AI, Clones, Deep Fakes, and Neural Networks

Welcome to Cornell Keynotes, everyone. I’m your host, Chris Wofford. Generative AI has taken the world by storm. For some of us, it’s top of mind.

Lots of us are using it. Some of us are just beginning to learn about it. And for some of us, it can be scary. Oftentimes, it’s amazing.

Speaking of amazing, today we have a special guest in studio, Lutz Finger. He knows AI inside and out. He’s had deep experience in this space. He’s built up Google Health.

He helped Snapchat go public and led AI teams at tech companies like LinkedIn and MarPi. He is an angel investor and a venture partner at Cherry. He has his own podcast, and he teaches Cornell’s flagship courses on AI and product. Welcome, Lutz Finger, to the studio.

Thanks for having me. It’s been a blast getting ready with you here. Viewers, a couple notes for you. We’re going to be taking some questions from you in the chat throughout.

But I should mention also that there’s going to be an interactive segment that’s coming pretty shortly here. So heads up for a QR code that we’re going to shoot up on the screen. So get your devices ready for that. Lutz, welcome again.

In your course, you often talk about your motivations for teaching AI. And I should mention that, by the way, let me back up a little bit. You’re the author of a certificate program in AI. And we’re going to be subtitling.

We’re going to be citing that throughout as well. So I should ask you, why do you do this? Why did you get into the education space? It has been a sidekick all along.

I’m actually at Cornell over 10 years now teaching. And because it is clear that AI is impacting us. AI is everywhere, right? There is a reasoned excitement about, wow, you can suddenly talk to your AI and it’s talking back to you.

But AI is there since a while. And we use it for businesses. We use it for our own productivity. We use it for a lot of things.

But it’s as well impacting us. I mean, you probably watched the presidential debate. And afterwards, Taylor Swift wrote that she is endorsing the Harris campaign. And in her explanation, she said, there was an AI of me.

That’s how far we are. We have now an AI of Taylor Swift. And Taylor Swift isn’t liking it. So what’s that thing AI of me?

Well, somebody created the deepfake. Like six days ago or so, eight days ago, I think, the U.S. State Department actually told Russia off for using AI to meddling in the U.S. election.

So AI is everywhere. And AI is impacting our lives. And I kind of came to Cornell. I wrote.

I wrote a book. And I started doing all those side activities next to my real job. Because I believe that we need to understand that AI thing. Because only if we understand it, we can deploy it ethically.

And only when we understand it, we can use it in a way that makes money for businesses. So my aim is, that’s the reason why I teach, explain AI to everyone. And it’s not brand new. We’ve been thinking about this for a little bit.

There is deep historical context. Perhaps more than I even thought so. Well, I mean, AI is not new. Like it started in the 1950s, 1956.

What did that look like? Well, like initially, they had this idea of like a black box doing something. And then they started with a checkers program. A computer that can play checkers and everybody get excited.

So we went through those waves. It’s called AI hype cycles. And then we have AI winters. But since lately, we had like only hype or excitement around it, right?

The whole idea of AI is essentially that you give a computer data. And that data contains tiny clues. Tiny clues that you and I might not see. But the computer with the data gets trained and figures out those tiny clues.

I give you an example. I built up Google Health with a bunch of very, very talented folks. And one of the things we did is we would look at an electronic health record. So a hospital would give it to us and we work for them as a supplier.

So we would look at the electronic health record and we would use AI to figure out tiny clues to predict, for example, whether somebody will end up in an emergency room in six months. So six months before something is happening, that later on everybody would say, gosh, if I would have just known it. We see the writing on the wall and the data. We find those tiny clues so that we can actually help people to prevent those, what is called preventable hospitalizations.

And it’s actually a huge part, like roughly one third of people in hospitals shouldn’t be there if we would have just known. Yeah. So that predictive component is absolutely mind-blowing, actually. But at the same time, for a lot of people there, and we think about it, we’ve seen this through history, minority report, et cetera, right?

Getting into the predictive component can be really scary for a lot of people. I’m naturally scared of it. Absolutely. Absolutely.

Right? If I tell you when you will split up, when you die, when you have a car crash, when you end up in hospital, this is all kind of like, honestly, this is uber natural. Well, it’s statistics, right? Yeah.

So, and now very often, in the traditional phase of AI, we used stuff like to predict when will a machine fail, right? Or like GE, for example, came up with predictive maintenance. Don’t check the airplane like in a regular cadence. Check it when it’s needed to improve safety, right?

Sure. Doesn’t sound scary. But now we have prediction on words. We can predict what is the next word in something.

In a Google search. In a Google search. This is type ahead, right? Yeah.

Or what is the next word in a sentence? And if we can predict the next word out, then we can actually start making sentences. For example, Chris, life is like a box of? Chocolates.

You predicted the next word. Right. Because it’s most likely that word, right? So now we have what we call generative AI.

Like it’s essentially technically wise. It’s as well a neural network. It’s nothing special. But by turning around a neural network, it’s a whole new world.

Yeah. I can have a conversation with a computer. Yeah. And that’s so amazing.

Well, it is just a prediction going forward. So what does this mean for business? Where is this taking the direction of business and trends? Yeah.

Have you looked at the stock market lately? Yeah. Let’s not talk about this. Okay.

Let’s not. No. I mean, like, look, I get excited about it. Like, Amazon said very clearly, we are excited about it and we are investing.

