Episode 7: "Interview with Lori Mazor, CEO of Synthetivity"
Listen to this episode:
Episode description:
In this episode I talk with Lori Mazor, CEO of Synthetivity, a company that develops AI and metaverse applications as well as providing design, consulting, and workshop services in the space. We discuss chatbot personas, ethics, prompting, bias in Midjourney, and more.
10% off of Synthetivity workshops, use code: QUICKBITS10
https://www.synthetivity.net/
Episode script:
SCOT: Hello, everyone and welcome back to AI Quick Bits: Snackable Artificial Intelligence Content for Everyone. My name is Scot and today I'm going to be speaking with Lori Mazor.
Lori is the CEO and founder of Synthetivity, a company that develops AI and metaverse applications as well as providing design consulting and workshop services in the space. Lori's projects range from physical buildings to technology products and services. She regularly guests lectures on design, strategy and operations. She is a multimedia artist whose creative work has been published in The New York Times, Inside Higher Education, Crane's New York Business and The Wall Street Journal.
Lori is a recipient of the American Institute of Architects’ Award for Design Excellence and Crane's New York Business named her one of its “40 Under 40 Rising Stars.”
Lori, thank you so much for speaking with me today. How's it going?
LORI: It's going really well.
SCOT: I have several things I'd like to discuss with you today, but I would like to start with a recent newsletter post you wrote in your newsletter, “Humans of AI,” about chatbot personas and I believe in the post you experimented with ChatGPT, speaking with you as a benevolent cheerleader, a tough love coach, a difficult parent , and even the bully. Would you talk to me about why you wanted to experiment with different tones from ChatGPT?
LORI: Yeah, absolutely. I mean, I think my premise is that I am a teacher first and foremost and, you know, I am trying to teach people from a place of learning myself and sharing in real time the things that I'm doing and the explorations I'm doing out of curiosity and what lessons I'm learning.
And I think, you know, some of the lessons are the amazing creativity that the generative AI applications enable and some of them are the scary aspects of it, things that we might want to think about and be more cautious of as we think about rolling out these applications in our businesses and in our schools, and in our personal lives.
SCOT: I totally agree, especially something like a bullying tone, I believe OpenAI [ChatGPT company] is pretty good about putting disclaimers or guardrails. When you were, in your post with the bully persona, it did say like, hey, you know, this isn't really something I do. But it was pretty easy for you to say, you know, give me an example of something that we shouldn't do, or something like that. And it did give you a bit of a tone of a bully.
And I enjoyed your conclusion about needing an ethics framework for personas and the transparency into the establishment of the base persona when someone starts to just use the product. Like the setup, give the user power to craft their chatbot to kind of their preferred style. I thought those were good suggestions.
LORI: Yeah. And I think, you know, one of the things that I learned over the process of teaching is, you know, I incorporated the original GPT-3 model into a teaching and learning application for high school students. And when I did that, I had a developer basically kind of write a program that scripted the bot in terms of what the bot would be able to do functionally, how it would engage with students, what tone of voice? And then literally, he gave it a personality. And so I think I had that experience before ChatGPT really became popular. So I can kind of see under the hood and I know that when I am communicating with a machine that that machine is basically a statistical machine, right?
I mean, we are talking about a technology that is about predictive language. So all it's doing is looking at how we've used language historically and then predicting what the most likely word is to come next after a word or a phrase that you've put in. And so, I think that's really, it's really important to recognize what it is and what it's not. And that, you know, humans have then given the bot personality, a style of engaging with you, and the benevolent cheerleader is actually the natural style of GPT-4, for 3.5 and 4, which is, you know, the ChatGPT that you get when you sign up either for free or pay the subscription. But, you know, one of the most powerful ways to use it is actually to tell it first who you want it to be in order to speak in that voice. And, I experimented with a bunch of good bots and then I also tried experimenting with some not so nice personalities.
What was really interesting was I put a poll up on LinkedIn the day after I did the newsletter asking people what kind of voice, what kind of coach, motivates you to do your best work and no one picked the bully. And, and actually I kind of phrased it as the disgruntled customer. Because, you know, I think we all receive negative feedback either from our customers or our colleagues or sometimes our bosses.
