00:00:11:01 - 00:00:36:21
Speaker 1
Hello everyone and welcome to another edition of Ascii Anything presented by Moser Consulting. I'm your host Angel Leon, Moser's director of Personnel. In today's episode, we're going to be talking about a topic that I think is taking the nation, if not the world, by storm, and that is generative A.I.. And with me to discuss this very interesting topic is Moser's director of strategy for our Application Services Division.

00:00:36:22 - 00:01:00:10
Speaker 1
Steven Peavey. Steven has been with Mosher Consulting for six years. Prior to becoming a director, he was a developer designer, Scrum Master product owner and engagement manager. Outside of work, he enjoys spending time with his family, which include his four children. One of his life goals is to bicycle from Alaska to Mexico. That is quite the life goal and I want you to tell me all about it.

00:01:00:15 - 00:01:01:24
Speaker 1
Before we start with the episode.

00:01:02:01 - 00:01:29:19
Speaker 2
Yes. So in Prudhoe Bay, I believe that's the location. It's very northern tip of Alaska. And I would like to start there and bike all the way down to the border between the U.S. and Mexico. So I think it would take at least three months. I would need and a lot of training before working. And yeah, there's just something that I read this blog article, somebody else that had started there and they biked all the way to the southern tip of South America.

00:01:29:21 - 00:01:41:08
Speaker 2
Mm hmm. And it took them it was on and off for two years. Just the dedication to that and, you know, not many people have done that. There's just something that that calls me to that. So I need to start working my way up to that first.

00:01:41:22 - 00:02:04:19
Speaker 1
As somebody who's lived basically all the way down to the tip of South America, I lived in Santiago, Chile, for about a year and a half, and you feel like you are all the way down, especially when you're traveling even by airplane. The closest trip from Santiago to the U.S. was Miami, and there was always a four and a half hour or five hour flight.

00:02:04:19 - 00:02:23:06
Speaker 1
But it felt like it was 12 hours because it just feels that way. Because when you look on the map, how far down on the wall you are? It's amazing. So, yes, I think I know what you're talking about. I may have watched it. I can't I'm trying to break my brain about some interesting thing that I remember about it, but it'll come back to me.

00:02:23:06 - 00:03:01:23
Speaker 1
So very interesting. Like all you're going to say is very cool. So it's great to have you and ask you anything to talk about what I think is a topic on everyone's mind. I've seen this topic on the news talk shows, night shows, you know, late night shows, etc., Even honestly, even some of our own internal meetings. I've been a part of a couple of meetings where it's been brought up the topic of, you know, chat and the little chat box that you can type in information on and you ask your request, Can you type me X, y, Z on x, y, Z, and then you get something back.

00:03:02:00 - 00:03:06:24
Speaker 1
So let's start with the basics. What is generative A.I.?

00:03:07:03 - 00:03:30:21
Speaker 2
Yeah, so generally it is a type of artificial intelligence that allows for people to create new content. That new content it can be text, which what you were just talking about, like with the chat bots. Mm hmm. It can be imagery. So if you've seen some text to image generators like Dolly to or mid journey, that would fall under that.

00:03:30:23 - 00:03:55:24
Speaker 2
But it can also create code or music, video and even 3D generated objects. So traditional A.I. usually is for analyzing things, analyzing existing data, but it never really created new content from that. And now we're starting to see that change. And that's where generative AI comes in. There's usually three core pieces from my understanding of it and just has a little disclaimer for the listeners.

00:03:56:06 - 00:04:15:11
Speaker 2
You know, I first became aware of generative AI last August, so I'm by no means an expert in this, but from my reading up on it, there's there's three core pieces. You have the A.I. model itself, so you'll hear things like stable diffusion. That's a really popular model that's been used on some of the popular commercial products you've probably seen in the news.

00:04:15:12 - 00:04:41:14
Speaker 2
And then you have a large dataset and then processing power. So they've actually been trying to do this since the sixties, from what I found. One of the very first chat bots was from the sixties called Eliza, but it was very limited. And what it could do, it was limited because it didn't have enough data. The A.I. models weren't as complex as they are today, just the processing power overall, it just wasn't there.

00:04:41:19 - 00:05:05:17
Speaker 2
But they've been trying since then. And then also, you know, just has a timeline for the listeners. 2014 is when we made a huge improvement with Gans or generative adversarial Networks that came from Ian Goodfellow, and basically it pits two A.I. models against each other. It's a zero sum game. So just to break it down, there's the generator and the discriminator.

