Introduction and Conference Overview
00:00:35
Speaker
Hello, everyone, and welcome. I'm Josh from Sweden. Hi, I'm Ray from Belgium. And we are glad you're back with us. So about three weeks ago, Ray and I went to a Clojure conference in Belgium.
00:00:49
Speaker
And our guest today gave a keynote at that conference. And that was really impressive. And we decided that we would really love to have a deeper conversation with her to get more into the themes that she introduced us to in that talk.
00:01:05
Speaker
So we are delighted to welcome Ana Kolum on the show today to talk about, drumroll please, AI. Ana, welcome.
Democracy and Technology with Ana Kolum
00:01:15
Speaker
Would you mind introducing yourself and saying a few words about what you do?
00:01:19
Speaker
Of course. Hello. And first of all, thank you so much for inviting me ah to your podcast. So yeah, my name is Anna, as you said, and I'm a researcher working at the intersections of democratic processes and digital technologies, and more specifically in the last few years in data and AI. I have a bit of a background in deliberative ah processes and other type of democratic innovations and um how they intersect with um technologies.
00:01:45
Speaker
So, um Currently, I'm working at the Data Time, which is a nonprofit that's focusing on how we can ah take a more responsible approach to reusing data so that it can help improve people's lives.
00:01:58
Speaker
And prior to that, um I was at the Edelowicz Institute, which is a research institute based in London that also does research to ensure that AI is working for people and society.
00:02:09
Speaker
Fantastic. And hopefully all of our guests know who Ada Lovelace was, but she is widely considered the first computer programmer of all time, right? Exactly. yeah
00:02:22
Speaker
Yeah, so um so Anna, um we were talking um about Clojure at this conference, and um I'm just wondering like whether, as far as I know, you don't have a Clojure or a Lisp background, even though Clojure and Lisp, or at least Lisp, was the original AI programming language.
00:02:44
Speaker
um But ah I'm wondering what your connection is to like this conference and how you got invited. Not that it's a gatekeeping thing here. I'm very happy to have you there.
00:02:57
Speaker
But not how did you arrive at this conference? That sounds that sounds very bad. We can cut this bit. No, it's it's a good question. i also ask myself because as as I am not a programmer and actually I did not about ah did not know about Closure Language before I was invited. So yeah, it's a good
Cross-disciplinary Conversations and Ethics in AI
00:03:18
Speaker
question to ask. um I think that the organizers were looking for a talk that brought a bit of a critical lens to the conversation and could zoom out a bit from the technology into more of the political, the social and the philosophical implications. So yeah, it was actually Frank Bennett, who was the former director at the Ada Lovelace Institute,
00:03:37
Speaker
who connected me ah with the organization. And yeah, I'm so glad I could make it because ah we don't really get a chance to talk and converse across disciplines.
00:03:50
Speaker
But I think it's key. And I mentioned that in the talk as well. We do need to work collaborative more and more. So yeah, I was delighted to be able to be there. And I think like as technologists, like we really need to hear this message because I think we sometimes can fall into the trap of like, I've got an interesting technical problem in front of me. I'm going to solve it.
00:04:11
Speaker
I have no idea what this thing is going to be used for and I'd rather not know. So i think that message, like with me, it it landed so well. Yeah. I think on that note, just to sort of follow up a little bit is that I think in technology, there's technologists have kind of long had this like dividing point where they say, Oh, you know I will, I will not work for military, um, military software. you know, I won't do like missile guidance systems or,
00:04:40
Speaker
you know Some people will, some people won't. It's a fairly sharp dividing line, I think. you know Those kind of ethical considerations are they're felt in the developer community. But I feel like this kind of discussion about like the ethics the ethics of AI, the ethics of what we're doing as technologists is still buried quite deep in just like you say, algorithmic um considerations or is this an interesting problem?
00:05:09
Speaker
So, yeah, I mean, that was really an interesting thing for me, Anna, about your work and your talks were about zooming out and thinking about, well, what's the purpose of this stuff and what's the context we're operating in?
AI's Impacts and Constructive Pathways
00:05:23
Speaker
So, yeah, maybe as you could give us a ah quick recap. um We don't want to go through the talk and really to get all the points, obviously. the The stuff is streaming on YouTube. So, you know, all the kind of stuff. and we'll We'll drop a link to to that talk in the show notes.