And, one of the ways that you could then use the chatbot in a positive way is to take that negative feedback and say, you know what, can you reframe this for me in a positive way that would motivate me, right? And so, I think a lot of the ways that I use this is kind of as another voice in my own head. And sometimes it's good to, you know, write and to speak aloud the things that we want to hear that are gonna inspire us or motivate us to do good things.
SCOT: Yeah, I think it's interesting that there's so much, even though it is a sequential model with predictive text and whatnot, it's so good that if there's enough out there about someone, people are experimenting with like, pretend you're Steve Jobs or something like that and speak to me in that type of mentorship type of way. And, of course, it's going to be based on the past, not what you know, a current alive Steve Jobs would say. But I think it's pretty remarkable how real it feels or how alive it does feel versus, if you were to say Shakespeare or Socrates, maybe it has a little less to draw upon.
I would also just say that humans, even without conversation will anthropomorphize. I mean, a lot of people with their car, you know, they give it a name, that sort of thing. So, we're way past, I mean, I think it's a given that people will feel like these machines are alive. Many people, even if they do know that they're not, there's sort of that like that just natural human tendency. And, I do feel like because of that you're right on to point out the need for an ethics framework.
LORI: Yeah. And, you know, I have a 15 year old daughter and so, I knew that Snapchat was an application that she uses with her friends and I knew they had just come out with an AI component. And, you know, I think there's a lot of people who are concerned and, and so I had that conversation with her. I'm like, are you using this? She's like, no, not yet, but I just saw it and I'm like, all right, you need to give me feedback. Let's do a controlled experiment here. I want to see how you and your friends are gravitating towards this and we're going to have a conversation about it and then we'll make a decision whether this is something that's good or bad.
So, we never say absolute “no”s in our house. It's more like, all right, let's experiment. Let's be intelligent about it. But I also was really surprised in teaching my first class to high school students and, you know, they are so savvy and I think, in many ways, they've grown up in a world where they know that they can't trust anything that's on the internet. And so they've already established really good habits about double and triple checking references and kind of living in this post-truth world that we're in.
And I think, one which then AI really exacerbates both on the image side with the ability for anybody to develop a deep fake, to the chat side where you do anthropomorphize the machine. But I'm hopeful about the next generation and, I think they've got a good lens to look at this with and they're a little bit more skeptical than I think those of us who've kind of grown up thinking that we can trust the world.
SCOT: Yeah. It's funny you mention that, I have a teenage son as well and a soon to be teenage daughter. And, for a while there, when I was introducing them to devices when they were younger, like, ok, here's your computer here, you know, you have a phone for the first time and you're sort of worried like that they're going to get phished or click on something for some spam. And now I'm kind of like, oh, my son, he knows that's spam. I don't have to worry , you're right, their lens is completely different. They can sniff out that sort of stuff.
OK, I'd like to move on a little bit here. You have something that you do called Midjourney Mondays, sort of like a weekly blog where you create with Midjourney, you, you have images created. And one of your recent posts I thought was just stunning. These, these dancing images that sort of look like I would say like if I went to a dance performance at a performing arts center, the costumes are are wild. They're just really interesting images. As much as you're comfortable with, could you take me through or talk to me about the creative process you use with Midjourney?
LORI: Yeah, this to me has really been magic and, I was a corporate executive, I was running operations for The City University of New York throughout the pandemic, dealing with serious issues and managing a large staff. And when I left that job, I decided to take a break and take a sabbatical and it just happened to coincide with Midjourney being launched and, you know, I was trained as an architect. I had also been in the performing arts, I'd been a dancer my whole life until I was in my mid twenties.
And so, I hadn't done anything creative like that in so long that it was amazing to me that I could use language, which is a skill that I have been perfecting over the years in terms of communicating with clients and customers. And, those skills now could conjure up a physical image, right? And I wasn't limited to being an architect and designing buildings. I could design fashion and I could design, you know, I could do choreography. And so for me, it's just given me this playground that I can explore ideas that I'm really interested in. I can literally do it sitting in front of the television with my family and, you know, just like writing prompts and seeing them um come up on the screen. I think the dancing images were probably like 13 or 14 weeks into my process. I started with the idea of folded paper buildings and I just kept like pushing that to the point at which it started to become clothing. And then Midjourney also kept evolving. And so it got better and better at generating realistic looking people. So I could start putting clothing on bodies.