00:05:05:17 - 00:05:27:14
Speaker 2
Those are the two models. And what I mean by zero sum game is if the generator gains something, the discriminator loses something, so the generator creates the data, and then the discriminator will determine if that generated data is real based on the huge data set that it has. When we talk about you've probably heard about things maybe in some of these trending news articles about datasets and copyright.

00:05:27:17 - 00:05:49:23
Speaker 2
Mm hmm. Especially with art. Look, these data sets, one of them is called Lion five B, and what they mean by the find B is there's over 5 billion images. Mm hmm. And maybe it was Dolly, too, but one of the text to image generators they used Lion to B, which contains over 2 billion images that have English text descriptions.

00:05:49:23 - 00:06:19:08
Speaker 2
And it's over six terabytes in size. It's massive, and it requires a lot of processing power, too. So that was 2014, 2015. We saw diffusion models come out. It's an evolution from Gan. And then 2021 we saw Dolly version one that came out at the end of December. We saw latent diffusion models. That is just that it reduces the generation time and costs to create images and text.

00:06:19:08 - 00:06:44:07
Speaker 2
And then we started to see a lot of publicity around generative AI in the news. You probably became aware of it last year in the second half of last year. And the reason for that is we had Crayon or Dolly Mini that was from Openai that was released to the public July of last year. MIT Journey was released to the public in July, end of September, Dolly two was released to the public.

00:06:44:22 - 00:07:01:18
Speaker 2
And then on the open source side, we have stable diffusion, which is part of that diffusion modeling that was released to the public in August. And since then, it's just it's hard for me now even to keep up with everything that's trending right now. You know, I'm having people bring up things to my attention that I don't even know about.

00:07:01:18 - 00:07:26:06
Speaker 2
Oh, that's interesting. It's moving at such a faster pace. And even last year I mentioned August is when I first became aware of it on Hill. I don't know if you remember, but last September, this was the first big trending article was a man from Colorado won the state fair art competition and he submitted his piece. He ended up winning and come to find out he used mid Journey, which is a text to image generator.

00:07:26:06 - 00:07:47:21
Speaker 2
And a lot of people were upset because they said that he wasn't the actual artist and the work that the generated output of it. It was because of all of the artists prior to him that published their work that he was able to even do that. And you start having people bring up copyright concerns, issues, and that's still an ongoing thing as well.

00:07:47:21 - 00:07:55:05
Speaker 2
But last September really was when I think generally I really started to become people became aware of it.

00:07:55:17 - 00:08:24:09
Speaker 1
Speaking of that copyright issue, because one of the things I actually read today was that Getty Images is actually suing mid Journey because they take images from their files that are already copyrighted. And one interesting thing that I found about that article that I read was that one of the pieces of evidence was a picture that generated from mid Journey that I can't remember what it was.

00:08:24:09 - 00:08:48:02
Speaker 1
It had to do with sports. I think it was a boxing match or something. So it's two boxers. They look to be in mid-flight, but then right at the right bottom corner of the picture there is this blurry image that resembles the Getty Images logo. And so it's funny because you can tell that t t y from Getty, the g are kind of like badly drawn.

00:08:48:02 - 00:09:25:13
Speaker 1
It's like if my one year old took a crayon and just kind of drew a Jenny together, but then you can tell that t t y and then images, it's kind of slightly blurred. But of course the middle journey is pulling that out from their own algorithm. And then whatever billion images that it has. So it's interesting that we talk about that in the art world, because the art world itself, it feels from the little bit of reading that I did, that it's actually up in arms say, No, we can't have this because it's taking the work of all of these artists and basically just stealing it.

00:09:25:19 - 00:09:26:21
Speaker 1
Plagiarizing. Yeah.

00:09:26:21 - 00:09:49:17
Speaker 2
And the counterargument to that and I'm not on either side, right. But the counterargument is that this is the same way that people learn how to draw is they look at existing artwork. And it's just that in this case, people know that the watermark isn't part of the original art. And in this case with Journey, it thinks it's part of it, which is why we're seeing the watermarks.

00:09:49:17 - 00:10:14:23
Speaker 2
Other competitors like Dali to there is a large amount of text image art generators and they even in their dataset when they train their program, they tell it do not include the watermark. So you can peel back. Yes. Yes. And you can work on that. One other thing I want to add to that brought generative AI to the public's attention was GPT, which came out.

00:10:15:13 - 00:10:27:19
Speaker 2
It was released to the public November 30th and four days later they had over 1 million active users, which is incredible. Yeah, it's amazing to see just how fast this is grown.