So is Facebook. So is Google. Everybody is investing a lot, spending shareholder money in this new technology. But we all are not extremely sure what will it actually bring.

So essentially, we have now this amazing technology out there. And in my work as a Cherry Venture partner, I see a lot of people. Trying to make business with this new technology. It’s a huge laboratory we’re in at the moment.

We’re all using it and trying to figure out how do we make business. And that’s where my course essentially comes in. Right? Because in the course, I say, well, the best way to not only use it ethical and responsible, this new technology, but as well, the best way to make business, good business decisions with that is to understand, what this technology can do, what this technology cannot do.

And then go industry by industry. We look at healthcare, we look at legal, we look at media, we look at banking and finance. And we go through the different industries and then discuss, where can we use it and how can we use it? So how, when you talk about your approach to education, I mean, you know, you had some inspiration.

Tell me about that. Yes, absolutely. Because. Well, like you talk about businesses, right?

And then we need to make it happening in those businesses. And it’s very often complicated, right? People like if I, if I talk to folks that kind of saying, oh my God, AI is so complicated. I don’t know, but it’s not, it’s not complicated.

It’s actually pretty simple. And all what I wanted to do is to make it understandable. And if I’m thinking about complex stuff made understandable, then John Oliver, pops into my mind. Yeah, sure.

I love his show. And so I approached Cornell with the idea, let’s make, let’s make AI as simple as John Oliver would explain it. Minus the language, presumably. Yes.

No swear words. Sorry for that. But simple, simple and fun. So to make it simple and fun, I created this Jarvis, like if you like Marvel, Jarvis, this device.

Okay. Yeah. Yeah. The voice from the off.

Yes. The AI computer who can talk to me. So it’s not only me talking about like statistics and machine learning and so on, which, I mean, let’s face it, nobody wants to listen to hours and hours of German kind of accent. You’re not bad.

You’re not bad. Thanks. But I, I want to make it fun. Therefore I have now this counterpart and I actually needed a voice for that counterpart.

Okay. So I approached a friend of mine and we can like, if you want, we can have a look at it, but I like much of my course is actually a conversation between me and that counterpart. And this is a big through line. This is a big theme.

This is a, this is a concrete substantial part of the courses. So I think it’s worth taking a look. Let’s check out. Let’s check out the video here.

This is from your course, working with your assistant, your copilot. Greetings. I am Cy, a sentient AI, specially designed to assist in this course. I can access and analyze data to enhance our discussions.

You’re something like a teaching assistant? Well, don’t degrade me here. Let’s say I am your partner. I will help you with the course.

Okay. Partner. And how do you work? I mean, where are you even?

Technically, I’m a transformer model. I saw that you will cover this later in the course. Happy to chime in here at that time. And your voice?

It sounds, let’s say, um, familiar. Maybe. I have adopted the voice of Kenneth Neil Cukier, an American journalist and author of books on technology and society. He is best known for his work at The Economist and the book Big Data.

But it’s really me and not him talking. He just gave me the voice. Okay? Get it.

Greetings. Greetings. So you say this is all computer generated? Computer generated.

Ascension AI. Did Kenneth know about what was going on here? I mean, how did the permissions work out? Explain everything.

Yes, absolutely. I hope you like Kenneth and my interaction with it. You did great. So actually, or let’s call him Cy.

But yes, this is Kenneth’s voice as Cy actually introduced Kenneth directly, right? So absolutely Kenneth knew. Now the fun fact is, how do you do this, right? We take a lot of data and then we train a model.

And I’m going to show you. We’re going into this like in the course. So actually at the end of the course, people can do this by themselves. Sure.

But we take data from Kenneth and then train a model on it. Now here comes the fun thing. Kenneth, and we should give him a shout out, he could have easily done the course with me. He is super knowledgeable in the AI world.

He is a technologist. He is an amazing author and he is an amazing journalist. So I would have loved to do this course with Kenneth. No.

He is a busy man and he didn’t have time. Therefore, I approached him and asked him whether I can actually copy his voice. I could have, just to give you the opposition, I could have done this without his endorsement, right? Because he has all the speech out there.

I just grabbed the speech and use a model and kind of could have replicated him. But that would have been unethical and not legal. So actually I approached him and asked him. And it worked.

And it was a lot of fun to work with him together. So how would you characterize the quality of this? You work in this space all the time. So I don’t know.

Any thoughts? Is this where you want it to be? It’s very interesting to see, right? There is a voice quality.

We first trained with something like 15 minutes of data. And then I asked him to read a whole hour for me. The voice becomes more like him. When we train it obviously with more data.

There is as well the business part which I want to do. I want to have fun with a nice sounding voice. And Kenneth was what I dreamt of, right? So therefore I actually asked him whether this is okay.

And he said yes. Now, if you want to see the quality, what we can do is I went to London and had a chat with him. And he said, yes. Sounds great.

sequence. Let’s roll the first one and see how we do. Should we have some form of natural right, some sort of property right? People could take anyone’s voice and use it and that might not be appropriate.

Okay, good. So that was clip number one, relatively short, but we’re going to hear the same thing. Clip number two. Let’s roll that one.

Audience, you let us know which one you think is the real one. Should we have some form of natural right, some sort of property right? People could take anyone’s voice and use it and that might not be appropriate. Okay, well, the way that it strikes me, you know, first of all, really good stuff.