Then I started realizing that, you know, the bodies that came out of the box were all white and all thin. I really wanted more diversity and I started to recognize the problems with the models the way that they'd been trained. You know, in a bit of disillusionment. So, I kind of went through the typical Gartner hype cycle with Midjourney, which is like, it's amazing and then it's so terrible. But I think through that process, I kind of figured out ways of creating my own code of ethics for how I engage with the application.
SCOT: Yeah, that was going to be my next thing I was going to ask you about, keep going, please.
LORI: Yeah, I mean, there's a lot of bad stuff that has already happened, right? In terms of workers who were exploited in order to, you know, be digital factory workers around the globe. In terms of artists whose work was used without their permission and, you know, I want action to happen on those things. I try and be an active part of the community that's changing how that happens.
But I also think that in the meantime, I want to kind of establish my own rules for what I think is an ethical way for me to engage with the technology and, you know, something that aligns with my values, which is, I want to do things that are accessible to everyone. I believe the world needs to look like the world really looks. I really want to try and override systems so that I'm an active participant in training the model to see and register a wider variety of human beings.
SCOT: That's awesome and if you don't mind, I actually would like — because you do have a Midjourney Mondays Manifesto and I would love to actually go through some of these bullets that you came up with.
So, you have “I shall not” and I'm going to go through these and we can talk about some of them in a little more detail, but “I shall not use an artist's name in a prompt to replicate their style and present it as my own work.” That's a very common sense. In the artist community, there's chatter every day about intellectual property and copyright and I think that's a great one to take.
“I shall not use words of which I do not understand the meaning or cultural history to generate images that look cool.” I think that's also very timely right now, and I can speak to that. Have you used the Midjourney “describe” feature?
LORI: Oh, yeah, yeah, absolutely.
SCOT: It can be problematic, you know, like I took one of your dance images and put it into the “describe” feature of Midjourney and, it's interesting what it may find, an art style, it does come up actually with some artists’ names, African American and Japanese artist names, I guess it saw some similarities into those styles.
But I have actually seen some questionable or downright problematic, you know, things come back from images [with the “describe” feature] like, oh, it happens to have a black person in here, but it's a, it's a lifestyle sports, outdoor sports sort of photograph maybe, and it'll say something, like “African wildlife influence” or something like that. And it doesn't say that if it's a white person running. So I think that this is a really important thing to talk about here.
LORI: Let's just pause on that for a minute because I think this is the big realization that I had and that I think a lot of my colleagues have had: we now have evidence of implicit bias, right? If anybody had a question and didn't believe that it existed, I can show it to you. It's now coded right in the system. And one of the things that I think a lot of ethicists talk about is the fact that we've now frozen in time, our history and that history is sorted.
So, I'll use my example of white women. One of the first studies that I did, I just typed in, I wanted to see an elderly woman wearing some of the clothing designs that I had. And so I typed in, “elderly women” and then I put in my designs and they all came back white and I didn't write white. But, you know, if you look at the way that we use language, if you look at the way that we classify ourselves in racial and ethnic categories, white is the default. And it's very rare that anybody would call themselves out as being white, right? We definitely ask people who are not white to give themselves a label, but we don't do that often for white people.
And so I think like what you're seeing in Midjourney is you're seeing a mirror on history, you're seeing a reflection of where we are as a society. And I think that's the first step in change. So my intention is to use these in a way and, you know, I talk with students all ages 17 to 71. It's like, “look at what is happening. This is the way the world is. If you want to change it, you have to actively counterbalance.”
SCOT: I think you're right. I mean, basically the predominant digital imagery or imagery that's scanned or whatever, even from the invention of photography and video, basically people pointed their cameras at white people mostly. So that's obviously reflective in the data. I had a very similar experience and I would say I think many people have had with Midjourney, I was getting some really cool images back. This is during version four, so the faces and fingers and stuff were a little stranger, but I was still really enjoying the the imagery. It was black and white imagery of people eating food on an Italian beach and they were really cool, but they were all white people. And I had to specify black people or Asian people, etc. if I wanted any type of diversity in the images and if I just wanted white people, then I didn't specify and that is the, that was the default Midjourney, the same experience [as you].
