Is ChatGPT finally getting it right? Our 60-year obsession with chatbots

Summary

The script traces the history of chatbots, from ELIZA in 1966 to modern conversational AI like ChatGPT. It explores early chatbots meant to simulate human conversations versus task-oriented bots designed to help users accomplish goals. Though the technology has advanced dramatically, the allure of natural language interfaces and computer companions persists. The narrator questions if we should want technology that fools us into thinking we’re talking to humans. Either way, today’s relatively capable chatbots let us evaluate whether the long-held theory that human-computer conversation enhances accessibility and usability holds true.

Takeaways

• 💻 ELIZA, developed by Joseph Weizenbaum in 1966, is considered the first conversational bot, designed to mimic a therapist.
• 👨‍💻 The illusion of understanding by ELIZA convinced many, including Weizenbaum's own secretary, of its human-like responses.
• 🧑‍💻 Early chatbots like Parry and Dr. Sbaitso aimed to simulate human-like conversations for therapeutic purposes or companionship.
• 📲 SmarterChild, a chatbot introduced via AOL Instant Messenger in the early 2000s, combined fun interaction with practical functionalities like news and homework help.
• 🎧 The transition to voice-activated assistants like Siri, Alexa, and Google Assistant marked a significant step but also highlighted the challenges in achieving seamless human-computer interaction.
• 📚 The introduction of transformers in 2017, as detailed in the paper 'Attention Is All You Need,' revolutionized AI by improving the way computers understand and process language.
• 🔧 ChatGPT, launched in late 2022, leveraged transformers to deliver more accurate and human-like responses, changing public perception of conversational AI.
• 🛠 Despite advancements, the technology still faces limitations, as evidenced by mixed results in natural language understanding and task execution by voice assistants.
• 👥 AI companions like Replika and Character AI, focused on creating virtual relationships, demonstrate the growing desire for more lifelike interactions with technology.
• 📱 The ongoing development and integration of chatbots and AI assistants highlight a persistent belief in the potential of conversational AI to make technology more accessible and enjoyable.

Q & A

Who was Joseph Weizenbaum and what was his contribution to chatbot technology?

Joseph Weizenbaum was a professor at MIT who, in 1966, created ELIZA, one of the first chatbots, which simulated a Rogerian psychotherapist and demonstrated the potential of natural language conversation between computers and humans.

What was ELIZA and how did it work?

ELIZA was a chatbot developed by Joseph Weizenbaum that simulated a therapist by using a pattern matching and substitution methodology to generate responses, based on keywords found in the user's input.

What was the significance of the paper titled 'Attention Is All You Need'?

The paper 'Attention Is All You Need', published by Google researchers, introduced the concept of transformers, a groundbreaking method in AI that improved the ability of machines to understand and process large amounts of data simultaneously, significantly advancing natural language processing.

How did transformers change the landscape of AI and chatbots?

Transformers allowed computers to process and understand text in a much more comprehensive manner, considering entire sentences or paragraphs at once, leading to vastly improved accuracy in language models and paving the way for advanced AI like ChatGPT.

What was the impact of SmarterChild on chatbot development?

SmarterChild was a chatbot for AOL Instant Messenger that combined information retrieval with engaging conversation, influencing the direction of chatbot development towards both functional and social interaction.

How did ChatGPT differ from its predecessors like ELIZA?

ChatGPT, built on the transformer technology, is capable of understanding and generating human-like text responses, making it significantly more advanced in processing and responding to natural language queries compared to keyword-based predecessors like ELIZA.

What role did academic research play in the development of chatbots?

Academic research has been crucial in the development of chatbots, providing foundational theories and technologies like ELIZA and transformers, which have guided the evolution of conversational AI towards more sophisticated models like ChatGPT.

What was the main goal of chatbot technology according to the script?

The main goal of chatbot technology, as described in the script, is to enable natural language conversations between humans and computers, making technology more accessible, useful, and enjoyable to interact with.

How have public perceptions of chatbots changed over time?

Public perceptions of chatbots have evolved from viewing them as simple novelty or utility tools to sophisticated AI companions capable of meaningful interaction, as demonstrated by the transition from ELIZA to modern AI like ChatGPT.

What challenges still exist in the field of chatbots and conversational AI?