00:05:39
Speaker
um yeah But if you could just give us a quick recap of the of the major points, that would be great. And then we can we can take it from there. Yeah, exactly. And as you were saying, I think a key point was that today it doesn't work anymore not to think that you're working on a narrow ah problem that you can just focus on that and not worry so much about it.
00:05:59
Speaker
the other connections. know Today, writing a line of code can have many implications because we live in such a data-fired society that you know the data you use and the assumptions you make or others make on your behalf on you know what's the equation or what are the variables that you are including in an algorithm can really have huge implications beyond whether a pipeline is working or not. So I think i think that was that that was a key message I was hoping to to give. And yeah, overall, the objective for me was to to step away from this kind of, I think, a haze of hype that we are going through. And maybe maybe now we're getting out of it, perhaps, hopefully.
00:06:37
Speaker
But yeah, I think this is kind of clouding or has been clouding our knowledge and and our attitudes towards AI. ah So I was hoping we could have a more grounded conversations on you know data-driven systems um cut through the narratives that serve certain powers very well. and And for that, I had to focus quite a bit on current risks and harms. And I worried that maybe gave a pessimistic tone to to the talk, but I thought we had to do that so that then we can focus on the positives, snow so we know what what is it that we need to avoid. um
00:07:08
Speaker
So, yeah, I went through some of the current risks that we've seen in society, largely related to using algorithms to make decisions related to social services, um which have had huge implications. I gave a few examples from the UK, the Netherlands and Australia on, you know, um subsidies, childcare benefits, and yeah, algorithms used in decisions related to who gets a mortgage or not, who gets a loan or not, um and and the huge harm that they caused.
00:07:39
Speaker
But then um i focused a little bit ah towards the second half on more constructive pathways. for the way forward. So I hope i i hope that also ah was received because I do think that there's potential and therefore if we put our energy in this potential, then we'll make a better use of these technologies.
00:07:59
Speaker
Yeah, absolutely. And that did come through very well on that. And I think just to jump off what the yeah one of your points that you made, you you said like you hope it didn't come across as pessimistic.
Regulation, Public Participation, and Corporate Responsibility
00:08:11
Speaker
like Honestly, I hope it did, because that is the reality we're living in. And and one of the things I feel is, you know you mentioned a kind of hype bubble.
00:08:21
Speaker
you know I just wanted to to kind of point out the fact that AI itself, artificial intelligence, is a very broad marketing term. There are many, many different technologies that fall under that broad umbrella.
00:08:33
Speaker
you know Everything from machine learning models that are very tightly focused on a specific problem, like finding the optimal route to deliver freight or something. Now what we're seeing is large language models, which can be used to both understand unstructured text like human language, but also generate it.
00:08:52
Speaker
We should be worried about what are the risks or what are the harms rather that AI is resulting in right now Because there are very real you know negative social impacts, you know some of which you pointed out.
00:09:09
Speaker
So when we say AI, I think as technologists, we need to be very clear about what we mean by that and um you know what the applications actually are. I guess what I'm trying to get at is, what do you think we should think about when we hear the word AI?
00:09:27
Speaker
like How would you like people to think about it? I think forget about robots and sci-fi, um even even if you know they they are a part of it. But the the real harms, as you were saying, are about water consumption, energy energy consumption, how they are worsening our um hopes and you know to to reduce emissions on the one hand. And then the societal harms that we're discussing about um you know how, ah for example, algorithms include
00:10:01
Speaker
prejudices and biases in the data and therefore ah can be used to make decisions that can have really serious consequences for people being able to access social services, jobs, etc. So I guess ah we should think about AI as a system of technologies that currently they use vast amounts of data. It's true that they are doing impressive things that we had not seen before, but they are ultimately technologies designed, developed, and applied, and made sense of um and evaluated by people, by humans.
00:10:36
Speaker
um So therefore, they are not neutral. They are not intelligent in the sense that you know they can make ah safe ah and objective decisions.
00:10:48
Speaker
um So we just need to look at it as a tool um that can be used for good or bad depending on how us as people use it. So one um following up from that, Anna, is um in like classical like technical ah tech technological breakthroughs like um like gene modeling and CRISPR, stuff like this, um embryonic technologies, biology, all those kind of things, one sees one sees a kind of this feeling that people are becoming God.
00:11:19
Speaker
And that that somehow there's a sort of reaction against that and regulators and governments generally get together and decide that there should be some limits imposed on the application of those technologies.
