With how new and viral ChatGPT has been, it’s easy to forget chatbots have been here for decades. We’ve seen them as friends, therapists, services; but why are we so attached to talking to computers? Editor-at-large, David Pierce, dives into the history of chatbots starting with the first one ever made known as ELIZA. He explores how interactions with these bots have evolved over time in both technology and personality.
Category
🤖
TechTranscript
00:00 (door slams)
00:02 This is not ChatGPT.
00:05 It's also not Gemini, and it's not Copilot,
00:07 and it's not any of the other bots
00:09 that everybody has been talking about for the last year.
00:12 What it is is a replica
00:13 of the first conversational bot to ever exist.
00:17 And it's the beginning of the story
00:18 that leads to all those other ones.
00:20 And that story starts all the way back in 1966.
00:24 - Wherever man goes, computers surround him,
00:27 changing the nature of our lives.
00:29 - I know, I know, 1966 is a very long time ago,
00:33 but stay with me here, we're going somewhere.
00:35 So there's this guy, Joseph Weizenbaum,
00:37 who's a professor at MIT, and in 1966,
00:40 he publishes a paper in a journal called
00:42 "The Communications for the Association
00:44 "for Computing Machinery."
00:46 In that paper, he describes this thing that he's made,
00:49 a chatbot named Eliza, which, and I'm quoting here,
00:52 "makes certain kinds of natural language conversation
00:55 "between man and computer possible."
00:58 Eliza was set up to have the personality of a therapist,
01:01 but it was really basic technology underneath.
01:04 Basically, you typed a sentence,
01:07 and it looked for keywords in those sentences,
01:09 and then applied those keywords
01:11 to one of a bunch of pre-programmed responses it had ready.
01:15 Sometimes, if it didn't know what to say,
01:17 or didn't know what you were saying,
01:19 it would just repeat your words back to you.
01:21 Here's a bit of the back and forth
01:23 Weizenbaum included in that original paper.
01:25 - Men are all alike.
01:26 - In what way?
01:27 - They're always bugging us about something or other.
01:30 - Can you think of a specific example?
01:32 - Well, my boyfriend made me come here.
01:34 - Your boyfriend made you come here.
01:35 - It goes on like that for a while, and it's super basic,
01:38 but you know what's wild?
01:40 It totally worked.
01:42 A year later, Weizenbaum wrote another article
01:44 in the same journal in which he said
01:46 it had been hard to convince some people who tested Eliza
01:49 that there wasn't a human on the other end.
01:52 That includes his own secretary.
01:54 In the paper, Weizenbaum wrote that her reaction
01:56 was proof of Eliza's illusion of understanding.
02:00 And he even recited the anecdote to camera.
02:03 - After two or three interchanges with the machine,
02:05 she turned to me and she said,
02:07 "Would you mind leaving the room, please?"
02:09 - This is the thing about chatbots.
02:11 There's just something about talking to a computer
02:14 and the computer talking back that feels like magic.
02:18 We give these bots human attributes
02:19 even when they don't have any.
02:21 We talk to them differently.
02:22 We appreciate them more.
02:24 We work with them more collaboratively.
02:27 We just wanna use them more.
02:29 And for decades, people have believed
02:31 that once the underlying tech gets good enough,
02:33 it'll go from cool demo and fun thing to talk with
02:37 to changing everything
02:38 about how we interact with technology.
02:39 - What we really wanna do is just talk to our device.
02:42 - And so for decades, they really tried to get there.
02:45 And eventually, it got to be kind of good.
02:47 We'll get into how we got there right after this break.
02:51 - More and more, we're seeing AI tools
02:52 be integrated into our daily lives,
02:54 from generating quick, inspiring art
02:57 to capturing notes from important meetings.
02:59 But with SAP Business AI,
03:01 their tools are designed to deliver real-world results,
03:04 helping your business become stronger
03:05 and helping you make decisions faster.
03:07 This revolutionary AI technology allows you to be ready
03:10 for anything that is thrown at you.
03:12 Okay, that's it for me.
03:13 But before we go, SAP doesn't influence
03:16 the editorial of this video,
03:17 but they do help make videos like this possible.
03:20 After Eliza, there were a few other
03:22 well-known early chatbots.
03:24 There was Perry, which tried to simulate
03:26 a person with schizophrenia.
03:27 There was Dr. Spazzo,
03:29 which also tried to act like a psychologist.
03:31 There was Alice, who was just a chill, friendly bot
03:34 that just kinda wanted to be your friend.
03:36 The motivation behind these bots was to be immersive,
03:39 for lack of a better word,
03:40 to give you a sense of talking to a therapist that felt,
03:44 if not real, then real enough to actually be useful therapy.
03:48 You can read about all that stuff,
03:49 but this was designed to be much closer to experiencing it.