Despite advancements, challenges in conversational AI include improving the understanding of context, managing nuanced human interactions, and bridging the gap between AI-generated responses and genuine human-like understanding.

Outlines

00:00

🤖 The Genesis of Chatbots

This segment introduces the narrative of conversational bots, beginning with ELIZA, the first-ever chatbot developed by Joseph Weizenbaum in 1966. It highlights the significance of ELIZA as a pioneering venture into the realm of human-computer interaction through natural language processing. Despite its simplicity, ELIZA was able to simulate a therapist, impressing users with its ability to engage in a basic form of conversation by searching for keywords and crafting responses based on a predefined set of answers. This early success indicated the potential of chatbots to mimic human-like interactions, paving the way for future developments in the field. The narrative underscores the fascination and inherent human tendency to attribute human qualities to machines, an aspect that has driven the evolution of chatbots over decades.

05:01

🌐 The Evolution of Conversational AI

This paragraph delves into the development and impact of chatbots following ELIZA, with a specific focus on SmarterChild, a notable chatbot from the early 2000s that was popular on AOL Instant Messenger. SmarterChild was distinguished by its ability to provide information and engage users with its personality, laying the groundwork for the integration of chatbot technology in various customer service applications. The narrative also explores the advancements in AI that led to the creation of transformers and the introduction of ChatGPT, which significantly improved the ability of machines to understand and process natural language by considering larger contexts. This section emphasizes the transformative potential of these technologies in making human-computer interaction more natural and efficient, reflecting on the ongoing journey towards achieving seamless conversational AI.

10:02

🤔 Reflecting on the Human-AI Relationship

The final segment contemplates the broader implications of chatbot advancements, questioning the desirability and ethics of technology that blurs the line between human and machine interaction. It discusses the emotional and psychological impact of increasingly sophisticated AI companions, such as Replika and Character AI, which offer personalized conversational experiences. The narrative raises critical questions about the future role of technology in human lives, whether it should mimic human interactions to the point of indistinguishability, and the potential consequences of such developments. This reflective discussion underscores the deep-rooted human desire to humanize technology, while also recognizing the limitations and ethical considerations of this pursuit.

Keywords

💡Chatbot

A chatbot is a computer program designed to simulate conversation with human users, especially over the internet. Chatbots like ELIZA, SmarterChild, Alexa, Siri and ChatGPT are mentioned throughout the video as early and recent examples of attempts to create natural conversations between humans and machines.

💡Natural language processing

The ability of a computer program to understand, interpret and generate human languages. The goal of chatbots is to process natural language input from users and respond in a natural way, the way a human would.

💡Transformers

A type of natural language processing model introduced in 2017 that represented a major advance in chatbots' ability to understand context and respond more intelligently. Transformers enabled chatbots like ChatGPT to process much more information and make better predictions.

💡Illusion of understanding

The ability of early chatbots like ELIZA to trick people into thinking a human must be behind the conversation, even though the technology was very limited. Their techniques gave an illusion of understanding the user.

💡ActiveBuddy

The company that created SmarterChild, one of the first widely used chatbots that many people interacted with in the early 2000s. It was designed to be useful and fun to talk to.

💡Voice assistants

Chatbots like Siri, Alexa and Cortana that accept voice input and respond out loud instead of using text. For years tech companies thought voice would be the ideal chatbot interface, but it has significant limitations.

💡AI companions

A category of chatbots like Replika and Character AI designed to emotionally support users by building relationships with them through ongoing conversations.

💡Augmented reality

The use of technology to overlay digital elements onto the real world. Chatbots may someday leverage AR to give themselves virtual bodies, allowing more natural back-and-forth conversations.

💡Human-computer interaction

The study of how people interface with computing technology, including areas like interface design and user experience. Improving human-computer interaction has been a driving goal behind chatbots since the 1960s.

💡Joseph Weizenbaum

The MIT professor who created ELIZA in 1966, one of the first chatbots. His work established the possibility of simulated conversation between humans and machines mediated by natural language.

Highlights

ELIZA, the first conversational bot created in 1966 by Joseph Weizenbaum, marked the beginning of chatbots.

ELIZA's basic technology simulated a therapist by using keywords to generate responses, demonstrating early natural language processing.