00:10:27:19 - 00:10:52:11
Speaker 1
And we're going to probably take a look at that here in a second. But this leads me to my next question is what impact do you think this type of AI is going to have for the workforce? Because we talked a little bit about artists themselves, but you mentioned Chad Djibouti, which basically is a text to text A.I. where you requested to write you X, Y, Z on X, Y, Z.

00:10:52:11 - 00:11:14:15
Speaker 1
And so it comes back. I just saw a late night show that happens on Sundays. And they talked about this and somebody made a song about Eminem rapping about Cats, and it was super hilarious. It was actually it was rhyming. It was everything you wanted to be, and it was Eminem singing to Cats. So what can you tell us about the effect that it'll have on the workforce?

00:11:14:20 - 00:11:39:08
Speaker 2
Well, before I so I started, I also have another article that I've been working on, and originally I thought that it would just be illustrators and close professions that are close to that, like maybe like painters or digital artists. But after doing some research and looking at a few different resources, you know, to your point, I think there's a lot of other occupations that will be impacted by this.

00:11:39:08 - 00:12:08:23
Speaker 2
So my estimate is that 18% of the workforce is going to be affected by generative AI this year, which was around 28 million people. You mind if I walk you through how I turned out? I'm still. Yeah. So I started by looking at the Bureau of Labor Statistics. They use a federal coding system called the Standard Occupational Classification System, and it breaks down occupations into four levels of aggregation major groups, minor groups, broad occupations and detailed occupations.

00:12:09:01 - 00:12:31:08
Speaker 2
So that's been really helpful. And then also I've been looking at OWN that which uses the data from the bureau and it gives us information. It describes each person's occupation and to a framework called the content model. And it has six domains. And one of those domains in particular is really great when we're trying to come up with how generative I can impact occupations.

00:12:31:17 - 00:13:03:12
Speaker 2
So that domain is the work activities domain. And under that they have general, intermediate, detailed and even down to the task level of what people do. So what I did was I started at the task level, but there were tens of thousands of tasks. So I worked my way up to the general work activities and there's, I think 41 and I basically put this on this rating scale of how likely is it that generative AI will impact and this one general work activity, and I found three in particular.

00:13:03:16 - 00:13:28:19
Speaker 2
The first is thinking creatively. So occupations where you do something that thinking creatively is important, that you are more likely to be impacted by generative AI. And we can get a list of occupations where we can see what the level of importance is for thinking creatively. On the flip side, we have two things where I think I will not impact it, and that is performing.

00:13:28:19 - 00:13:53:10
Speaker 2
General physical activities and handling and moving objects. So anything where we have to use our hands or body chat, GPT or Dolly to or mid journey, they're all things where we're at our keyboard, we're typing things in and it's not requiring us to move things around or use our body or hands. So right now in his current state, we don't have robots walking around, not mainstream.

00:13:53:10 - 00:14:18:02
Speaker 2
If you've seen Boston Dynamics, they're getting there. So it's on the horizon. But for now, you know those other two performing general activities and handling moving objects will subtract from the likelihood. So with those three things, I basically I took those three and I looked at every occupation and that gave me a list. And I can look at the major groupings, the minor groupings.

00:14:18:02 - 00:14:39:21
Speaker 2
I can even go down to the detailed occupations and see which jobs are most likely to be impacted by generative AI this year. And with those findings, the one group that's most likely to be impacted is the computer and mathematical occupations. We have 2.8 million people that fall under this grouping. But there's another component to this called the job zone.

00:14:39:21 - 00:15:01:09
Speaker 2
And a job zone is the level of preparation needed to do a job. And it goes from jobs number one, all the way up to five. So if you have a job zone of five that requires the most amount of preparation needed, that would be somebody like a surgeon or a CEO and then going down to a job zone of one, this requires 0 to 3 months of preparation.

00:15:01:14 - 00:15:22:18
Speaker 2
This might be somebody that could work as a cashier, just very little prep work needed. The computer mathematical occupations. They had an average job zone of four. So there's a lot of prep work needed and we'll circle back around to the impact I think this will have on the people under this one occupation. But there's 23 of them.

00:15:22:22 - 00:15:46:18
Speaker 2
I have data for 22 out of the 23 and there's one occupation, one major grouping. I think that will be very likely to be impact and that's computer mathematical. But then we have legal business and financial, we have education and even community and social service, and then it starts to work its way down. Out of the 22, those were the top five.