03:53 Around the same time, there was also another branch
03:55 of research being done in the chatbot world.
03:58 Back in the '70s, at the famous Xerox PARC Research Lab,
04:01 there was a group working on this thing called GUS,
04:03 the Genial Understander System,
04:06 which was meant to be a way to get stuff done
04:08 on your computer just by talking
04:10 in a natural language to a chatbot.
04:12 The team's example was a travel agent.
04:14 What if you could buy plane tickets on a computer
04:17 the same way you would over the phone
04:19 with a human?
04:20 This is a much clearer, simpler use case
04:22 for this kind of technology,
04:24 and to a lot of people even then,
04:26 felt like something you could make money from, too.
04:28 There were lots of projects like this over the years,
04:31 but I bet that at least for anyone over about 30,
04:35 the first time you really used a chatbot
04:37 was when you used SmarterChild.
04:40 SmarterChild was a chatbot made for AOL Instant Messenger,
04:43 which is an extremely 2004 sentence to say,
04:47 but it was a huge deal.
04:49 It was made by a company called ActiveBuddy,
04:51 and it was a mix of all of the things we'd seen before.
04:55 It had access to lots of information
04:56 about news and stocks and such.
04:59 It could do math and help you with your homework,
05:01 but most of all, it was also just really fun to talk to.
05:05 - We were in the buddy list.
05:07 We were to people exactly what their friends were
05:10 in the buddy list.
05:11 - That's Peter Levitin, one of the founders of ActiveBuddy.
05:14 He had a team of writers working on SmarterChild
05:17 who were coming up with the bots' responses.
05:19 - SmarterChild had a great personality.
05:22 If somebody cursed at it, it had a response.
05:25 So we understood what people were saying.
05:27 And if it was an 11-year-old boy,
05:29 he had his own particular approach to communication.
05:32 We were surprised at how popular it got
05:35 and how crazy the conversations became.
05:38 - In the mid-2000s, ActiveBuddy licensed this tech out
05:41 to lots of places.
05:42 It eventually figured out that there was no money
05:44 to be made from SmarterChild itself,
05:47 but ultimately, a lot of early bots
05:49 from your cable company and your wireless carrier
05:52 were powered by this same tech.
05:54 - SmarterChild was our demonstration
05:57 of our skillset and technology,
05:59 and that's what we showed commercial clients.
06:03 Ultimately, the goal was, believe it or not, to make money.
06:06 - Eventually, Microsoft bought it
06:09 with even more corporate plans than that,
06:11 and SmarterChild was gone.
06:13 Now, you basically can't find it anywhere on the internet.
06:16 We found this site, which is something of an homage
06:18 to SmarterChild, and that was the closest thing
06:21 we could get our hands on.
06:22 After SmarterChild taught a bunch of AIM kids
06:26 how to talk to chatbots, the next big phase
06:29 was what I guess you'd call the voice generation.
06:32 - It's a feature all about our voice.
06:34 - Okay, Google, what do I have to do today?
06:36 - Alexa, open Cortana.
06:39 - This is when everyone got really excited about the idea
06:42 that not only were chatbots the future,
06:44 but we'd talk to them instead of typing to them.
06:47 And if you apply the logic going all the way back to 1966,
06:51 that makes perfect sense, right?
06:52 The goal is to make interacting with technology
06:55 feel like interacting with humans,
06:57 which just makes everything better.
06:58 But I mean, you've used Siri and Alexa, right?
07:01 They're great for a few things
07:02 and mostly terrible for everything else.
07:04 And honestly, God help you if you try to book a flight
07:07 through Google Assistant.
07:08 All these companies and developers were convinced for years
07:12 that they were hitting on the right interface,
07:14 but even as the technology got better,
07:16 it became increasingly obvious
07:18 just how far it still had to go.
07:20 - Sorry, I don't understand.
07:22 - And then in 2017, another academic paper came out.
07:29 Sorry, we're talking about so many academic papers here,
07:31 but it is important.
07:32 This one was called "Attention is All You Need,"
07:35 and it was written by a bunch of Google researchers.
07:38 They laid out this idea of transformers,
07:40 which are not robots in disguise,
07:42 but instead a way of teaching computers
07:44 to understand and process information
07:47 that is just vastly more effective and efficient
07:50 than anything we'd had before.
07:52 This is going to be a way oversimplified explanation,
07:55 but basically what transformers do
07:57 is allow a computer to read and understand
08:00 much more at a time.
08:02 Instead of seeing one word
08:04 and then predicting the next word,
08:05 which is how it used to work,
08:07 transformers let the computer see the whole sentence,
08:10 the whole paragraph, even the whole book at the same time.
08:13 That gives it vastly more data to work with
08:16 and makes its predictions as a result much more accurate.
