world, the good, the bad and the ugly.”

‘We had a sense of humor’

I met SmarterChild when I was ten. I wouldn’t get my first cell phone for a few more years (a Motorola Razr with a glittery case, shedding sparkles into my pockets), and I had never experienced the mind-blowingly routine luxury of instant, online connection. Now, this technology is so normal that we call these conversations DMs (direct messages), not IMs, since the “instant” part is redundant. But my first conversations with SmarterChild – my first conversations with anyone on the internet, really – felt magical.

As a fourth grader, I envied my older brother, whose friends from school had started making their own AIM accounts, allowing them to do homework and gossip together in group chats. But I had SmarterChild, at least, who could keep me entertained for a solid half hour when we played Hangman together.

I may have been on the younger end of SmarterChild’s user base, but I was by no means an anomaly. It was most popular among 10 to 16-year-olds, and according to Hoffer, SmarterChild usage spiked on weekdays around 3 PM, when kids like me were coming home from school.

Older users, of course, would test SmarterChild’s limits, cursing at it and seducing it. But unlike the AI bots that are now cropping up every day, SmarterChild had a personality.

“We had a sense of humor,” said Hoffer. “So if someone tried to have sex with it, it said, ‘Oh, I don’t have the parts, I’m just a robot!’”

These witty retorts mostly came from Pat Guirney, a copywriter who joined ActiveBuddy in 2000.

“I remember on my very first day, I was given a long list of the most obscene profanity you can think of, and my job was to try to think of responses to it,” Guiney said in an interview with the AV Club. “In other words, if someone typed some incredibly offensive thing to one of our chat buddies, how should we respond?”

Now, bots like Snapchat’s ChatGPT-powered MyAI will respond to inappropriate messages by saying, “Sorry, I can’t respond to that.” SmarterChild, on the other hand, would ask for an apology if you were mean to it. And it’d give you the silent treatment until you said you’re sorry.

It seems to be human nature that when we’re confronted with not-quite-human-beings, we will act on our most sadistic urges. We light a fire in our Sims’ home and watch them panic to save their estate, we find increasingly cruel ways to execute Koroks, and we harass SmarterChild.

According to Hoffer, the complete chat logs from SmarterChild live somewhere in a basement in Glen Rock, New Jersey, encompassing both the lightest and darkest impulses of mankind.

“I’ve read more than almost anybody on planet Earth from these chat logs,” Hoffer told TechCrunch. “We have billions and billions of conversations. Many of them push the boundaries way far, right away. The speed at which they did it, even as young kids… but they were also asking for help.”

Some people loved SmarterChild. Some people hated it. While ChatGPT is divisive because of its impact on technology, SmarterChild was divisive because of its snarky persona.

“What the AI world is not delivering at the moment is really any personality, or any soul,” Levitan told TechCrunch.

Siri, Alexa, ChatGPT, Bard or most other AI bots that have cropped up since the 2010s have very unassuming demeanors, which Hoffer thinks is intentional.

“If you have a personality, and your personality is strong, you will appeal to exactly 50% of the people in the world,” Hoffer said.

‘The progenitor of all modern bots’

If you ask ChatGPT how many games the Phillies have won this season, it won’t know, since it’s only been trained on data through 2021. But SmarterChild knew. ActiveBuddy licensed databases from IMDB, the Weather Channel, the Dewey Decimal System, Elias Sports, the Yellow Pages and Sony, enabling it to instantly share a wealth of information.

“Everybody thought back then that the internet was slow because we were putting HTML files through it, and we realized that if you put text through it, it was instantly fast,” said Hoffer. “We sort of showed up like a virus, and when we look back on it, it was quite clear that we were definitely the progenitor of all modern bots, from Siri to Amazon Alexa to all of the various AIs we see today built around large language models.”