I think everyone would be wise to read your Midjourney Mondays Manifesto. There are other bullet points. I think everyone should think about how they are using these tools and it’s easy not to think about it. You do something cool, like you know, maybe like recently in the news an artist's voice was used in a song that's not really that artist but I posted on LinkedIn, that's kind of a cheat code to go viral, you know, like you're using someone else's name or likeness without their permission.
You have a bullet that says “I shall not generate photorealistic images intended to represent current events that did not happen.” I think that's also a huge bucket, misinformation, deep fakes.
Also I like that you put — basically imagery in general, human imagery has for probably from the beginning of it has has perpetuated objectification and violence against women. So, “I shall not present hyper sexualized images of women who are not controlling the gaze.” I think that's also very important.
So I think everyone could maybe, it would behoove them to make their own you know, manifesto of how they will use a generative AI tools.
LORI: Yeah.
SCOT: I would like to also get into you through your company Synthetivity, you do workshops and boot camps. What specific generative AI tools do you teach? I assume ChatGPT and Midjourney, but are there other others?
LORI: I start with free tools because that's accessible to everybody. So I teach ChatGPT and DALL-E, and you know what I'm really teaching is communication skills. I'm teaching people how communicating with a machine is different than communicating with a human. I am teaching them, a lot of people will call it prompt engineering, but I actually call it prompt poetry because it's more like the social sciences and the humanities than it is engineering and my father is an engineer. I definitely think he prefers numbers.
SCOT: I certainly think engineers are kind of like, “Prompt engineering? What are you talking about there?”
LORI: And I teach people about the art of losing control and this to me is one of the toughest lessons to teach to corporate executives, particularly women. And I think it's because, you know, I ask everybody in the room before the class starts, “Who likes to be in control?” And, everybody raises their hand. I don't think there's a person who is in corporate America who's in a leadership position, who doesn't want to control the outcome of their actions. But the thing about generative applications is that they're all about statistics, probability, and chaos.
And you can't control the outcome and you know, there's a certain metric that's called temperature, which is kind of how narrow the field of statistics you get when you're asking for an image to be generated or words to be generated and you can set that band narrowly or you can set it widely and, you know, that creates way more variety if you have a high temperature setting.
But I think what we all are learning with this is something that I learned actually as being a client being on the client side of architecture, which is that, half of the work is in what you communicate as your intention. And then you really turn things over either to a third party who might be doing work for you or to somebody on your team who might be taking things. And you then get something back and at that point, your role is to curate. You can comment, you can give feedback. You go back and forth and you have a conversation and I think this is an art that everybody in business should learn. Because I think if we are so focused on controlling outcomes, we really miss opportunities that come to us serendipitously. I think creativity is going to be the superpower of this era.
SCOT: I think you're right on with prompt engineering, being not the correct word or maybe being something like prompt poetry. To me, there's a direct line between the conversational style of these large language models and then the input that we would do to create imagery with these machines. I mean, why would people learn nuanced differences in different ways is to prompt all the products that are coming, it's going to move to natural conversation of what you want to see as an image or whatever. So why wouldn't it be?
I get there's a short term, I'm sure plenty of people are having a lot of success with their paragraph long prompts with weighted, you know, colons and numbers and everything and they get very specific and it can happen, you can do it that way, but I just don't see that being the way in 3 to 6 months that that's how people are going to be successfully prompting these tools, it's going to naturally get to where it's communication, like you said, and it's going to be natural communication because that's how these, that's the UI that is basically being rolled out and developed right now is, is essentially chat and voice.
Well Lori, this has been a very insightful and wonderful conversation. I really appreciate you coming on the show. I want to let people know that there is a code for 10% off of workshops at Laurie's company. The code is just QUICKBITS10, I believe the website is synthetivity.net — Is that correct, Lori?
LORI: You got it!
SCOT: Synthetivity.net
QUICKBITS10 will get you a discount on some workshops there on Lori’s website and I can't tell you how much I appreciate you coming on the show.
LORI: Oh, thank you and thank you for such thoughtful and insightful questions. It was really great talking with you.
SCOT: Have a good day.
LORI: You too.
SCOT: Bye.