00:11:35
Speaker
and even some limits imposed on the research. um These are broad you know societal problems that that your you and your organization are trying to address.
00:11:46
Speaker
What kind of things or how do you think society and the world should respond to the implications that ai is being used for potentially and problematic things you know that where people can be harmed or where where the technology itself is regardless of all this you know it's going to destroy all human all human beings i don't think that's reality but no the current situation maybe be but not in the way we think pardon me I said maybe it will, but not for the reasons everybody's afraid of.
00:12:21
Speaker
Okay, maybe, yeah. But ah yeah, but but given that given the situation that we have, and like you say, there are fairly glaring problems with the technology, modular even modular, the energy use, you know the actual application of this thing. How do you think we should respond? And maybe it's you know talk about some of the tools we have as a society to respond to to the situation we see in front of us with AI.
00:12:46
Speaker
Yes. I mean, I think one key thing that I mentioned is, of course, regulation. The problem is that the technology is advancing at such a pace that governments are struggling to keep up, but they are, they are, and and we ah we now have a...
00:12:59
Speaker
EU AI Act, which is not perfect, but it's it's something. And it's identifying high risk uses and putting mechanisms in place to mitigate them.
00:13:09
Speaker
But we also need um a lot more processes of accountability, evaluation, auditing, ah throughout the life cycle of technology, even at the initial stage of you know asking the question, what problem do we want to solve with this technology? Who is involved in asking this question and and who's involved in the design?
00:13:30
Speaker
So another thing I mentioned in the talk is public participation. And sometimes there's a reaction to that, you know like, oh, the public don't know. It's too complex to explain. But I think we now have enough evidence um from fields like democracy deliberative democracy, for example, which consists of bringing together representative mini-publics, so randomly selected groups of people representative of a society to first learn about the topic, um then um deliberate, aid converse, and make recommendations about you know how something should be used for what.
00:14:07
Speaker
So public participation, I think, is key. and we In my previous role at the Ada Lovelace Institute, we did a review of all the, well, not all, but kind of a review of the literature that we were able to identify through our research that related to public participation. And people also want to be involved. It's not enough to yeah do a survey or to just be consulted. They they want to be part of decision-making processes.
00:14:34
Speaker
And people do understand and have quite nuanced views about the technology once they learned and have a chance to really be involved meaningfully about it. So I think that that's really important as well. How do you think we make companies accept that there are other stakeholders beyond their shareholders and investors?
00:14:54
Speaker
I think that's a crucial question to me because it feels like we can we can talk to the at the side about public participation and government regulation, but yeah what what how do we stop um like people like sam altman for example um from essentially just going his own way and not respecting any of this um democratic um know the sort of institutions and processes that you're talking about i'm not trying to be too pessimistic either but i hope that there are ways around this yeah but i think it's an interesting you know it's it's it's the elephant in the room isn't it really um
00:15:34
Speaker
Yeah. So, of course, there's the side of enforcement and you can, for example, in the procurement processes, have clauses that require that a technology has gone through this process before it can be purchased by a local authority, for example, and has gone through a series of testing and assessments, etc.
00:15:53
Speaker
Then there's the culture, that then there's the within the organization's kind of movements. you know um Isabel Ferreres, who's a professor at UC Levin University, has done work on the democratization of companies and organizations. And of course, we all know it's it's not easy. it's you know It's a lot of power that's held in these companies, but that's another avenue, workers, um I guess, mobilizing.
00:16:21
Speaker
and reclaiming their kind of rights and ethics in in how they do their work. and I guess then it's also helping organizations understand beyond enforcement ah the benefits of including people. I know this may sound a bit naive, but there are movements and work done by different organizations. I was, for example, part of a task force led by PAI, Partnership on AI, based in the US, that brought together a lot of people from different countries to put together some guidelines for private sector workers on what tools are available to that are not radical, but you know just ways to embed participation in the design um and the development of technologies.
00:17:05
Speaker
So yeah, it's many fronts, I guess. There's no like one way. um There's also conversations about, for example, having a global people's assembly on data and AI. There was one on climate and environment, two, now three, maybe years ago.
00:17:18
Speaker
And now there's been a few conversations on umm doing that. So there's both, you know a lot has to come from the demand side, I guess, and as well from enforcement. Yeah.