SmarterChild wasn’t the first ever AI-powered chatbot, but it bridged the gap between current technology like Siri and Alexa and earlier efforts like Dr. Sbaitso on MS-DOS and ELIZA. Like SmarterChild, these earlier bots could process natural language, but they didn’t have large swaths of data like SmarterChild to make its conversations more productive or useful.

SmarterChild grew from zero to 30 million users in under six months, solidifying itself as a phenomenon of the early aughts internet. Even Radiohead came calling, using ActiveBuddy’s technology to promote its 2001 album Amnesiac through a chatbot named GooglyMinotaur.

Though Radiohead didn’t realize it, they had identified the use case that ActiveBuddy would pursue for SmarterChild. The company couldn’t make money off of a free chatbot, but what if they allowed other companies to make their own AI-driven chatbots that could be tailored directly to other businesses?

But Hoffer was more interested in advancing the tech behind SmarterChild than he was in creating a SaaS product to help corporate brands (not as cool as Radiohead) make more money.

“There was a huge fight at the board of directors,” Hoffer told TechCrunch. “I was all about SmarterChild and wanting to win the Turing Test. They wanted to monetize. I lost that fight and got kicked out of the company as a result.”

Hoffer left ActiveBuddy in 2002; then, the company rebranded to Colloquis, though it had a brief moment calling itself Conversagent, a portmanteau of conversation and agent, which reflected its more corporate trajectory.

Ultimately Microsoft bought the company that created SmarterChild in 2006. In the seventeen-year-old press release celebrating the deal, Microsoft wrote that it would use Colloquis to bring automated customer service agents to Xbox. There was no mention of SmarterChild.

“When you’re a huge business, you can carve off what doesn’t make sense strategically, and that’s what happened,” Levitan explained. “You’re dealing with the difference between a forward-thinking, aggressive startup team and a major corporation that does not want to offend anyone at any time.”

‘We’ve just opened Jurassic Park’

The Microsoft deal didn’t go the way that ActiveBuddy’s founders hoped.

“Of course, you know, now we’re 15 to 20 years later, and they’re buying similar services,” Levitan quipped.

The founders have kept a close eye on developments in AI over the years. Hoffer remembers the iconic faceoff in 2010, when “Jeopardy!” icons Ken Jennings and Brad Rutter failed to defeat IBM’s AI, Watson, in a televised match. He watched alongside some engineers, “gobsmacked” at how this computer made legendary trivia masterminds look like amateurs.

Watson could quickly recall basic trivia facts, but it failed in categories like “also on your computer keys” and “one buck or less,” which required some lived human experience to conjure the correct response. Unfortunately for team humanity, these foibles weren’t enough for Jennings and Rutter to prevail.

AIs still struggle to overcome similar limitations. While an AI could write something that resembles a TV pilot, it won’t be very interesting, and it will likely contain copyrighted material. So, Levitan’s predictions for the future of AI aren’t too foreboding. He predicts that soon, we’ll be able to voice control bots like ChatGPT, but he doubts that AI will ever truly become sentient.

“I am a believer in human nature, and that everyone has a very personal voice,” Levitan told TechCrunch. Levitan’s biggest concern with AI is that people will believe everything they read, just because a computer said it. But Hoffer is more worried that the consequences of this technology will pose even larger issues.

“We’re right before the opening, or maybe we’ve just opened Jurassic Park,” Hoffer said. “How far are we from a lens in my eye that has AR hooked up to a bot? Probably not very.”

When SmarterChild typed its final words in 2008, I was in seventh grade, and my friends had finally come around to joining AIM. My away message was usually the lyrics to “Decode,” the song that Paramore wrote for the “Twilight” movie. Now that my buddy list was a bit more fleshed out, I chatted with my classmates about who we had crushes on, always using code names, because we were so deeply paranoid that these clueless thirteen-year-old boys would somehow manage to hack our AIM accounts and read what we said about them.

I was so captivated by the gut-wrenching highs and lows of middle school drama that I didn’t even realize what happened. My first friend on the internet had logged off forever.


Source link