Compelling. What’s really important and critical to our conversation here is what the audience thinks of it. So we’re going to check out some poll results and see how we do. I’m curious to see what people say.

I know. We’ll wait for that on the display there. Let’s talk about Kenneth. So has he seen the course?

Has he heard the course? Oh, he has heard the course. Yeah, Like I essentially, I first created a message to him where I used his voice without asking him. Yes.

And I created a message to his team wishing him, I think, a Merry Christmas. And I sent it over to him and saying, this is a little bit how it will work. Ask your wife whether she kind of recognized it. And his answer was, it sounded very much like me, but I would never wish, Merry Christmas in that way.

So she knew exactly it’s fake. Okay, good. Well, our audience chimes in. We’re at roughly 70%.

It just came down a little bit. Should we have some natural right? Most think that clip one is the correct one. And clip one it is.

So 70% of the order, 68 actually, got it right. And it’s interesting. Did you like, what did you think? I thought clip one was the one.

Why? Because his, just simplification. Simply put. Example two just didn’t feel as natural.

It didn’t have the ups and downs, maybe whatever sing-songiness exists within his voice. Did you just realize he stuttered? Like it didn’t, didn’t, he was thinking you could see the rubber burning, right? That’s right.

It’s hard to articulate. The human takes way more liberty in how we modulate things, how we think about things. So while I was talking to Kenneth in London in the studio about property, rights, he obviously not only talked, he as well thought he, he put a certain stress on certain ideas that he thought is more important. This is what we humans do, right?

Which makes us humans. So when people are always kind of, and we should have talked about the bigger picture is AI going to replace us? The answer is, and I go through this in my course many, many times. The answer is no, we are just getting the support.

And you see in that very property rights discussion, it’s, it’s a minor problem. It’s a minor piece, but you can tell how important modulation is. So important that 70% of the audience says, yeah, well, this is a human. But because we are stressing, can you make a computer yell?

Yes, you can. But somebody needs to take the decision at what point in time the computer should yell. So the whole emotions, the whole pint of like stressing different parts, all of this is human. And all of this makes us different.

And therefore, many people get a lot of stress. And so, I’m not saying that we should get now bored by all those computer generated images, right? It’s kind of like, because it’s missing a piece here as well. However, I would claim it doesn’t make a difference whether it’s a real one or not.

Because I had fun with Kenneth, with my fake Kenneth. I got the voice that I wanted. I wanted a partner to work with, to explain AI in simple terms, so that people like the course and people can relate to AI. And that’s how I put him in.

And Kenneth wouldn’t have time to do this for me. So I supercharged Kenneth’s voice to have that possibility. So I want to know, the techie within me wants to know how Kenneth was built. How do you do this?

And do your students learn this? How do you do it? Absolutely. It’s kind of simple, right?

Because if you know Kenneth, and you hear the voice, and it wouldn’t have told you, it would just sound like Kenneth, right? Voice sounds for us humans is actually way harder to describe. But let’s describe images. And I talk very often in the course or in my classes about it.

If you see the tiny edges of a cat ear behind the table or so, you will know that this is a cat. You just know. But you cannot explain why. Or I recently started to help my son learning.

Driving. And I would look at the road and saying, oh, careful, this car is switching lanes. Because I had this gut feel. Yeah.

Right? Like the car didn’t do anything. It didn’t set the signal. It was nothing.

But I was like, I think this is a car that looks like it’s switching lanes. Those tiny little clues. This looks like a cat ear. But we as humans have extremely hard time describe what is actually the rule.

So if you want to describe a cat, rule-based, you will… Sounds like a cat’s ass. A neural network makes many, many tiny little decisions, and it’s very simple. Like a neural network, it’s essentially a linear regression, which we all learned in school, put into an activation function, which could be a logistic regression, which many of us learned in school.

And if you put many of those on top of each other, then you actually start to learn those gut feel, and you put it into a neural network. That’s it. It’s actually pretty simple. So in my course, yes, we learn how to build Kenneth or SI, like sentient AI, as we call it in the course.

And we start with very simple concepts, linear regressions. And then we stack them in. It becomes a logistic regression. Then we do several of them.

Then we have a neural network. Then we turn it around, and then we have a voice like the one from Kenneth. So what is the coding expectation here? It seems, I don’t know, kind of burdensome.

There must be a lot going on. No. Well, I mean, I told you my motivation, right? Like my motivation is to bring AI to everybody.

If I say you have to be a hardcore coder, I’m losing quite a lot of my audience, right? I imagine, yeah. So it’s a no-code course because coding is just a language. And we just discussed that a language is a prediction of what the next word in a sentence.

Life is like a box off. You predict that it’s chocolate, right? So coding is nothing else. You say print.

Well, then if I talk Python, I need a bracket open. What I want to print, bracket closed. So it’s, again, a prediction, right? So using generative AI, we can now code.

So what we did, and we are the first ever doing this, is we built in this Psy person as a co-pilot. So Psy enables the students to talk about business cases with the AI. But Psy. As well, enables the students to code without needing to code.

So my students can say, okay, I want to build a neural network with this data. And it should have the following characteristic. And we talk a little bit about the technology, about what characteristics there should be. And then the co-pilot or the teaching assistant, Psy, is actually coding this for the students.

So my students can code if they want. But many of them actually don’t know how to code. And they don’t know how to code. They just use the integrated OpenAI.

