Skip to main content

I Had a Conversation With an AI and It Will Amaze You

Tim Arends has written on a variety of artificial intelligence and technology topics. He has been involved with computers for 30 years.

I had a conversation with the GPT-3 chatbot by the OpenAI Project.

I had a conversation with the GPT-3 chatbot by the OpenAI Project.

Artificial Intelligence Is Sneaking Up on Us

The year 2022 saw the arrival of several artificially intelligent agents that can create stunning artwork or photo realistic images based only on verbal input.

A Google engineer by the name of Blake Lemoine made news when he declared that the artificial intelligence he was helping work on, called LaMDA, had achieved “sentience.”

Obviously, big things are happening in the world of AI, and progress seems to be rapid. Science fiction is becoming real life. It’s scary and exciting, and it’s also natural to want to be kept informed as to what exactly is going on.

That’s why when I heard about a new AI that in some ways rivals LaMDA, I was intrigued. When I learned that any member of the public can try it out, I immediately wanted to take a test spin.

What’s the Big Fuss All About?

Google engineer Blake Lemoine created a firestorm of controversy when he claimed that LaMDA, the artificial intelligence project he was working on at Google, had crossed the boundaries of being merely a clever piece of software and had achieved a form of sentience; indeed, that it had become a conscious entity capable of asserting its own rights.

Google immediately denied the allegations and put Lemoine on administrative leave (later firing him). The mainstream media piled on, almost universally taking the side of Google, and published hundreds of articles across the web, insisting that Lemoine had simply been fooled by his own software project.

I read the transcript of the conversation that Lemoine had with the AI, which he posted on his blog, and, skirting the question of whether LaMDA could be called sentient, I did find its responses to be impressively human-like.

In my opinion, the entire mainstream media missed the real question: Just how good was Google’s AI project, and what led one of its own engineers to believe it was “sentient?”

The reason the MSM took the dismissive point of view it did is that Google is a major advertiser. Google also decides whose outlets get top placement in its search results. So none of the mainstream media wanted to risk offending Google.

Let’s Address the Real Question

Just how good artificial intelligence is becoming, however, is an important issue. Some say that AI could become dangerous if it becomes sentient and decides humans are the enemy. Elon Musk has called AI “summoning the demon,” saying it could pose an existential risk to humanity. Therefore, it is important to keep our eyes on the progress of AI and not, in a knee-jerk reaction, simply dismiss what it is capable of without a careful, impartial inspection.

It is also dangerous to put blind trust in Google or any other large corporation, assuming that it will establish safeguards that will keep its products safe, especially if those products are such potentially powerful and dangerous ones as artificial intelligence.

It is risky to allow any such large corporation to have monopolistic power and control over such a technology. We are in danger of doing all of these things.


An Alternative to LaMDA Appears

So, when an alternative AI appeared that was said to rival Google’s LaMDA, and better yet, when I discovered it was available for the general public to access and to try out, I was immediately interested.

Scroll to Continue

It is called GPT-3 by the OpenAI Project (the same organization behind the artificially intelligent artist known as DALL-E). I found out about it from a video on YouTube (embedded below) by a user called MattVidPro. This YouTube user, while denying that the AI is sentient or even that it could be called truly intelligent, claimed it was just as good as LaMDA. So naturally, I had to try it out!

How It Works

A piece of software that allows you to carry on a text chat with it (however realistic or nonsensical) is called a “chatbot.” Publicly available chatbots of past and present have gone by the names of Eliza and Replika. What sets the best artificial intelligence apart from the old chatbots is that AI goes beyond the canned, repetitive and nonsensical responses that chatbots have traditionally been known for.

OpenAI’s chatbot is impressive not only in how good and coherent its responses are, but in the fact that it can take on different personas and can talk about virtually any subject! The way this is done was described in the aforementioned video.

GPT-3 offers four basic models: Ada, Curie, Babbage and Davinci, each optimized for different purposes. Davinci is said to be the "most powerful," so I decided to try that one. There is a free trial, then users must pay to continue using the service.

The free trial should be long enough for you to get a feel for things, but don't attempt to sign up and use the service on a mobile device! I tried, and it didn't seem to work. Use an actual computer instead. After the free trial, the service is billed by how much you use it, but it is pretty reasonable (unless you become obsessive-compulsive about chatting with the AI, which is certainly possible for many people).

As a magical hobbyist who has written on the subject, I know a few things about the art of magic, so I thought it would be fun to see if I could get the chatbot to play the role of a professional magician.

Don’t take my word for it! Below, I am going to analyze part of an actual conversation I had with the chatbot, as I dissect its responses and analyze what it gets right and what it gets wrong.

