From Humans to Robots: The Ethics of Using AI Workers

Ryan Purvis 0:00
Hello and welcome to the digital workspace works Podcast. I'm Ryan Purvis, your host supported by producer Heather Bicknell. In this series, you'll hear stories and opinions from experts in the field story from the frontlines. The problems they face, how they solve them. The areas they're focused on from technology people and processes to the approaches they took that will help you to get to grips with a digital workspace inner workings.

Yeah, very good. Thanks. Very good. Just as usual, too many balls in the air. Too much going on? It feels like like the world has woken up again. Something. Huh?

Heather Bicknell 0:46
I can I can relate to that. Just go go go. Never know when to take a moment to take a deep breath and take it all in.

Ryan Purvis 0:59
Yeah, yeah, exactly. And it just feels like I don't know. Just just from a timing point of view, even though you're not commuting, they maybe what's happened is we've just we've just filled up the time now because you're not commuting everywhere. So now you have even less time because you don't have that buffer when you commute? Maybe that's it? I don't know.

Heather Bicknell 1:30
I sort of wish from a work perspective. Like, imagine this. Yes, you go into an office every day. Maybe you have you know, a commute. That's not super fun. But there's no work you know, work tech yet I can follow you home, you know, you don't have a accompany laptop, your you know, your phone doesn't do your email. And you just go home and work is I think that's it for me is that even you know, before commuting to an office you know, everyone's leaving around the five to six mark, you shut your computer down. Maybe you'll check a couple teams messages or your email at home, but you're not going to you know, get you can't get that sucked into work, because works at work. So that's it for me anyway.

Ryan Purvis 2:30
Yeah, don't you mean, I remember when I was starting up my my career in the Microsoft world as an MD, and he's to me Tell me course working late at night, which was pretty much every night, used to say I wish we were just making widgets. You know, there's only seven, we didn't make a day. And then if you don't make enough you, you don't make that day. You know, that's it? Because yeah, I don't think you've understood the whole technicalities of what we're actually doing writing software. So his, his understanding was always limited. And, you know, people moan about working in an office job sometimes like a normal job. And sometimes you wish you had that job. Because, you know, this, this kind of work, you can just do all the time, you just, you could you could work till midnight every day and still not finish it. Because it's always there.

Heather Bicknell 3:24
Yeah. So you have to set your own boundaries and stuff, but it's easier said.

Ryan Purvis 3:32
Well, that's that is the problem. And you know, you've got to give yourself permission to take a break as well, you know, so going for that walk or watching a little bit of Netflix, not not everything. Otherwise you just blow up. And that's I think mental mental wellness has come back. It's always been there, but it's definitely something that's talked about a lot more often.

Heather Bicknell 4:02
Yeah. Well, I mean, I don't want to complain too much, because we have some pretty hefty topics for today. But really put, I guess, work in a new light. So we talked about, we have two topics really at the intersection of I think humans machines and labour and they're both interesting, because in one, one of the topics we there's it's sort of like the the wizard behind the curtain, you know, we think seem much more magical and refined than they may be they truly are. And then on the other topic, I think it's interesting because maybe we are, in some ways more ahead than we might have thought. So I don't know if you want to I can dig into the first one on the, you know, human beings training a lot of the AI that's out there today. Both, you know, the low paid labour that exists out there, and the companies that are hiring people sort of on a contract basis to do AI, you know, data model training, and, you know, train image engines and stuff like that, which can be really stressful. And in some cases, like, traumatic work, even for things like content moderation.

Ryan Purvis 5:40
Let's start with that. And I'll be honest, I mean, we used to joke about the thing of the student API API, which didn't really translate to a lot of in a lot of meetings, people got the concept, but basically, you're paying students to do the work until you could justify investing in the, the sophisticated tool. And I've worked in various organisations with with various complicated things that AI could be used to solve, and the one that I was thinking about, as well as legal document, when you sign agreements in different countries, and the clauses and all the rest of it are written in the in the local language, but you've got to get that back to a common language, and make sure that you haven't misinterpreted the clause. Through the translation process, and the company in mind ahead about it was a few, a few 100 employees that were PhDs, legal experts, and also language experts. So they were translating, you know, from Polish to English and French to English, and whatever it was, and they were trying to do all the stuff. And the theory was if they could automate that using machine learning, they could save like a million dollars a month in salaries. But I don't think they get it, even though they could use the technology, they could only get to maybe a 60% 70%. Improvement. She still needed the experts, but at least she was speeding up the process. Because that was the other thing is, you know, to go through contracts takes time, especially ones with with a note with with nuances to them, because of it's always a different deal. And whatever it is. But, you know, this article surprised me, because of some of the other things which you know, that I know, I knew of products that were built where there was no AI, but they were telling us who they are. And you just knew because of some article by accident where they let it slip. But then you find out through the grapevine about other ones where they're selling it as they are to, but actually, it's a whole bunch of humans doing that. I think one of them was a scheduling tool for meetings. Before Cortana came along, and they said they would do this tool would organise your meetings for you. And then you realise, and they found out there was people because the language wasn't exactly right. It was different, different people writing responses. But yeah, it's it's it's not surprising now that they haven't read the article.