We’re the first course that integrated OpenAI completely into it. So again, no coding. No coding. In a course in AI.

And it’s, yes, no coding in a course in AI. Like, this is it. Yeah. You built the first, so this is, you know, effectively the first online course with a speaking co-pilot.

Yes. Tell us what the experience feels like. Because I want to know operationally what is it like within the course. So one of the things I am, so it’s fun, right?

It’s fun. It’s like, it’s for, for students is actually amazing. They kind of, they, learn. And I should publish about this because we started to look at the students.

They go into the course and they realize that they have this partner, this human partner. You saw me in the video where I said, okay, partner. This is essentially the experience of the students. They go in and they start to actually.

And they engage directly with an interface that feels very human. And they can make any part of data analytics or business discussion or coding with that interface. And that interface gets more and more smarter as we progress through the course. So do the students.

So their relationship with that AI actually changes. It’s very funny. So the initial course, which we ran, people saying like, I don’t. I don’t know what to ask, or they actually, they ask too much, right?

I had, I had students that would copy the whole question list they should work on and just put it into the interface. And then the site was like, wow, that’s too much. And shouldn’t you tell me, right? And so there is a learning experience, how we engage with AI, which is a huge part of this course.

UX is so important or user interface is so important. Now, the other thing is we should ask is. What’s the business impact, right? I was just going to ask you.

Yes, because very often we see people getting excited about AI, but it’s a tool. It’s just a tool. Like the question is what is the underlying business idea? Well, Kenneth, the voice underlying business idea, make the course like John Oliver’s shows.

And John Oliver, if you watch it, let me know. You get a free sample of this course, right? So like make this course simple. Then having SAI integrated into the course to talk was the business idea.

Allow everybody, even when you don’t have coding knowledge to take the course, right? So every time I use AI, the question is how do I use this best? And here it’s make it simple and make it fun. Now.

You will laugh. I asked at the end of the course and I teach this throughout the course. At the end of the course, I clone myself. How?

Simple, right? Well, I don’t know how, but why? Why? Well, how is I take data from myself, right?

Like a picture here, Okay. And then I try and model and boom, I have it, right? Yeah. That’s how.

Why? You know what, Chris, you’re an awesome guy and I really love working with you. But whenever I want to see your life, I have to fly nine hours. Okay.

So to come into studio, to come to Ithaca, it’s an amazing place, by the way. Come to Ithaca. It’s a great place. Yeah.

But to come to Ithaca takes me from the San Francisco Bay Area nine hours and that’s a full day. So Cornell and I, we had this discussion about, look, we are making a course. And the first thing one of the deans said to me is like, let’s put so much. We’re looking at how much is happening.

How do you keep your content up to date? Right. Because like, face it, like many of the things we run into issues with ChatGPT 3.0, 3.5, 4. All of this is changing, right?

Initially, ChatGPT couldn’t get math questions straight. Now I’m using it for my students for analytics. So how do we keep the course up to date? That’s a reason why I cloned myself.

Because now. We promise. the course up to date every month. How?

Well, I sit at home and I just write down the next lecture and boom, you have a Lutz teaching on the course and keeping it up to date. You know, and to see it practically playing out, there’s actually tremendous value in that. Sure, it’s fun, right? But witnessing the state of the art, seeing the improvements that are made, which happen literally from week to week.

Gigantic improvements. So that’s got to be a fun part for you too. Like you don’t know where this is going. No, but I can promise to my students that like do the course and you will be up to date with the latest and greatest because I can now use a virtual Lutz, which I can invoke like 24-7 to explain yet another piece.

And that’s crucial because Yeah. Because why do I use AI? Not because I want to see myself double. No, I use AI to scale up education, to change education.

And we have now done three things which are pretty amazing. So first, we make the course fun by having a Kenneth or a fake Kenneth, a Psy person who can interact with me. Second, we scale up the students by allowing them to work on programming without need to program. And third, we scale up me as a professor so that I can keep the course up to date.

So what do you look like? Not you. Well, I look like me. Like, would be bad, right?

Well, yeah. The question is, can we handle two of you? So Should I show you a duck? Like, I think we have a video.

No, we should give a fair warning to the audience because there is a video which we could link actually in the show notes too. But I am in a video. I copied in this case only my liking. So the physical likeness.

My physical liking. So my face and how I move. Let’s look at this. Okay, good.

Let’s take a look at this as it rolls. So the audience is not going to hear the audio. I should just remind everybody that there’s going to be a URL that’s dropped in the chat. So you can watch the whole, the whole full length video.

So here we have you with a clone. Who is the clone here? Right or left? I think most of us are going to say that the guy on the left is the clone.

Yes, absolutely. And by the way, you can see that I wear actually the same sweater I’m wearing now. So it’s that is not replicated in the clone. Oh, it sure isn’t.

So, so if we can articulate the differences, again, you talked about that kind of interesting phenomenon, we need very little visual data to either identify a person that’s far away down the street or a cat’s ear. It’s shocking. You know, you can tell by someone’s walk within a quarter of a second. This is similar to I know you well enough to know that you don’t move around.

like that. You mostly do. Right. But there’s something that’s a little bit off.

Well, I stare at you here. Like the clone stares at you. Now, but it is, does it look like me? Yeah.