00:17:29
Speaker
And i you've mentioned the word democracy a few times. And i i mean, we got to go into what that word means, right? Because um I think all too often when people hear the word democracy, they think of you know our kind of standard parliamentary or representative liberal democracy where it's like,
00:17:47
Speaker
If I get to vote, then we have democracy. And I think we really need to reclaim that word. And, you know, in the closure community, we have a tradition of defining words.
00:18:00
Speaker
I'm going to do a little etymology here. And and democracy comes from two Greek words. One is demos, which means people. And the other is kratos, I believe. Yeah. Greek listeners will be mad at my pronunciation.
00:18:14
Speaker
But interestingly, kratos does not mean rule necessarily. That's one of the things that can mean, but it can also mean force or power. And so like to me, democracy is the power of the people.
00:18:26
Speaker
And that doesn't mean only elected representatives. That doesn't mean as as both of you pointed to the C-suite of a company or the board of directors or, you know, Sam Altman, you know, or other kind of Elon Musk, you know, these industrial titans, if you will.
00:18:43
Speaker
So like, what does democracy actually mean to you? Because you kind of mentioned democracy in the workplace. You mentioned these people's assemblies. So like, how do we get real democracy?
00:18:55
Speaker
Wow, yeah, of course, that's a huge question. But yeah, I think you're right that we've reached a point where it is clear to many people, definitely people working in academia in these fields, as well as many practitioners, activists, civil society organizations, that representative democracy is just one type of democracy. It might be the least...
00:19:17
Speaker
the least worse, um but it's still not good enough, especially in today's world. The problems we we have are too complex and too interlinked.
00:19:28
Speaker
And we know that the interests that are driving political parties are not working to solve this complexity. For example, yeah, with the climate crisis, there has to be decisions and policies that are quite radical in terms of how people move, for example, around a city, or just changes that we just have to make if we want to to meet our targets for zero emissions.
00:19:55
Speaker
But They are so radical and they are difficult to explain that so often politicians don't dare to take them. There's also, of course, lot of interest and lobbies from um industries um that that have a big role to play.
00:20:11
Speaker
But yeah, these kind of four or five-year cycles ah don't work because politicians want to just stay ah for the next cycle and rather not implement something that might mean they're kicked out. So yeah, deliberative democracy is a type of democracy that you know is really now quite common. We're going through what someone called the deliberative wave. So lots of governments, local authorities and civil society trying to implement these these approaches.
00:20:35
Speaker
So for me, democracy is really about that, how we together ah can make informed deliberative decisions about what societies we want and how they should be governed.
00:20:49
Speaker
And it means also doing it in ways that are inclusive, because we know a lot of people can vote or don't vote. And there's many reasons why sectors of society are left out of so-called democracy.
00:21:01
Speaker
But yeah, we can also apply that to how decisions are made in private areas. companies, in organizations. so So you can talk about democracy in terms of how we govern ourselves ah in a country or in a region or in a village.
00:21:16
Speaker
But you can also talk about it in in terms of how we make decisions in organizations.
Tech Workers, Democracy, and Data Stewardship
00:21:21
Speaker
I think that's a great point, actually, because um it feels to me, at least, that but We see this, that there's a democracy in certain types of companies in the sense that there's a there's a sort of shareholder democracy.
00:21:37
Speaker
There's, um you know, investors have shares in the company and depending on how many shares they have, they vote on board appointments or, um you know, compensation packages, etc.
00:21:49
Speaker
so So there is some sort of democracy in the corporate world, but Clearly, there are other stakeholders in corporations, you know, the people where the technology, you know, the technology affects the public or whoever the technology is used by, um as well as um the the workers who produce the the goods and services that a company makes.
00:22:16
Speaker
So I feel like one of the things that we want to talk about in Polytex is that obligation by the tech workers in these spaces. And it feels to me like they're a crucial player in this world where they're actually doing the work, they're producing the algorithms, they're doing all the data processing.
00:22:33
Speaker
And I would like to see but you know those people involved more and organized more to understand what the implications of their work are. And what is your as your think tank or your organization done anything about worker organization or what the implications of that could be?
00:22:55
Speaker
At the moment, I'm at the data time and no, we we don't focus on this. um The mission of the organizations is a bit more specific in terms of how data can be reused. And in terms of the human aspect, we do focus a big ah part of our work is on data stewardship and the role of data stewards in organizations.