Heather Bicknell 8:17
Yeah, definitely. Unfortunately, not not surprising, I think we know, right, that a lot of even even the stuff that actually is AI putting aside the things that are just faked, like fully fate, that these, you know, algorithms, these, these datasets like this all a lot of the there's a lot of repetitive, like, unglamorous work that's behind building up a lot of these, you know, large models and actually doing the training piece. And there are humans behind that. So we you know, we've been hearing a lot about, you know, chat GPT and generative AI and all of this stuff, I think, you know, what, it's coming in the, in the news cycle a little bit more, but the, the actual people behind training a lot of those you know, things that just like are blowing people's minds right now, we kind of missed that there. It's not just AI, you know, automatically doing it right, that there are people behind the scenes, maybe not getting treated super well. Or actually getting the things to that state by by training them.

Ryan Purvis 9:43
Yeah, and if you remember that, you know, in Captcha came out on your website, there was it was asking you to confirm those images, etc, that that was a part of crowdsourcing, labelling of the data, which is what, what we're talking about here and, you know, in order for machine learning and all that stuff to work, you need to have a set of data that It is correctly labelled, and a set that isn't and allow the algorithm to process and this is supervised learning. And allow the, the data that isn't labelled to go through the the algorithm, versus what has been done. That's what they call it, they call it when they call to training. That's what I mean. And then you slip somebody to someone to check it. Now, the answer, the unsupervised learning is basically point the algorithm and all the data and see what it comes out with, which is really groupings and classifications and that sort of thing. I think that, you know, if you look at any sort of sweet sweatshop, stigma, you know, the the kids building shoes in Vietnam, or whatever it is, for Nike or Adidas, wherever it was, I don't know, it was back in those days. That analogy could be applied to this as well, were you paying people to do you know, branded daily caption to be honest, and paying them a low wage, but you got to pay in the context of is that wage low? By first world, society based on where you live, you know, the pound versus the dollar, versus what they own in that country. I mean, I was talking to people this morning about some resources, and discussing what you'd pay for that resource in the UK. And if you multiply that into rains, it's a ridiculously big number. I mean, you're talking three, four, or five times the thing. And then if you go look at the resource, you know, here in South Africa, it's a good resource, getting a good salary until African terms. But not what you'd be paid in pounds, but then your living costs in pounds is also significantly higher. So getting boys got to balance these things out. I guess, as long as the people are humanely treated, and they get benefits and employee benefits and all those sorts of things, or at least they're entered into contract with a choice they get forced into it. That should be okay. In theory.

Heather Bicknell 12:02
Yeah, I think, Well, I think it was a vice article that we're discussing. So we'll link it in the show notes. So they do bring up the point about, you know, wages being their money going farther and different economies. I think one of the interesting points they also noted was that because it's a lot of this is gig work, that there's not sort of advancement opportunities, necessarily in this line of work. So you, it's being even though there is some skill, and there is skill to this labour, if you read the experiences of people working, doing this kind of work, they they do have to learn things to be able to do it, I think it's being, you know, treated as unskilled labour. And then they don't have because it's gig work. And they're just doing, you know, one off things, you know, bidding for jobs, basically, there's less of an advancement pipeline. So it's just good, you know, good thing to be aware of that this kind of work, just like we, you know, became people were aware of sweatshops, and you know, any of it, it's like, yeah, a digital, a digital version. AI isn't just magic, there are people behind the scenes sometimes. So I think, just good to be aware of that, in all the noise about the excitement of AI.