00:15:46:18 - 00:16:20:22
Speaker 2
We can give you the data on our website. We have a table articles and we can we'll certainly include the data there for the listeners to read up on if you're interested. There is a trend, though, that I saw with these major groupings is that the occupations most likely to be impacted, they had a job zone of four or five, and if a job requires a lot of preparation needed, I think in its current state it's still in its infancy gender today I it's not going to replace workers like I think the common fear is, however there was one major grouping that had 18 million Americans.

00:16:21:03 - 00:16:39:15
Speaker 2
It had a average job zone of two, and it was right in the middle of the likelihood of being affected, and that was office and administrative support staff. So 18 million Americans, it's right in the middle. That's certainly something for us to pay attention to. Are people within that major grouping there. And we have over 18 million Americans.

00:16:39:15 - 00:16:44:07
Speaker 2
So something to definitely keep an eye on when we see generative A.I. mature.

00:16:44:19 - 00:17:09:14
Speaker 1
You mentioned a lot of job roles that could be affected by this, but I agree with you in that we don't necessarily we aren't going to see A.I. replace these roles right now. One of the things I was watching actually the other night was how it should be used first as a complement to these roles. So take a lawyer, for example.

00:17:09:19 - 00:17:29:10
Speaker 1
We talked about lawyers a little bit. So to prepare for a case, it's a lot. I mean, it takes months and months to prepare for cases where now you bring in this A.I. and it can help you write those documents that you need to prepare for court, etc., all those things. So it's really a complement to the role.

00:17:29:10 - 00:17:53:06
Speaker 1
So if you are a 100% lawyer and you can give that maybe ten, 15, 20% of that typing right up that now it could affect maybe your legal assistance, your, you know, which is what you're mentioning in your office administration. Now, it gets to affect those roles specifically, but not so much the lawyer itself, because then they just basically can take their laptop out, type up.

00:17:53:06 - 00:18:12:06
Speaker 1
Okay, I need you to type this for me, this format, and I will type that. Now you still need to review it for accuracy, for, you know, fact checking, etc. But it's going to be there. I mean, it could in the long run replace some of these jobs, but it could also maybe change the dynamic of the job.

00:18:12:11 - 00:18:20:20
Speaker 1
Not so much from maybe the admin assistant, that legal assistant to do so much typing, but now it becomes maybe a reviewer of some sorts.

00:18:21:04 - 00:18:36:17
Speaker 2
What generally the AI is doing is it's reducing the level of effort needed to do certain tasks, right, your job and it's not going to fully automated your occupation or even the tasks, but it's certainly going to help speed up the process.

00:18:37:01 - 00:19:04:06
Speaker 1
Well, in another and you mentioned this at the beginning of your last answer was cashiers. We're seeing that right now and it doesn't have anything to do with generative A.I. has to do with now we see more self-service checkout counters and I don't want to say it's coming right, but you can see there's a light at the end of the tunnel where now these self-checkouts are like, Oh, only for 15 to 20 items.

00:19:04:16 - 00:19:27:22
Speaker 1
Well, what's going to prevent a major supermarket from saying, Well, now it's all like you get to do your own bagging your own self-checkout completely. And I've seen some places already that have it like that that you can elect. That's not just a minimum item, it's you can elect to go either way and they give you the little checkout gun for you to check out the bigger items on your cart, etc..

00:19:28:08 - 00:19:52:07
Speaker 2
Yes. And you bring up an interesting point, too, is it's not just reducing the level of effort, but there are other things that companies are measuring. Sometimes it's to increase profit or to reduce costs. And there's always an outcome that they're looking to achieve and generative AI, it's going to be used as a tool to help achieve those outcomes.

00:19:52:17 - 00:20:06:24
Speaker 1
Yeah. So finally, what do you think happens next? Because we've talked about what it is. What's Kate, what's really capable of, how can it affect our workforce? But what do you think happens next? Where are the next steps?

00:20:07:04 - 00:20:31:15
Speaker 2
I think for the near term for this year and going back to to my theory of jobs and the level of preparation, the let's start with the job zone of one. We have almost 10 million Americans. They do things like door to door sales workers, fast food workers, dining room attendants and bartenders. If I were to affect them, it would be severe.

00:20:31:21 - 00:20:34:13
Speaker 2
But the likelihood of that happening is very low.

00:20:34:17 - 00:20:34:24
Speaker 1
Right.

00:20:35:16 - 00:20:58:18
Speaker 2
And then we have a job zone of two. These are a three months to one year of prep work needed. We have 65 million Americans that fall under a jobs zone up to the likelihood of them being impacted by generative AI's low and if they were to be affected, it would be major. So just one notch below the severe under jobs and one, but it would still be a huge impact.