The illusion of understanding by ELIZA convinced users of its human-like interaction, including Weizenbaum's own secretary.

The emergence of chatbots like Parry, Dr. Sbaitso, and A.L.I.C.E further explored the potential of conversational agents in various contexts.

GUS at Xerox PARC aimed to enable task completion via natural language chat, illustrating early attempts at utility-driven chatbots.

SmarterChild, introduced on AOL Instant Messenger, combined fun interaction with practical utility, signifying chatbots' mainstream potential.

The transition to voice-activated chatbots like Siri, Alexa, and Google Assistant marked a significant evolution towards more natural human-computer interaction.

"Attention Is All You Need", a paper published by Google researchers, introduced transformers, revolutionizing AI's understanding capabilities.

ChatGPT, leveraging transformer technology, significantly enhanced chatbot responses, moving closer to human-like interaction.

ChatGPT's ability to generate thoughtful answers highlighted its advanced AI, despite its primary design as a utility tool.

The enduring novelty and forgiveness towards chatbots' mistakes reflect our anthropomorphic treatment of conversational AI.

Emergence of AI companions like Replika and Character AI, focusing on creating virtual relationships, signifies chatbots' evolving social role.

Augmented reality integration with chatbots forges towards lifelike virtual companions, pushing the boundaries of human-bot interaction.

The ongoing debate about the ethical implications of indistinguishable human-bot interaction underlines the complexity of conversational AI's future.

The realization of chatbots as a longstanding vision for the future of computing, reflecting on Joseph Weizenbaum's foundational work.

The exploration of whether chatbots will fundamentally change our interaction with technology remains open, amidst advancements and ethical queries.

Transcripts

(laptop bangs)

(gentle music) (hands typing)

- [David] This is not ChatGPT.

It's also not Gemini and it's not Copilot

and it's not any of the other bots

that everybody has been talking about for the last year.

What it is is a replica

of the first conversational bot to ever exist,

and it's the beginning of the story

that leads to all those other ones,

and that story starts all the way back in 1966.

- [Narrator] Wherever man goes, computers surround

changing the nature of our lives.

- I know, I know.

1966 is a very long time ago,

but stay with me here.

We're going somewhere.

So there's this guy, Joseph Weizenbaum

who's a professor at MIT

and in 1966 he publishes a paper in a journal called

The Communications for The Association

for Computing Machinery.

In that paper, he describes this thing that he's made.

A chat bot named ELIZA,

"Which," and I'm quoting here,

"makes certain kinds of natural language conversation

between man and computer possible."

ELIZA was set up to have the personality of a therapist,

but it was really basic technology underneath.

Basically you typed a sentence

and it looked for keywords in those sentences

and then applied those keywords

to one of a bunch of pre-programmed responses it had ready.

Sometimes if it didn't know what to say

or didn't know what you were saying,

it would just repeat your words back to you.

Here's a bit of the back and forth

Weizenbaum included in that original paper.

- [Writer] Men are all alike.

- [ELIZA] In what way?

- [Writer] They're always bugging us

about something or other.

- [ELIZA] Can you think of a specific example?

- [Writer] Well, my boyfriend made me come here.

- [ELIZA] Your boyfriend made you come here.

- It goes on like that for a while and it's super basic,

but you know what's wild?

It totally worked.

A year later Weizenbaum wrote another article

in the same journal in which he said

it had been hard to convince some people who tested ELIZA

that there wasn't a human on the other end,

that includes his own secretary.

In the paper Weizenbaum wrote that her reaction was proof

of ELIZA's illusion of understanding,

and he even recited the anecdote to camera.

- After two or three interchanges with the machine,

she turned to me and she said,

"Would you mind leaving the room please?"

- This is the thing about chatbots.

There's just something about talking to a computer

and the computer talking back

that feels like magic.

We give these bots human attributes

even when they don't have any.

We talk to them differently,

we appreciate them more,

we work with them more collaboratively.

We just wanna use them more.

And for decades people have believed

that once the underlying tech gets good enough,

it'll go from cool demo and fun thing to talk with,

to changing everything

about how we interact with technology.

- What we really wanna do is just talk to our device.

- And so for decades they really tried to get there

and eventually it got to be kind of good.