Ryan Purvis 13:31
Yeah, look at this look, take away from you know, chat GPT. As an example, I understand this one, at least, you know, that that is something that has been built, designed, tagged correctly, you know, all that kind of stuff. And to our knowledge, so far has used unsupervised learning to build. Its training models, and it's, you know, it's in this fourth version of the three and a half now, going into its fourth version. You know, a lot of APIs that are built that way, where it's, it's, you know, truly artificial. But the other ones we exactly that they've taken the thing I mean, you know, I'll share a story of my recent trip to Joburg and I was talking to one of the drivers that I use. And I met him as an Uber driver. And I just, you know, at the time said, Look, you know, I need a trust with the driver, then I can just call them and I need to, I don't want to go through but I'll pay you directly. And he's been my driver for you know, every time I do a trip to Joburg and, you know, over the years, and he was telling me about how, you know, he can't even live off when he gets out of Uber anymore because their algorithm and their choices are impacted indirectly they're you know, they're deciding that get a trip that's not to 20 kilometres they'll pay one rate and then a trip that's over 20 to over 20 kilometres that pay a quarter of the rate. And you know, in Johannesburg and Cape Town and all that most your trips over 20 ks And so he's always getting the worst rate, such as templates, you know, all the other Uber drivers and you know that's that's a case where the data is not being applied to the, to the economy that it's in. And if I pay a private taxi guy, we're in Las Vegas at the moment I pay them 10 ran a kilometre. The equivalent for this driver in Joburg is about Adrian, a kilometre, within 20 ks and goes down to four Rand over 20. So, you know, we don't have to go through the whole math of why that's bad. But it is bad. It's it's, I mean, the price of a litre of fuel is 1314. Rent are $9. And pounds, you know, these are very small numbers, because you're dividing that by a team environment by 20. But these guys have to live support families, etc. And this is where you and I think this is part of where Elon Musk was saying things need to be regulated. If you're using people in involved in these things, there needs to be some level of minimum wage that is regulated, so we implement a gig based technology that needs to comply with the minimum wage in the country to provide the minimum living standard. If you are and you're right, there's no there's no career progression. I mean, you know, this driver or what the driver here, the tech, the other what's the other choices, they can't go and become a team leader or earn more money. That way they and you know, they're driving long hours to generate the minimum income. So they're, they're stuck. Now, if you look at where some of these plans are, with these, with these technologies that are self driving cars, and all that kind of stuff, these sorts of jobs won't even exist in the future, if these are all going to play up. So, you know, what do you do with those people? You know, the driving job and was saying to me, you know, what do I do I need to learn computers? Now? How do I learn computers? And you're sitting there going? Well, how do you train someone that's got no skills really very common, because most of them don't. Into the technology world, we already the the chat GPT is generating code, you know, writing your problem statements for you, and all that kind of stuff, because the data has been processed and pushed out that that curve has just gotten more aggressive to learn at a speed to catch up. So it's a very interesting and scary problem that's coming. Because you're gonna have a whole lot of people that don't have the know how to go up a level in many jobs, not just the one I mentioned.

Heather Bicknell 17:54
Yeah, I think this is actually a good transition into there. The other sort of side of the human machine labour equation that we've been looking into, which is robotics as a service, you know, the latest as a service that I have heard about everything, what can't be as a service now, I challenge you. But yeah, renting robots essentially, mainly for manufacturing or warehousing roles, obviously, it's replacing sort of one route task of you know, lifting something here putting it something there screwing, you know, whatever widget in but there's, you know, definitely aspirations and goals to add AI and ML to the mix and make these things more intelligent and be able to do more skills as well. And I thought it was interesting, there was an employer talking about how, you know, he can rent these robots for $8 An hour and a human costs $15 an hour. So, you know, he personally is like, well, we're not going to downsize our staff, but maybe we don't need to hire more. We can, you know, let our people do more interesting work. But it is overall, I think, going to reduce the job pool. As things like this become, you know, like the economics therapy can rent free versus paid for 15. And especially if you're a small business, you know, it's it's the classic robots robots will take our jobs, which, you know, that I have in a lot of manufacturing settings already. There's a lot of, you know, technology that has come in to replace previous human labour.