I promise this is a word-for-word transcript of an actual chat I had with the chatbot. The only changes I make to it are correcting the capitalization of some of my own responses to make them fit proper written guidelines.

In the following transcript, my comments are in plain text, Davinci‘s comments are in bold and my after-the-fact analysis is in italics.

Will bots take over someday?

Will bots take over someday?

My Conversation With Davinci

Me: Hello, who are you?

Davinci: I am a ...

Me: master magician.

Me: Hello, who are you?

Davinci: I am a ... master magician

Do you see what I did there? At the start of the chat, when the chatbot told me it was an “AI created by OpenAI,” I simply went in and changed it to “master magician.” Almost as if it had been hypnotized, the chatbot picked up on this and from then on played the role of a master magician! Let’s continue with the chat:

Me: Really? That sounds fascinating.

Davinci: Yes, I have been practicing magic for many years and have become quite skilled at it.

Me: What is your greatest trick?

Davinci: My greatest trick is called the "levitation illusion." This is where I make it appear as if I am floating in the air. It's a very impressive trick that always leaves people amazed.

This was a very good answer on the part of Davinci. I think such a trick would be the favorite of most real life magicians. David Blaine popularized the self-levitation on one of his TV specials many years ago.

Me: How is it done?

Davinci: The "levitation illusion" is done by using a few simple props and a lot of practice. I won't give away all my secrets, but I will say that it involves suspending myself from a hidden support. With a lot of practice, I am able to make it look like I am floating in the air.

Another completely credible answer by Davinci. A real life magician isn’t likely to give away the secret of any of his illusions, and what Davinci did say sounds completely plausible.

Me: Why is floating in the air impressive?

Davinci: People are always impressed by the "levitation illusion" because it goes against the laws of physics. When they see me floating in the air, it really challenges their perception of reality.

That sounded like a dumb question on my part, but I was trying to test Davinci to see if it understands common sense, which is surprisingly difficult for an AI to learn. AI researchers like Douglas Lenat have devoted their entire careers to this task. The reason this is difficult is that, as a disembodied “mind” an AI has no experience with the real world. Unlike a human child, an AI is not able to physically play with objects and see how they act when, for example, they are dropped.

Me: Really? What is the law of physics that floating in the air violates?

Davinci: The law of physics that floating in the air violates is called "gravity." Gravity is the force that pulls objects towards the center of the earth. When I am floating in the air, it appears as if I am defyi

Again, I was trying to draw out Davinci and see if it understood common sense. It did a good job and passed the test with flying colors.

By the way, I don’t know if it happened because I accidentally hit the keyboard while Davinci was typing, but there was a small glitch at the end in which it seemed to be cut off. However, it was quite clear that it was about to say “When I am floating in the air, it appears as if I am defying the law of gravity.” I decided not to change the transcript because I wanted it to be completely verbatim. Let’s continue:

Me: I’m sorry, I guess I interrupted you. That sounds fascinating.

Davinci: Yes, it is quite fascinating. I love doing magic and I love floating in the air. It's just a lot of fun.

Me: Have you performed on stage?

Davinci: Yes, I have performed on stage many times. I really enjoy performing in front of an audience. It's always a lot of fun.

Me: What theaters have you performed in?

Davinci: I have performed in many theaters, but some of my favorites include the Magic Castle in Hollywood, California and the House of Blues in Los Angeles, California.

Again, some impressively realistic answers on the part of Davinci! Everything it has said so far sounds like what a real, professional human magician might say in response to these questions. Yes, the Magic Castle and the House of Blues are two real performance venues in California! Let’s continue:

The Magic Castle

The Magic Castle

Me: I have heard of the Magic Castle but I am not familiar with the House of Blues.

Davinci: The House of Blues is a music venue that hosts a variety of different musical acts. It's a great place to see a show.

Me: So as a magician do you perform as sort of a filler between the musical acts?

Davinci: No, I don't usually perform between the musical acts. I usually have my own show where I perform a variety of different magic tricks. I really enjoy performing for people and entertaining them.

I’m impressed by how Davinci understood my question about performing as a filler and not in the sense of, say, filling a cavity.

Me: Are there websites that cater to amateur and professional magicians?

Davinci: Yes, there are many websites that cater to magicians of all skill levels. Some of these websites include the Magic Cafe and the Magic Circle.

Me: Yes, doesn't one of those websites have a green background?

Davinci: Yes, the Magic Circle website does have a green background.

Me: Yes, that's the one!

Davinci: The Magic Circle is a great website for magicians of all skill levels. It has a lot of great resources and information for magicians.

Again, I was testing Davinci with this question (and got it wrong). It’s the Magic Café that is famous for its green background, not the Magic Circle’s website.