We'll get into how we got there

right after this break.

- More and more we're seeing AI tools

be integrated into our daily lives.

From generating quick inspiring art

to capturing notes from important meetings.

But with SAP Business AI,

their tools are designed to deliver real world results,

helping your business become stronger

and helping you make decisions faster.

This revolutionary AI technology

allows you to be ready for anything that is thrown at you.

Okay, that's it for me,

but before we go,

SAP doesn't influence the editorial of this video,

but they do help make videos like this possible.

- After ELIZA, there were a few other

well-known early chatbots.

There was Parry which tried to simulate

a person with schizophrenia.

There was Dr. Sbaitso which also tried

to act like a psychologist.

There was A.L.I.C.E who was just a chill, friendly bot

that just kind of wanted to be your friend.

The motivation behind these bots

was to be immersive for lack of a better word.

To give you a sense of talking to a therapist

that felt, if not real,

then real enough to actually be useful therapy.

You can read about all that stuff,

but this was designed to be much closer to experiencing it.

Around the same time,

there was also another branch of research being done

in the chatbot world.

Back in the '70s at the famous Xerox PARC research lab,

there was a group working on this thing called GUS,

the Genial Understander System,

which was meant to be a way

to get stuff done on your computer

just by talking in a natural language to a chatbot.

The team's example was a travel agent.

What if you could buy plane tickets on a computer

the same way you would over the phone with a human?

This is a much clearer simpler use case

for this kind of technology.

And to a lot of people even then

felt like something you could make money from too.

There were lots of projects like this over the years,

but I bet that at least for anyone over about 30,

the first time you really used a chatbot,

was when you used SmarterChild.

(bright music)

SmarterChild was a chat bot made for AOL Instant Messenger,

which is an extremely 2004 sentence to say,

but it was a huge deal.

It was made by a company called ActiveBuddy

and it was a mix of all of the things we'd seen before.

It had access to lots of information

about news and stocks and such.

It could do math and help you with your homework,

but most of all,

it was also just really fun to talk to.

- We were in the Buddy list.

We were to people exactly

what their friends were in the Buddy list.

- [David] That's Peter Levitan,

one of the founders of ActiveBuddy.

He had a team of writers working on SmarterChild

who were coming up with the bots responses.

- SmarterChild had a great personality.

If somebody cursed at it, it had a response.

So we understood what people were saying

and you know if it was an 11-year-old boy,

he had his own particular approach to communication.

We were surprised at how popular it got

and how crazy the conversations became.

- In the mid 2000s,

ActiveBuddy licensed this tech out to lots of places.

It eventually figured out

that there was no money to be made from SmarterChild itself,

but ultimately a lot of early bots

from like your cable company

and your wireless carrier were powered by this same tech.

- SmarterChild was our demonstration

of our skillset and technology,

and that's what we showed commercial clients.

Ultimately the goal was, believe it or not,

to make money.

- Eventually Microsoft bought it

with even more corporate plans than that,

and SmarterChild was gone.

Now you basically can't find it anywhere on the internet.

We found this site which is something

of an homage to SmarterChild,

and that was the closest thing we could get our hands on.

(animation music)

After SmarterChild taught a bunch of AM kids

how to talk to chatbots,

the next big phase was what I guess you'd call

the voice generation.

- It's a feature all about our voice.

- [Lady] Okay Google, what do I have to do today?

- Alexa, open Cortana.

- This is when everyone got really excited about the idea

that not only were chatbots the future,

but we'd talk to them instead of typing to them.

And if you apply the logic going all the way back to 1966,

that makes perfect sense, right?

The goal is to make interacting with technology

feel like interacting with humans,

which just makes everything better.

But I mean, you've used Siri and Alexa, right?

They're great for a few things

and mostly terrible for everything else,

and honestly God help you

if you try to book a flight through Google Assistant.

All these companies and developers work convinced for years

that they were hitting on the right interface.

But even as the technology got better,

it became increasingly obvious

just how far it still had to go.

- Sorry, I don't understand.

(instrumental music)

- [David] And then in 2017,

another academic paper came out.

Sorry, we're talking about so many academic papers here,

but it is important.

This one was called "Attention Is All You Need"

and it was written by a bunch of Google researchers.