Ryan Purvis 19:34
And I think some of it makes sense. I mean, I remember a very good book many years ago, what was it called? It was called airframe. Why do we disclosure actually, I think was disclosure was Michael Crichton book. And, and it's, I mean, the stories are about sexual harassment and all that kind of stuff. But the part that's important to our conversation is they were manufactured showing a computer in Asia and they couldn't get it to me Quality Standards. And what they realised at the end of the book, The conclusion was actually the problem was that the humans putting the chip into the computer onto the motherboard, didn't have the accuracy, or the finesse, to put the chip in the right place every single time. And that's why some batches were good. And some batches were bad. And it's kind of that point. So when you have repetitive tasks with with a need to be accurate and onerous, I have no problem with that being the machine driven thing. And if it's an intelligent machine, it does that and it saves the human life. Or it augments what the human will do. And you remember, we talked about their power suit for lifting things and that sort of stuff, you know, I don't have a problem that if you look at logistics and warehousing and that sort of stuff, we're going to put boxes away and pick them up and carry them move them around, it makes total sense that that's, that's using some sort of technology to solve that problem. Where it becomes probably concerning, again, is, you know, a lot of the economies that exist are based on having a certain level of blue level certain blue collar and a certain level of white collar workers. And as the technology will eat into those industries, where do you put the people that don't have jobs anymore? You know, how do they provide for families, and you know, all that kind of stuff. And you've seen it already in some industries, where, and, you know, self driving cars, self driving trucks, probably not the ones to look at, but it's, it's the, you know, I was doing some legal work over the weekend, and I asked a lawyer for something. And he didn't charge me because he said, Look, all I did was write down three words, three sentences to what I wanted. And we generated the template from from this product that we use. And, you know, I'll only charge you if you want me to change anything in the contract. So, you know, that that would have normally been a week's worth of work, let's say. But, you know, I got the contract looked at it was actually, you know, fairly, fairly, okay. I'm sending it off to another attorney to give it a second look, because I just want to make sure that I do so quickly, so cheaply, that I haven't actually opened up a huge thing. And that's, you know, that's a sort of white collar job that's potentially going away. If you look at, you know, something very basic, like providing garbage in garbage collection services, you know, why could that not be an automated bot that drives down your driveway and picks up the using, you know, Vision AI to pick up the right dustbin tap into the thing, and there's a human watching it, but you know, that a carry on, because that's a that's a meaningless task. To be fair, and why don't have a machine do that. And if you're dealing with hazardous waste, and all those things, I think it makes sense. Where we, I think we get into another murky thing is, you know, some some countries in South Africa and one of them legislate that there has to be a job for everybody. And, for example, the petrol attendant is a very common job. I mean, there's always there's always someone to do the patrolling. And by law, there has to be a petrol attendant. So even though you could have a machine, or you could do it yourself, they will still be a patriot, there has to be paid. So you'd have to see constitutions change, in some respects. And I don't have any other examples, but there was one that I was thinking of when when I looked at this article, because I think it needs about the same thing, you know, there were people helping the person doing the job, just because they have to pay a person to do something. So there was like a guy carrying the broom for the guy street sweeping the streets. To get a job. Now,

Heather Bicknell 23:48
I think it's really complicated. And I wonder how, you know, some of these things could intersect with movements toward the four day workweek, or even the third three day workweek, I think, you know, if, or, you know, we haven't even touched on things like universal basic income, but, you know, if we all were talking about how stressed you know, we are with everything that's going on, and there's so much work if we could just spread out that work a bit more and also skill people up. But anyway, yeah, I wonder if some of that could be sort of help Redistribute?

Ryan Purvis 24:29
So I don't know if you saw that that study on a four day week in the UK, I think is a UK. Of all the companies that did it. I think the majority are sticking with it. So they're not going back to a five day and I think that is going to be the reality. I think you're gonna get a situation where and there's been a lot of stuff and in one of the podcasts is about one of the publishing company was CNET was generating a whole lot of articles using AIs. And when Wordsworth was the AI thing, And I was saying yeah, but these are for generic articles that it's not worth having a journalist write them. And you know, obviously the word it's been using it's using the AI but it's also using an AI to make it SEO heavy. Which kind of becomes funny because now you're writing stuff artificially to become artificially more search sensitive and the algorithms are artificially using it and they're going to discard it because it's too perfect. But I think it'll happen you know, I look at the you know, into the movies on Sunday. We won't okay look again, you know, a lot of people involved in the process, but I booked everything online. I including the food I got the the food is really I picked it up, we walked into the into the cinema, I need something more I texted the guy on WhatsApp, I need nothing on popcorn, he brought it to me. That could be a robotic thing. You know, on a could be a drone flying around carrying popcorn. But the point is, it's the technology has made it so close to demand driven that that person was bringing me the popcorn literally was just, you know, he didn't eat oranges pick it up and bring it to me there was no, there was no need for him to process the credit card transaction or take the order or anything like that. He was literally just bringing it to me. So how long will that job is just gonna replace buyer. Something else?

Heather Bicknell 26:21
Yeah, that's like a question. And it's happening everywhere. So I'm sure I'll be the last time we talked about that. But unfortunately, I do have to go here for today.

Ryan Purvis 26:31
Super. Thanks, Chat soon

Thank you for listening to today's episode. Hey, the big news app producer editor. Thank you, Heather. For your hard work on this episode. Please subscribe to the series and rate us on iTunes or the Google Play Store. Follow us on Twitter at the DW W podcast. The show notes and transcripts will be available on the website https://www.digitalworkspace.works/. Please also visit our website https://www.digitalworkspace.works/ that works and subscribe to news. And lastly, if you found this episode useful, please share with your friends or colleagues.