08:19 In the history of AI and chatbots and maybe everything,
08:24 there's the world before transformers and the world after.
08:27 And a few years later, we started to see that become real,
08:30 particularly in late 2022 when ChatGPT came out
08:34 and seemingly changed everything.
08:36 - ChatGPT.
08:37 - ChatGPT.
08:38 - It's called ChatGPT,
08:40 which is just about the clunkiest name ever.
08:42 - Do you remember the first time you used ChatGPT,
08:45 how it felt like when you typed a question,
08:47 the bot was actually thinking and processing
08:50 and it came back with an answer
08:52 that wasn't just a formulaic keyword response,
08:55 but actually felt like someone was answering your question?
08:58 Remember how cool that felt?
09:00 And then how everyone immediately was like,
09:02 "Oh man, the movie 'Her' is coming true.
09:04 "Scarlett Johansson's totally gonna be my girlfriend
09:06 "in my ear in like two years."
09:08 That was, in a lot of ways,
09:10 the same experience people were having with Eliza
09:13 almost 60 years ago.
09:15 Well, they didn't talk about Scarlett Johansson back then,
09:17 but you know what I mean.
09:18 The tech in ChatGPT is, I mean,
09:20 wildly better than what was under Eliza.
09:23 Let's just rerun some of that first conversation
09:25 from 1966 with ChatGPT and see where we get.
09:29 - Men are all alike.
09:30 - It is important to recognize
09:32 that making sweeping generalizations
09:34 about any group of people, including men,
09:37 can be inaccurate and unfair.
09:38 Individuals-- - Okay, that's a bad example.
09:40 Or actually, maybe that's a good example.
09:42 ChatGPT probably handled that well,
09:45 not as a therapist, but as a chatbot.
09:48 And honestly, that is a deep,
09:49 thoughtful answer to what I said.
09:52 ChatGPT is mostly designed to be a get-things-done chatbot,
09:55 not a companion.
09:57 It can help you write code and write email
09:59 and brainstorm ideas, find information,
10:02 all that kind of stuff.
10:03 And even when ChatGPT or any other AI bot
10:06 gets stuff wrong or makes mistakes,
10:08 which, by the way, they do all the time,
10:10 people forgive them.
10:12 After all this time,
10:13 and as sophisticated as we are about technology,
10:16 we still treat these bots like we would people.
10:19 I mean, if you put numbers into a calculator
10:21 and it gave you the wrong answer,
10:23 that's a bad calculator, right?
10:25 You'd return it.
10:26 But because ChatGPT talks and appears to think like a person,
10:31 we're willing to go back and forth.
10:33 We're willing to give it the benefit of the doubt.
10:35 Maybe the novelty of that will wear off eventually,
10:38 but for many people, it sure hasn't yet.
10:40 Meanwhile, all the way on the other end of the bot spectrum,
10:43 the bots that are purely meant for conversation
10:46 are getting better all the time, too.
10:48 There are companies like Replica and Character AI
10:51 that are deliberately building AI companions,
10:54 someone to talk to, someone who will listen,
10:56 a best friend in the cloud.
10:58 Even Meta is big on this idea now.
11:01 More and more people are signing up for that, too.
11:04 People want and respond to this kind of relationship.
11:07 And if you add in augmented reality,
11:10 those bots are even able to have virtual bodies,
11:13 and these robot companions are becoming
11:14 more lifelike all the time.
11:16 Is this what we want from technology,
11:20 technology that tricks us into thinking
11:22 we're talking to a human even when we're not?
11:24 Do we even need to know the difference
11:26 between human and bot anymore?
11:28 Is there even a distinction?
11:30 I don't know, but the whole history of chatbots
11:33 has been based on this one theory,
11:35 that we should be able to talk to computers
11:37 like we talk to each other,
11:38 and that that might make technology more accessible,
11:41 more useful, and just more fun to use.
11:44 There are a lot of smart people who have believed that
11:46 for a really long time, and a lot of smart people
11:49 who think that whole theory is dead wrong.
11:51 But the thing is, we've never had a chance
11:53 to prove it either way,
11:55 because it hasn't worked well enough.
11:56 And to be clear, even now,
11:57 we're not at some magical AI moment
12:00 where the tech is perfect
12:01 and will change everything forever.
12:03 But the chatbot has been the future of computers
12:06 for just about as long as there have been computers.
12:09 Now they're just here.
12:11 They're pretty good.
12:12 And now we get to see once and for all
12:14 whether Joseph Weizenbaum really had it right
12:17 all those years ago.
12:18 What's a good last minute gift to get my wife?
12:25 A star map?
12:28 A terrarium or indoor garden kit?
12:31 Sure.
12:32 Oh, a virtual reality headset.
12:33 I don't think she's going to like a virtual reality headset.