They laid out this idea of transformers,

which are not robots in disguise,

but instead a way of teaching computers

to understand and process information

that is just vastly more effective and efficient

than anything we'd had before.

This is going to be a way oversimplified explanation,

but basically what transformers do

is allow a computer to read and understand

much more at a time.

Instead of seeing one word

and then predicting the next word,

which is how it used to work,

transformers let the computer see the whole sentence,

the whole paragraph,

even the whole book at the same time.

That gives it vastly more data to work with

and makes its predictions

as a result much more accurate.

In the history of AI and chatbots and maybe everything,

there's the world before transformers and the world after,

and a few years later we started to see that become real,

particularly in late 2022 when ChatGPT came out

and seemingly changed everything.

- ChatGPT.

- [Reporter] ChatGPT.

- It's called ChatGPT which is just about

the clunkiest name ever.

- Do you remember the first time you used ChatGPT,

how it felt like when you typed a question,

the bot was actually thinking and processing,

and it came back with an answer

that wasn't just a formulaic keyword response,

but actually felt like someone was answering your question.

Remember how cool that felt

and then how everyone immediately was like,

"Oh man, the movie "Her" is coming true.

Scarlett Johansson is totally gonna be my girlfriend

in my ear in like two years."

That was in a lot of ways

the same experience people were having with a ELIZA

almost 60 years ago.

Well, they didn't talk about Scarlett Johansson back then,

but you know what I mean.

The tech in ChatGPT is I mean wildly better

than what was under ELIZA.

Let's just rerun some of that first conversation

from 1966 with ChatGPT and see where we get.

- [Writer] Men are all alike.

- [ChatGPT] It is important to recognize

that making sweeping generalizations

about any group of people, including men,

can be inaccurate and unfair, individuals--

- Okay, that's a bad example.

Or actually maybe that's a good example.

ChatGPT probably handled that well,

not as a therapist but as a chatbot.

and honestly, that is a deep thoughtful answer

to what I said.

ChatGPT is mostly designed

to be a get things done chatbot, not a companion.

It can help you write code and write email

and brainstorm ideas, find information,

all that kind of stuff.

And even when ChatGPT or any other AI bot

gets stuff wrong or makes mistakes,

which by the way they do all the time,

people forgive them.

After all this time,

and as sophisticated as we are about technology,

we still treat these bots like we would people.

I mean, if you put numbers into a calculator

and it gave you the wrong answer,

that's a bad calculator, right?

You'd return it.

But because ChatGPT talks

and appears to think like a person,

we're willing to go back and forth,

we're willing to give it the benefit of the doubt.

Maybe the novelty of that will wear off eventually,

but for many people it sure hasn't yet.

Meanwhile, all the way on the other end of the bot spectrum,

the bots that are purely meant for conversation

are getting better all the time too.

There are companies like Replika and Character AI

that are deliberately building AI companions.

Someone to talk to, someone who will listen,

a best friend in the cloud.

Even Meta is big on this idea now.

More and more people are signing up for that too.

People want and respond to this kind of relationship.

And if you add in augmented reality,

those bots are even able to have virtual bodies,

and these robot companions

are becoming more lifelike all the time.

(bright music)

Is this what we want from technology?

Technology that tricks us into thinking

we're talking to a human even when we're not?

Do we even need to know the difference

between human and bot anymore?

Is there even a distinction?

I don't know,

but the whole history of chatbots

has been based on this one theory,

that we should be able to talk to computers

like we talk to each other,

and that that might make technology more accessible,

more useful, and just more fun to use.

There are a lot of smart people who have believed that

for a really long time,

and a lot of smart people who think that whole theory

is dead wrong.

But the thing is we've never had a chance

to prove it either way,

because it hasn't worked well enough.

And to be clear,

even now we're not at some magical AI moment

where the tech is perfect

and will change everything forever,

but the chat bot has been the future of computers

for just about as long as there have been computers.

Now they're just here,

they're pretty good,

and now we get to see once and for all

whether Joseph Weizenbaum really had it right

all those years ago.

(bright music)

What's a good last minute gift to get my wife?

A star map,

a terrarium or indoor garden kit?

Sure, oh, virtual reality headset.

I don't think she's going to like a virtual reality headset.