00:23:17
Speaker
So, you know, we kind of, we do... focus a lot on the human aspect of who's going to be a custodian or a steward of this data and enable this collaboration so that data can be reused responsibly.
00:23:30
Speaker
um So we definitely think that building the human infrastructure is key, but we haven't focused so much on the aspect of workers' rights.
00:23:43
Speaker
Yeah. But I think that point on data stewardship is like a really important one, right? Because that you're talking about, i assume you're talking about data stewardship, both kind of on the regulatory level, but also like inside these organizations and companies that are producing and analyzing the data, right? So,
00:24:04
Speaker
I'm just thinking you know back to our conversation earlier in the Organizing Tech series where we talked to Per from the Swedish Association of Natural Scientists, and he talked about the role the union has to play in um pushing back against or combating discrimination, both gender and and race discrimination. And I think that part of organizing workers is often overlooked, that it it goes beyond making sure to secure our own working conditions. But also we need to think about, you know, like, you know, solidarity means that we are standing together for the good of all people. And I think like educating ourselves as as tech workers.
00:24:49
Speaker
on what responsible use and reuse, as you said, of data
AI Risks: Environmental Impact and Biases
00:24:54
Speaker
is. That is so vital. And then we need to use you know our organizations to promote these social goods. I think that, again, like going back to what we were saying, we we as tech workers often don't think about what the the what the code we're writing will be used for and what that means for the world.
00:25:14
Speaker
um And I'm just wondering, Ana, like we've we've talked a bit about the risks, but I think in your talk you highlighted what, seven or eight risks. um Do you like, can you just pull out, either walk us through them quickly or or just pull out the ones that you feel like you really want to highlight?
00:25:32
Speaker
Sure, of course. Yeah. Yeah. I can go through them briefly. So, for example, I started talking about the impact of AI-driven systems on rights, human rights and democratic integrity. um Talked about how there is that these technologies compromise the information environment.
00:25:50
Speaker
um Also subverting public health and consultation processes. We've seen some governments saying that they are consulting the public because basically They are scraping social media and saying that they know what the public wants, ah which is just the opposite of public participation. or You're actually surveying and throwing your own conclusions. Yeah, the impact on electoral processes and how it's impacting the ah political discourses. We see more and more power concentration and increasing inequality. So that all has consequences in terms of our democracies.
00:26:25
Speaker
Then I talked about the climate climate emergency. and the huge amounts of water and emissions that data centers and running these technologies requires, and how actually um not only Big tech companies ah have already acknowledged that they are not going to meet their targets in terms of emissions. But actually, there was an article maybe a month ago now that even the numbers that we had available are really underestimated. yeah So it's's yeah it's not looking ah good. And i what when i what I was saying is maybe...
00:27:02
Speaker
You know, there's ways that we can use AI driven technologies to protect biodiversity to help mitigate a climate crisis. But at the moment, I'm not seeing enough of how this is possible. So, you know, if it's possible, then we should put our energies there.
00:27:18
Speaker
Yeah, just on that point, sorry, about like how we're underestimating the the energy and water consumption. one One really interesting yeah thing I heard is that in these um ah computations of of kind of carbon emissions and energy needs,
00:27:35
Speaker
ah One thing that's missing from that, we just do not have data from NVIDIA, which is the the maker of microchips. They're powering all of these, um you know, the the actual hardware that's being used to for these machine learning systems and large language models.
00:27:53
Speaker
They just do not produce their data on energy use and carbon impact on the manufacture of those. um And yeah, so anyway, I just wanted to to bring that up.
00:28:05
Speaker
um But sorry, i yeah I derailed you a little bit there. No, exactly. That's the thing. We are also lacking a lot of data and information about and, you know, ways to really calculate accurately the ah scale of these impacts. Yeah.
00:28:19
Speaker
um Yeah, another impact I mentioned, which we talked about earlier, is about economic and social inequalities. And that's you know related to what I was ah mentioning on, for example, decisions on who gets a mortgage, who gets a loan, who gets a job, and how...
00:28:36
Speaker
um decisions um and data that's biased play a role in that. um For example, I was just reading an article ah yesterday about a new research study that has found that um some of these AI systems are really discriminatory in terms of um advising higher, um giving higher interest rates um for black applicants um to home mortgages ah because they were labeled, black and Hispanic borrowers were labeled as riskier.