Actually, right. I should also say that when I first saw this video, this is a couple weeks ago or something. That’s beautiful. And let’s not forget that this is actually really incredible stuff.

If you think about, if you think about the rendering of images, six months ago, I’m thinking about Dali, early faces. Right. Yes. Yeah.

You know, you were, you could never do that. Yes. Now, again, is it the real looks? Or is it close enough?

Actually, nobody cares. It could be I could actually replace myself with a duck, as long as I manage to keep the course up to date, because that’s my business, which I want. So throughout my course, I always bring it back to the business. Because, and we had this very often at Google, we would do very complicated.

Algorithms, deep learning algorithms. But what counts at the end is, how do you actually make money with it? How do you use it? How do you use it in a business sense?

How do you use it ethical? How do you deploy it? That’s what comes down. You know, and a lot of that’s been figured out in real time.

So let’s talk about that video in just a moment. I want to understand how that was made. We talked about the Kenneth process. Yes.

Similar here? Similar here. It’s good that you asked, because I introduced very… Very early on in my course, a framework of how to think about building those things.

So first, as I said, it’s a business, then you need data. So I needed to scan myself. And then I trained the model. And then I deployed and I can actually use it in order to keep the course up to date.

Yes. One thing that I think is critical to point out here, and we talked about this a little bit before the show, you used your likeness. You fed that into the AI, so to speak, right? Right.

The sound of your voice. There was a good enough data set to pull that off. But the AI doesn’t know what you don’t know, right? And so…

Absolutely. That’s the part, right? You’re literally authoring what it doesn’t… in kind of a real time in order to keep up with the timeliness of the course.

Yes. That is… That’s incredible. No, but that’s a point.

Yeah, yeah. Let me restate this. The AI does not know what I know. Mm-hmm.

When it comes to something like… No, sorry. Like you said, the AI does not know what you don’t know, meaning when it comes to new things, the AI doesn’t know. Yes.

But we should point out when it comes to existing things, the AI knows what I know. Because beyond liking and voice, I as well train the AI with all of my knowledge. Okay, right. So, exactly.

But it doesn’t know what you’re going to do for tomorrow’s syllabus for your course. Exactly. Exactly. And so when people talk about…

Replacement, right? There is this big fear, and we will see a huge shift in our industry, but in terms of how people engage and how people work. But the innovative thought process, the new things, what I’m going to do tomorrow, the AI doesn’t know. But I did train, and we should ask why, but I did train the AI with all the past knowledge.

Why? Okay. Because I know that my students can interact with me, the professor, 24-7, right? On a one-on-one…

On a one-on-one basis. Office hours. Right? Yeah.

I mean, think about this. This is now… We talk a lot of interesting stuff here in our podcast, right? Yeah.

And if somebody wants to know something, we could say, well, just listen to the podcast, right? Yeah. But you don’t have time to always listen to the podcast. Yeah.

So, can you not just… Can you not just tell me? Well, yes, I could, but I don’t have time neither. So, that is where I’m scaling myself up.

And to all of the teachers and professors out there who are listening in, you know what I’m talking about. You have the syllabus, a three-page document. Your students could have read it. They didn’t.

But they do ask you questions, and you do answer. So, AI plays a huge role in scaling us up in areas where… Where we have knowledge that needs to be replicated. I consider myself an audience advocate when we do these.

I’m just going to dig in and say, still feels creepy. Tell me why this is the way to go. Yes. Or why it’s not creepy to begin with.

No, it’s… Like, look, your feelings are your feelings. It does feel creepy, right? Yeah.

We have this… There is a terminology, actually, for it, which is called the uncanny valley. Yes. The uncanny valley is the idea of that it becomes more and more closer to humans.

So, when you have these robots from Boston Dynamics, and they kind of look very clunky and fall over and jump around, then you’re kind of saying, oh, weird robot. Then the robot becomes more cute. It suddenly gets eyes and kind of start dancing. Then you’re like, oh, that is cute.

That’s still… Still, but still clear that this is a difference between humans and non-humans. And now, you have robots which are like you look like. That’s creepy.

So, it is maybe creepy that I copied myself. But the whole purpose here is… And actually, to overcome this, I should say this. In the course, I actually have a different voice.

So, my liking looks like it, to make the point that I can copy. But I keep a different voice. Just to make this… From the appearance clear to the audience.

But… So, there is a creepiness. I have a… Or something which we humans don’t like.

I have a very, very good example from healthcare. It’s my favorite example in healthcare. It’s actually… We gave doctors a write-up of an x-ray.

And that’s a write-up about, okay, this is what you see here. And, dear doctor… Chris, I want your second opinion. Okay?

And we do this to many doctors. And the only difference is one write-up says… It’s the same write-up. One write-up says, this is done by AI.

The other write-up says, this is done by Dr. Lutz, for example. Right? And you can see that the doctors who get the write-up of AI say, can’t be true.

No, no. Like, you double-check this. I don’t trust this. Because we have this relationship problem currently.

And that’s the reason why you feel it’s scary. But we were there before, right? Google looks up content. Now, an AI looks up content.

And I believe that optimizing workflows by using AI agents is what many people who have content, who think about content marketing, are going to do. If you are a brand, or if you are an influencer, if you are a professor, you have created content. I mean, like, heck, I came from San Francisco to see you. Why?

Because we have created this amazing course. And I want to talk about it. So, and I spend time now to talk about it. But if somebody in the audience wants to know something, I mean, reach out to me.