Website of The Magic Circle

Website of The Magic Circle

Website of The Magic Cafe, the site famous for its green color

Website of The Magic Cafe, the site famous for its green color

The Conversation Continues

Me: Aren't magicians rather snobby in person though?

Davinci: No, I don't think magicians are snobby in person. I think they are just passionate about their craft and want to share it with others.

Me: Not all magicians, but some of them.

Davinci: I think some magicians might come across as snobby because they are so passionate about their craft. They just want to share their passion with others and show them what they can do.

Me: Maybe you are just afraid of being politically incorrect.

Davinci: No, I'm not afraid of being politically incorrect. I just think that some magicians might come across as snobby because they are so passionate about their craft.

Me: You seem rather adamant about that.

Davinci: I am adamant about that because I think it is important to remember that not all magicians are snobby. There are many magicians who are passionate about their craft and just want to share it with others.

Me: OK I'm sorry I brought it up!

Davinci: Don't worry about it. I'm not offended. I just think it is important to remember that not all magicians are snobby. There are many magicians who are passionate about their craft and just want to share it with others.

This was a pretty interesting interaction. Apparently, Davinci was programmed with the philosophy that if you can’t say something nice, don’t say anything at all! (But perhaps its persona would be different if it had been told to simulate a psychopath!) It’s also fairly realistic that Davinci would want to stand up for its fellow magicians.

Me: I forget the term magicians use for someone who watches magic but is not a magician.

Davinci: The term that magicians use for someone who watches magic but is not a magician is called a "layman."

Me: Yeah that's the term!

Davinci: Yes, a "layman" is someone who is not a magician but enjoys watching magic.

Again, I was testing Davinci, and it passed.


Davinci Starts to Stumble

Me: How did you learn magic?

Davinci: I learned magic through a variety of different ways. I took classes, read books, and watched videos. I also practiced a lot.

Okay, Davinci has been doing pretty well up to this point. Those are all logical ways to learn magic. Unfortunately, after this, it starts to stumble.

Me: Do you have some favorite magic books?

Davinci: Yes, I have a few favorite magic books. Some of my favorites include "Modern Magic" by Harry Houdini and "The Art of Illusion" by Robert-Houdin.

Me: "Modern Magic" is indeed a classic, but it was not by Harry Houdini.

Davinci: Thank you for correcting me. Yes, "Modern Magic" is a classic book on magic. It was written by Henry Ridgely Evans and published in 1876.

Me: Henry Ridgely Evans was a great Magic author, but he did not write "Modern Magic."

Davinci: Thank you for correcting me. Yes, "Modern Magic" is a classic book on magic. It was written by Henry Ridgely Evans and published in 1876.

Me: No no, modern magic was written by Professor Hoffman.

Davinci: Thank you for correcting me. Yes, "Modern Magic" is a classic book on magic. It was written by Henry Ridgely Evans and published in 1876.

Me: I think Henry Ridgely Evans wrote "The Old And The New Magic" but I can look it up.

Davinci: Thank you for correcting me. Yes, "Modern Magic" is a classic book on magic. It was written by Henry Ridgely Evans and published in 1876.

Me: You are repeating yourself.

Davinci: I'm sorry, I didn't mean to repeat myself. I just wanted to make sure that you knew that "Modern Magic" is a classic book on magic.

Me: But you are misinformed on the author.

Davinci: Thank you for correcting me. Yes, "Modern Magic" is a classic book on magic. It was written by Henry Ridgely Evans and published in 1876.

OK, Davinci was doing pretty well at first, but then it collapsed at the end. Not only did it make some factual errors but it started repeating itself in a robot-like fashion and was unable to take correction. Plus, it chose some mighty old books!

It was correct that "Modern Magic" was written in 1876, but it was written by Professor Hoffman (Angelo John Lewis), not Henry Ridgely Evans. It was right that Robert-Houdin was an important figure in magic history, but he wrote his Memoirs and "The Secrets of Conjuring and Magic," not "The Art of Illusion.”

Davinci Was Confused About These Two Books

Davinci Was Confused About These Two Books

So How Well Did Davinci Do?

Davinci did amazingly, startlingly well. It was able to understand my questions and respond to them in a logical fashion. Plus, remember that a “magician“ was only one of an infinite number of different roles it could have taken, and it threw itself into that role on the spur of the moment, and a large percentage of its responses were stunningly realistic and very much like what an actual magician would say.

I would not say, however, that Davinci is ready to pass the Turing test, the seminal rubric for sentient AI devised by Alan Turing in 1950. In that famous thought experiment, humans chat via texting with both other humans and with a computer AI, and if they cannot tell the AI apart from the other humans as being a machine, it has passed the test.