00:29:11
Speaker
um So this is just just an example of many prejudices that are playing. Automated redlining. Yeah, exactly. Yeah, and then I, yeah, bias and discrimination is of course related to that, you know, and um what data goes in these systems, how where it comes from, what languages are included.
00:29:33
Speaker
right right talked about the work of um Joy Bulamwini, who's done yeah really you very important research on, for example, how Some systems do not recognize, for example, black faces, just something as basic as that. So, yeah.
00:29:51
Speaker
And you showed the, you she has a really great video, right, where she does kind of the spoken word or or poem, like demonstrating this. That was very powerful. Yes. She's ah an artist as well as a researcher. And, yeah, I agree. It was so powerful. So I think, the yeah, the website is poetofcode.com if anyone wants to check it out. Joy is amazing, yeah.
00:30:12
Speaker
I also talked a bit about their risk to a human identity agency on autonomy and, um you know, just asking, encouraging everyone everyone to ask the question of what it means when our relationships are more and more mediated by by these systems.
00:30:27
Speaker
For example, in the research we've done with the public at the Ada Lovelace Institute and other research that we synthesized through a review of the literature we did, people are not necessarily negative about different AI uses, but they do worry that some applications, for example, in health care,
00:30:45
Speaker
or in other important aspects like the delivery of social services that that they will lose the human touch, um the ability to explain themselves, um you know and the nuance behind everyone's context and case.
00:31:00
Speaker
um So yeah, what what what are the implications for that? And then I talked about geopolitics and colonialism. yes
00:31:11
Speaker
Definitely unwrap that a tiny bit, please. Well, I, for example, quoted Keoni Mahelona, who's the chief technology officer of Tehiku Media, who's doing a lot of interesting work using AI and technology in very positive, I would say, and interesting ways um to to revive and make sure that the Maori language is representative. um And it's um used in a way that's co-owned by the community.
00:31:42
Speaker
um and And he has this sentence that says data is the last frontier of colonization. Because if we think about the type of data, the language of this data that's used, the where the power and these companies are located, it's really one-sided. like It's very homogeneous um in in terms of how much the world is represented.
00:32:10
Speaker
So that that's something. And then i um I attended a very interesting talk by some academic academics ah from different countries in Africa. It was South Africa, Kenya, Ghana, Nigeria, I think.
00:32:22
Speaker
And they were talking about different philosophies that are quite common in different African contexts. One that we hear more about in Western countries, Ubuntu philosophy.
00:32:34
Speaker
But it's not the only one, but it's this idea that we are... people because of how we relate to each other. So it's looking at the individual as relational to everyone else. So it puts a lot more emphasis on community rather than the individual.
00:32:52
Speaker
And they were saying, first of all, these systems are threatening these worldviews because they are encouraging very individual engagements mediated through technology. But they were also saying that actually, they could be used to inform how we govern and regulate these technologies. you know or how we design them in a way that center the collective and social justice and how we relate to each other.
Data Provenance and Ethical Data Usage
00:33:16
Speaker
So, you know, we can learn a lot from other parts of the world, but at the moment, um the trend as led by big tech companies does not really allow for that, as well as, you know, how workers in countries like Kenya are doing a lot of the data labeling in conditions that are just not not acceptable in terms of workers' rights.
00:33:41
Speaker
So yeah, there's a few aspects to that. Yeah, ah we were speaking a little bit about um data stewardship to go back to the risks again. did what What about the sort of, you were talking about like the provenance or the inability to answer questions about, you know, why do you you know, why does the AI think like this? Or what is it?
00:34:03
Speaker
You know, what is what are its sources for um for coming up with a particular answer? um Does the data stewardship part also talk about the gathering of data, the so you know the kind of and non-consensual gathering of data from from the public, but also from from from artists, from authors and other creators?
00:34:28
Speaker
create Yes, exactly. This is an area that we are also developing. The Data Time is very new organization, but we are definitely looking into the implications of data stewardship and data reuse in the context of AI and data the new technologies that, as you say, are just gathering vast amounts of data from anywhere without you know any concerns about you know whether they are licensed, whether they are open, whether there's any permission from those who might be holding that data to for it to be used. So yes, data stewardship in the context of AI also means a responsible use in terms of do we know where the data comes from? How has it been gathered?
00:35:11
Speaker
With what what we call a social license, you know, what what type of permissions or collective consent um or public participation in the process.