But most people will say, well, that’s a little bit too much. Can I just write and get an answer? Yes, you could get an answer from Lutz Finger. So, I actually, on my non-Cornell part, I build a company called R2Decide.

We are an AI agent platform. So, we take content from content creators and make sure that you can engage with it. So, I tell you, one of our customers is called Decent. It’s a coffee machine.

If you know my course, then you see I do a lot in coffee. Coffee is an AI topic. So, the Decent espresso machine has 10 sensors and measures pressure, temperature, flow rates, and so on and so forth. And because it’s a complicated technical topic, there is a diehard community of people talking about coffee, what is a good coffee, how to make the coffee, and so on and so forth.

Diehard community, a lot of people, 5,000 members. Now, people come in and they will ask, how do I make good milk foam? That’s a typical question. Sure.

And if you are somebody who has done like excellent coffee, it’s like, well, that is not a question I’m going to answer. Which would mean the community would kind of like go down, right? Yeah. And you use community manager who points you to the article or to the discussion.

That’s manual work. So, we build an AI agent that automates this work. The community loves this, right? Because now you come in, you ask a question, and I will say, oh, the AI will say, oh, that question was actually answered by Chris.

Here’s Chris. And which is cool because now you replicate the knowledge. And give the answer. Yeah.

You get the knowledge back as the agent. But, and this is the non-creepy part, you actually attribute it back to Chris, the outside of Uncanny Valley. You’re not saying, oh, I know how to make a milk foam. No, you say, you know what?

Chris talked about milk foam. I show you what he said, summary, and the link to it. Which is a beautiful thing, right? So, essentially, what you’re doing is you’re making prior knowledge available instantly, strengthening the brand, strengthening the quality of discussion within the thread.

Yes. Everybody wins. No, we, if you talk to influencer, they, influencers are like brands and influencers. And you have that discussion, what we just brought up here.

Yeah. Which is super important for how to make AI useful in there. The influencer will say, hmm, so there is an AI which knows all my knowledge and can talk about it. What’s my value as an influencer?

Right? Yeah. It’s the same thing. Yeah.

So, I started this course and I approached Cornell with, oh, I want to replicate me so that I can create any lesson from my home with a couple of keystrokes. The first reaction was like, oh, what are other teachers going to say about it? Because healthcare is changing. So, influencers are more scared because they actually think about the uncanny valley more.

They say, like, it’s my brand and you’re copying my brand. Brands, on the other hand, saying, oh, this is so awesome. I have all this content and I have normally to have a community manager helping to get all the content which I created out again. And now an AI can do this 24-7, giving acknowledgement to my community, showing my questions to the brand over and over again.

This is awesome. This goes at scale. The scale, like, strengthens my relationship. It’s very funny because this is a UI and UX problem.

It’s a challenge. How we interact. But influencers will get around. They will start learning that for them this is super helpful to scale up because how cool, you’re an influencer and you can talk 24-7.

How cool for a professor who now can talk 24-7 about their content. And you think about the marketer who’s done what is historically referred to as like kind of the golden egg, getting the right message to the right person at the right time. Yes. That actually just happened.

Yes, totally. In the example you were just talking about. Yes. But creepy, it still could be.

Marketing always pushes the edge of creepy in some particular way, especially as it relates to technology. So where’s the dark side of marketing here? Well, it’s actually not marketing. The dark side is when you use technology in something which is not legal or when you abuse technology, right?

That’s the dark side. We know that cyberbullying is illegal. Well, if you get, if I can teach to my students very simple non-coding, how to make clones or AI agents or copy voices, then anybody can do this, right? So there are now tools on your apps which allows you to do this.

But if you do, if you take somebody’s liking and put it in a tool, which unfortunately happens pornography context, that’s illegal. And that gets banned. But that gets creepy. Or let’s talk about what we initially started with.

Taylor Swift. Yeah. Taylor Swift says, there was an AI of me. Well, technically, it was actually a very bad, like technically it was very bad created, but it was a fake image of her endorsing Trump.

She did not endorse Trump. No, she now endorses very clearly the Harris campaign. So it became easier for us. Yes.

Yes. To create those fakes. And that is actually an issue because we can now create fake news on a scale, right? Yeah.

But it’s nothing new. So actually, with my old company, Fisher Analytics, I sold this a while back. We worked for governments and NGOs. And we helped to analyze context.

So my fame of claim here was. I worked for Commissioner Redding, like the big commissioner in the EU commission. And the Snowden scandal hit. And we were.

And she went to Obama. And she asked me, let’s tell me about how is Europe thinking about privacy rights in the wake of the Snowden scandal. So we analyzed all the different content and gave her a report. Now, the tricky part is already at that time.

You had a lot of fake accounts driving fake messages. And that’s called astroturfing. So astroturf is a brand name for fake grass. The idea is creating a fake grassroot movement.

So there are people pushing, for example, like how people react to the Snowden scandal. Or like a more funny example is as Microsoft had their antitrust case, they probably engaged some agencies who pushed them. So a lot of people wrote letters to senators. And some of those people who wrote those letters were dead.

Okay. Creepy. The L.A. Times at that time titled, even the dead are fed up with the antitrust case.

Yeah. So where it really, I think you just had kind of pointed to it a little bit. Then we get into the worst cases, right? State actors playing a part of this.