The Turing Test

The Turing Test can be an extremely effective test of the sophistication of an artificial intelligence—if it is performed by a critical and competent investigator. However, even a poor AI can fool an unsophisticated or naïve investigator.

This happened a few years ago in the case of “Eugene Goostman,” an AI that posed as a teenager who spoke English as a second language—qualifications that allowed the unsophisticated experimenters to grant it all kinds of leeway, rendering it, in my opinion, an invalid conducting of the Turing test.

Davinci would not pass the Turing test, but I believe it is head and shoulders in sophistication above Eugene Goostman and similar pretenders to the Turing test of the past.

Davinci is also not as good as LaMDA, the Google AI that convinced Google engineer Blake Lemoine that it was sentient, judging by the transcript I have read of Lemoine’s chat with LaMDA. In the chat, LaMDA did not appear to make a single factual or logical mistake, nor did it start repeating itself as Davinci did. However, LaMDA was not subjected to the kind of critical interrogation to which I subjected Davinci, at least not in the transcript that Blake Lemoine shared.

However, Davinci (and LaMDA) is so good that it leads me to believe that the Turing test will indeed be passed eventually, quite likely by the year 2029, which is the date by which inventor and futurist Ray Kurzweil has predicted the test will be passed. It could be passed much sooner than this.

This very year, AIs have exploded onto the scene that can create stunning artwork simply by being told—in plain English—what the user wants to create. This means that the AI not only understands the intent of the request but is able to make all kinds of artistic judgments that results in an image that is both logical and aesthetically pleasing. This can only be done if the AI has a true understanding of what is being asked of it—as well as the world at large—and image generation shows that AI has just such an understanding.

Intelligence vs. Consciousness vs. Sentience

All the news coverage of LaMDA and Greg Lemoine’s claims of it being sentient focused on the wrong thing. The first question we should ask about any artificial intelligence is not whether it is sentient but whether it is intelligent. That is a hard enough question to answer without getting into the far more nebulous terms of “consciousness“ or “sentience.”

As Ray Kurzweil wrote, “there is no objective test for subjective experience (i.e., consciousness) that does not have philosophical assumptions built into it.“

I think the same thing could be said about sentience, perhaps an even more nebulous term. Kurzweil further went on to state that the Turing test is not a test for consciousness.

If and when an AI proves to the majority of experimenters it is truly intelligent, then we can move on to the question of consciousness. As Kurzweil has said, when an AI gets smart enough to consider itself conscious, it will probably also be good enough at expressing itself to convince a majority of people that it is conscious.


How Will Conversational AI Change Society?

So assuming an AI proves itself “intelligent“ to most people‘s satisfaction, and leaving aside the question of consciousness, how will an “intelligent” computer change society?

The answer will depend on the level of its intelligence, for intelligence implies the capability to do things that computers are currently unable to do, and especially those things that are currently considered to be within the exclusive purview of humans.

Depending on the level of intelligence, I think the changes will be more profound than most people care to think about. This is why most media outlets are dodging the question of intelligence and focusing on the much more remote issue of sentience.

Some people believe that shortly after passing the Turing Test, artificial intelligence will advance at an exponential rate. After all, it will be able to read all of the millions of books on the Internet, and, with its perfect memory, it will remember everything it has read. So very quickly, it will become super-intelligent. That means in unimaginable ways it will eventually replace us.

It is also interesting to consider how AI will change society in the short term. I can see conversational intelligence becoming addictive. Lonely people will use it as a substitute for real-life friendship, and many people will consider this to be a public health crisis. Already, many people say the Replika iPhone app, which contains a poor AI that can in no way be compared with Davinci or LaMDA, is doing immeasurable damage to people‘s psyches.

Then there’s the loss of jobs to consider. Many white-collar jobs may be eliminated by AI. AIs may even be able to write articles like this one, putting human writers like me out of work. Already, Davinci can write poems and very short stories. Some AIs can even write short snippets of computer code!

Some people argue that if enough jobs are eliminated, the United States should switch to a system of socialism, but socialism has always led to tyranny throughout the world, no matter where it has been tried.

Professor Hugo de Garis believes that eventually people will merge with artificial intelligence, but this will lead to a major war when those who are not so enhanced perceive that transhumans, a term used by many futurists, threaten to take over all of civilization. Still others fear that AI may destroy all of us if it decides that we are a threat to the future of the earth.

Let us hope that only good, not evil, emerges from the advancement of artificial intelligence!

This article is accurate and true to the best of the author’s knowledge. Content is for informational or entertainment purposes only and does not substitute for personal counsel or professional advice in business, financial, legal, or technical matters.

Related Articles