00:20:58:21 - 00:21:23:13
Speaker 2
These are people like telemarketers, switchboard operators, people that do like answering service type duties. And then we have people under jobs zones of three. We have 26 million Americans. These are the people most likely to see the biggest impact, I think this year, the likelihood of them being impacted is right in the middle is being possible. And if they were to be impacted, that would be a moderate severity.

00:21:23:18 - 00:21:45:21
Speaker 2
And I actually have the data on the three occupations most likely to be affected and they are web developers, which here at Mosher Consulting, we're very familiar with what web developers to give us a shout out if you need help with two wealth development tasks. We have almost 100,000 Americans that that are web developers, and then we have fashion designers and then tutors as well.

00:21:45:21 - 00:22:00:23
Speaker 2
So those three right there and talking about the web developers, they're certainly the most likely to be impacted. But what about the other rules that are related to them or just the tech industry as a whole? I don't think we should count that out. I think we're going to see some changes or just around the tech industry as a whole.

00:22:00:23 - 00:22:23:04
Speaker 2
And then last, we have the job zone of four and five. With four, you typically need 2 to 4 years of prep work. We have 33 million Americans and from the data they are likely to be impacted by A.I., but the disruption severity would be minor. These are people like creative writers or video game designers or search marketing strategists.

00:22:23:11 - 00:22:47:17
Speaker 2
And then last, we have the jobs zone of five. This is you need at least four years of work needed to do the job adequately. We have over 7 million Americans. They are very likely to be impacted by A.I., but how noticeable it would be, the severity would be insignificant. And these are people like operations, research analysts, mathematicians and people in post-secondary titles like political science teachers.

00:22:48:10 - 00:23:08:11
Speaker 2
So that's I think that's a big change we'll see this year and the people who will be impacted. One other theory I have on here is at the job level, I think people who have more experience will be more resistant to disruption this year. So the job level, we have three, we have the entry level, the mid-level and then the senior level.

00:23:08:24 - 00:23:32:04
Speaker 2
And when I talk about changes, I'm talking about changes in their day to day process, but also at the company level, like hiring changes, people looking to change jobs but still stay at the same company. That would be a lateral transfer. And then also resignations like voluntary resignation, but also generative. The AI will have an impact on involuntary resignations as well.

00:23:32:04 - 00:23:51:18
Speaker 2
Yeah, so I see some changes there as well. And I think at a high level, those at the entry level position, they are most likely to be impacted for all five of those bullet points. And then the senior going down to the senior level, they're the least likely to be impacted, but still they're all going to be impacted regardless.

00:23:51:24 - 00:24:30:09
Speaker 1
Before we go, I mean, you talk about those entry level positions. How much would that it eat into what that entry level person might be doing equally, their salary or benefits? All those things that we nowadays look into. Of course, when we are looking at potential positions within another organization. And so when you look at, well, if I'm an entry level admin assistant and my part of my job is to write up assessments or anything like that, well how does generative, if I take a chunk of that away from me, So does that take away from my paycheck?

00:24:30:09 - 00:24:49:11
Speaker 1
Does that take away from the benefits that I am supposed to earn? So a lot of questions, obviously, around this topic that we're not even looking at or thinking of because we're so I think we are wondered by the I'll call it the magic of the air and what it can do for us.

00:24:49:13 - 00:25:15:21
Speaker 2
Yeah. You know, I, I think one thing that will happen even in the next couple of years is we're going to see job descriptions. We're going to see those changing over time. Generally, the high it's reducing the barrier to entry of creating content. And because of this, your employer is going to expect more from you. But these sort of things, I see the stuff changing over the time because of this technology.

00:25:16:03 - 00:25:32:23
Speaker 1
Very interesting topic indeed. Well, I like to think Stephen P and folks, Stephen will be with us for another episode on generative AI that you won't want to miss. We're going to go a little bit of a deeper dive on that in our next episode, so stay tuned for that. But Stephen, thank you very much for being with us today.

00:25:32:24 - 00:25:33:22
Speaker 2
Thank you. It was a pleasure.

00:25:34:07 - 00:25:50:08
Speaker 1
Thank you for listening into this week's edition of Ask You Anything presented by Mosher Consulting. We hope you enjoy listening to Stephen Peavy talk to us about generative A.I.. Join us next time when we continue to dive deeper with our resident experts on what they're currently working on. Remember to send us your ideas or topics your social media feeds.

00:25:50:08 - 00:26:08:11
Speaker 1
In the meantime, please remember to give us a rating and subscribe to our feed wherever you get your podcasts until then. So long everybody!