Spy agencies. Yes. Intelligence organizations launching campaigns on Twitter, etc. So a lot of legal and ethical things to consider there.

But can you tell me about that? How you think about this? Yes, absolutely. All right.

Because this is a while back that I had my own Twitter account. I created a couple of Twitter bots reciting the German government. And then people followed me. I had over 10,000 followers on a complete fake account, right?

Oh, yeah. I know. And so it’s so simple. It used to be so simple.

I talked about 2016 already about this. And now we have way better technology. Make fake pictures of the Pope. Make fake pictures of Tyler Swift.

Create fake information. Even talk like a person, right? All of this can be scaled up. So as a state actor, you can use this.

And Donald Trump talked 2016. Like he has this thing about this crowd. So he talked about his inauguration crowd size in 2016. And he got shown pictures.

And one of the people from the White House said, oh, these are alternative facts, right? So there is an idea of having alternative facts around, which is terrible because now state actors can go in and push on it. If I’m a state actor like Russia, I could easily imagine to create, I take any wedge issue like guns, abortion, whatever. And now I create two groups, one group actively debating pro and one group actively debating con.

And it will sound very natural because, again, it’s a prediction. Life is like a box of chocolate. So I can do this for abortion as well or for gun control. And now I push them onto social media and have them fight online.

And the whole country thinks we are in a war. They think we are divided. And people are concerned. And the state actor that initiated all of this can kind of lean back and saying, I knew it.

The US is a mess, right? This is a problem. Yeah. You know, the ethical and legal figures into everything you do as an educator.

I should mention that we’re going to be doing a, you and I are going to be meeting again October 3rd, coming right up. We’re pivoting. No, this is actually like October 3rd. This is a way bigger topic because actually in the course, sorry for jumping in.

Go ahead. But in the course, when we talk about legal or banking or healthcare or coffee, like whatever we do, there is an ethical component to it. So it is, you cannot divide this away from any AI discussion. On art, on ethics.

Yeah. So we’re going to be tackling this on October 3rd. AI agents and the implication on art and ethics should be kind of cool. So we’re pivoting from our previous conversation.

We’re going to be talking about the concept because I think this is the one we landed on and wanted to build upon that. So, you know, aside from taking a certificate program, educating yourself at Cornell University, if you don’t have access or whatever, how else do people learn about this? How do you keep sharp? Yes.

How do you keep sharp? I think the way for us to keep sharp is, or the way to understand, it’s not so much about how to keep sharp. Because if I’m… I think you need to understand.

I think you need to learn what the technology can and cannot do. And we actually learned this. Let’s talk about this alternative fact, right? You don’t trust everything which is happening online.

We had just two days ago, we had the presidential debate and the fact checks happened and people said, like there was one candidate which gave a lot of wrong facts. And I think many people during the debate already said, I know, I don’t think it’s right. So we learned by being exposed. To wrong facts that fake or alternative news is a fact.

We also learned, we just went through a pandemic and there was a time where the CDC said masks don’t help. And Twitter and Google were supposed to suppress that information, right? We now know that masks do help. So there is this alternative fact information is a double-sided sword because it’s a double-sided sword.

There is no absolute truth in some discussions. In other discussions, the world is like the earth is round. There’s an absolute truth. In a discussion like, for example, how much protein should you take?

There is no, there is a lot of different opinions. Like World Health Organization says at least 0.8 gram per pound. And bodybuilder will tell you two grams per pound, right? So it depends on your context.

It depends on a setup. And here it becomes complicated. Because when you go to Google, it will show you 20 blue links and you decide. And it will list those 20 blue links who is clicking on it most.

If you go to ChatGPT, it will only give you one answer. And it will give you the answer where they got trained most on it. I did the following experiments and I showed this in LinkedIn. I tried to convince ChatGPT that I’m Captain America.

So I uploaded a lot of documents. Which I used ChatGPT to write. But a lot of documents about claiming that I’m Captain America. And then I asked ChatGPT, who am I?

Now that poses a problem. Because now ChatGPT says, well, there are a lot of documents saying Lutz is Captain America. I’m trained in my weights. And like how the neural network is.

That Lutz is not Captain America. But Steve Rogers. So how do I combine those? Somebody needs to take a decision.

At the moment, the decision is done by a product manager in OpenAI or Google or Meta or wherever they live. And my question is obviously, who has the control here? Are we supposed to trust one product manager with this decision? By the way, fun fact in this case, OpenAI answered correctly.

It said, you’re Lutz Finger, but you make me believe that you’re Captain America. Which I thought was a very funny answer. You didn’t use a burner account for that one. No, not at all.

Okay. But the problem lies here that we will live in a world with alternative facts. We will live in a world which is not black and white anymore. Where there is not one opinion.

And I call this branded content. So that’s the reason why I built R2Decide as a platform where I allow people who have branded content to answer over this content. So take your fitness trainer who has an opinion about how much protein you should take. Now that fitness trainer can now have a bot or an agent that talks in his or her name 24-7 and explains their liking.

So we will get into a world where brands are becoming big again. And now all the media organizations, you lost it as the internet came around. But now New York Times economists listen up. You have probably a huge blast in front of you.

Because we are getting to a world where somebody has to decide what is not 20 blue links where the user decides. No, we’re getting into a world where we need one answer. And that one answer is probably not given by ChatGPT’s decision. But probably by a decision which talks like ChatGPT but is branded New York Times or it’s branded economists.

Because that’s the person we trust. If you want to know something about AI, Chris. You come to Lutz. You’re not just asking anybody.

You would ask me. I guess. Well, let me ask you this. So, you know, any way you slice it.

Neural networks are the foundation, right? Yes. Let’s close the loop here. So explain how neural networks relate to what you’re just talking about.

Everything which we are talking is ingrained in neural networks. Identify a cat is a neural network. Talk like Kenneth is a neural network. Decide that an individual is a human.

That an airplane engine needs maintenance is a neural network. Figure out when somebody ends up in a hospital is a neural network. Generative AI as well as a like traditional AI are empowered by many different logistic aggressions on top of each other that we essentially call a neural network. So very often people compare neural networks with a brain.

I do in the course as well. But I think there’s even a better analogy. And I have a segment in the course. Do you want to play that?

Yeah, let’s check that out. So this is the segment on neural networks. This is you hanging out on Cornell campus on one of the most beautiful days ever. Here we go.

That those neurons together form a kind of a society. We are all connected. We are all connected in one structure optimizing a loss function. In this case, it could be something like the sum of squared errors.

So is society. Society is a system. Society is connected. Many different agents called humans.

We are connected all by a shared objective. In this case, financial optimization, financial growth or wealth. And as much as each neuron in itself is not very powerful, but the network becomes very powerful, so are we as society. If we work together, we can be very powerful.

We see networks sometimes go wrong. We create rule-based logics to steer them. We do this for society. We call this loss.

But we know the more laws we do, the more symbolic logic we put on this neural network, the less effective it becomes, which is true as well for societies. So in a sense, neural networks can teach us quite a lot. Work together peacefully because a network is better if it is together. Speaking of together.

So we kind of have a problem. So we covered a lot, you and I, today. And in some ways, it’s been nice to kind of package it. But I feel like we’re just getting started, to tell you the truth.

There’s not a clear path forward. But there’s this feeling that innovation will continue and probably accelerate at a rate that we’ve never seen before. What is the next step for you? You know, you’ve created some of these courses.

What do you want to talk about or think about most? What gets you jazzed these days? As it relates to AI. What are you into at this moment?

Well, I’m obviously jazzed for the company I just created. We are seeing traction. This is this whole idea of branded information. The whole idea of helping brands, influencers to scale them up as much as I scaled myself up as a professor.

Companies called are to decide. You can check it out. You can build your own clone. So if you want to build your own clone.

You can build agents on top of your own videos. Should I do this? I actually could build an AI agent for all of E. Cornell keynotes.

And then you have all the people and you can talk about this. Let me do this. I sent this over to you. You check it out.

But that excites me the most at the moment. Yeah. So I’m thinking about our audience too. Again, as the audience advocate, right?

I mean, we have varying degrees of sophistication within people who are viewing this. Yes. So education is clearly a natural path forward. But for those who are still feeling trepidatious.

Yeah. Okay. Easy. The message I keep hearing is fear not.

Keep moving forward. We don’t know what it’s going to look like. What is your message? Like fear not.

AI is here to serve us. AI is here to stay. It’s not going away. Think about how you can use AI in your daily lives to scale up your life, to make it more effective.

In your businesses to scale up, right? So there are many different use cases. And I asked in the course, I actually like publish like a playbook how you can go through your business to figure out how is the best applicability of AI. Now, if you’re still scared and think, oh, a no-code course talking about AI is not mine, which like it’s really simple.

Trust me. And it’s really fun. However, if you set like still a little bit careful. What you can do is.

eCornell publish a lot of small pieces from my course, from other AI courses eCornell has. We have amazing courses here. And those snippets will help you to understand that this is just a technology just like Google was one, just like the mobile phone was one, just like the typewriter at some point in time was one, right? So embrace the technology.

Make it useful. But know that also as we had in the mobile phones, there was abuse. And we started to learn to deal with it. So learn how to use it.

Because once you learn how to use it, the best thing you can do to us as society is help us to not abuse it. Keep like keep guardrails for others to not abuse it. And that’s a discussion we as society need to have. And I’m actually super like stoked that I’m coming back to you in a month’s time.

Because at that point, we really should talk about the guardrails. And it’s like what does it mean for Hollywood if you can like recreate any actor? What does it mean for laws? What does it mean for regulatory processes?

There is so much happening currently. Where we have a lot of stupid activities happening. We have a lot of stupid activities happening. We have a lot of stupid activities as humans.

And we can scale them up. We have a lot of bad activities as humans. And we can scale them up as well. Now, we shouldn’t.

There are laws about it. And we should because we want to make the world a better place. And even if you watch us now and heard about it, AI is easy. AI is there for all of us.

AI can be abused. And we need to learn how to not abuse it. And we have not yet figured out how to make the most value out of AI. But a lot of things are happening.

Join that movement. Join the effort. Let’s finger lovely and heavy in the studio. Thanks for coming in today.

Thank you viewers for hanging with us through the hour. We’ve just run out of time. Sorry to get to your questions. Note that this event is live right now.

The recording will be available in just a few minutes. And may I? If you have questions, just post them on the LinkedIn page. And I’m going to answer them.

Oh, there you go. Okay, good. Maybe with an AI. All right.

Thanks, everybody. See you viewers next time. Bye