00:35:23
Speaker
So this is, yeah, AI is is definitely pushing the boundaries of maybe what the social contract was about what was open data, what was not, what could be used. Right, right. Because I've heard i've heard that if you if people were fairly compensated for the data that AI companies have collected, that it would become, you know, apart from it's uneconomical on its face with the hardware and like energy investments,
00:35:51
Speaker
um given the lack of benefits that we see in the world in general um know economic benefits i mean um then if if people were forced to be compensated or if the companies were forced to compensate their owners then it would also make it even less economical which to me is not like a squeal from the company saying oh you're preventing us from like doing our business, it's like, yeah well, yes, we are preventing it because it's incorrect. It's unethical. it's
00:36:22
Speaker
You're literally stealing um content and not fairly not fairly compensating people. So it feels like the business model of stealing data is not sustainable.
00:36:34
Speaker
yeah i mean I'm trying to think about, like in terms, again, about like what what the, like I said at the beginning about Whether we feel like military um military technology is an ethical thing or not.
00:36:49
Speaker
you How do we feel about the fact that we're making business, that we're doing business on essentially stolen data and stolen property? Yeah. And just like really quickly on that, Ray, like even without compensating um people for the the data that you're using in these systems, um there was just a ah paper like a month or two ago by this left wing think tank called Goldman Sachs.
00:37:17
Speaker
And they basically they basically said, look, this will never be profitable. Like, how is this going to generate a profit?
00:37:27
Speaker
Like, how is this ever going to generate return on investment?
Challenges and Opportunities in Data Labeling
00:37:30
Speaker
So I think even if we like, you know, even if we play into this neoliberal idea that corporations exist to maximize profit for their share shareholders, and that's their only responsibility.
00:37:42
Speaker
they're not even behaving responsibly in that. So, you know, but I'm, yeah, let's, let's definitely follow the thread of like this idea of, um you know, how data stewardship in terms of, well, there were two ah kind of phrases you use that I I'm hoping you can explain a little more. So one was,
00:38:04
Speaker
data labeling and the second one was data reuse. Can you kind of explain what you mean by those two? Because those feel like kind of technical terms of academic research, perhaps.
00:38:16
Speaker
In terms of data labeling, maybe there's that's not the right word. it's in terms of how, yeah, the one that's actually used in the context of um tagging, tagging the the data so that, you know, like, ah okay, that's going to be woman, that's going to be man, that's going to be dog, you know, um so that the tagging is very important, I understand.
00:38:39
Speaker
um I have no technical background, but in this developing these systems and a lot of the this work um yet is being done at the moment, but a lot of people based in countries where maybe there's less regulation or some big tech thing, they can pay less or outsource and pretend that they don't know what's what's happening.
00:38:59
Speaker
Isn't the labeling in that case, and just to sort of jump in a little bit, is that it's about whether it's like safe for the public. So those workers are subjected to watching child pornography. All those kind of like that all the worst things on the Internet that is meant to be essentially.
00:39:17
Speaker
kind of like You would imagine the AI should filter it out, but in fact, these people are the ones that are doing all that sort of traumatic work to support the the the safety of the technology.
00:39:31
Speaker
yeah it's Yeah, sorry, Anna, go ahead. No, was just going to say for AI to be able to filter it, it needs to learn what what's what. So that think that's where that tagging comes in. yeah Yeah, and the really interesting thing, so OpenAI, i of course, the company behind ChatGPT,
00:39:46
Speaker
It turns out they have two large language models. So one is the LLM chat GPT itself, and the other one is the filtering mechanism. So they have a separate machine learning algorithm that governs the filtering. And and that's exactly as as Ray was talking about there. And as you said, on ah um having human beings,
00:40:08
Speaker
look at the output of of the unfiltered chat GPT to label what is safe, unsafe. um So yeah, I mean, that is, yeah, horrible and very exploitative in terms of like who is actually doing this work.
00:40:26
Speaker
Yeah, on the other end, data reuse, ah in the way that I was referring to it and the way we refer to it at the data time is in a positive way. So we think that there is a lot of potential of using data responsibly, sustainably, for informing policies, understanding, for example, how can we move better, more sustainably within a city or, um i don't know, improving our biodiversity. There's so many potential uses.
00:40:53
Speaker
So yeah, we kind of want to reclaim ah data and reuse it um so that it can benefit all of us um rather than just specific interests.
00:41:04
Speaker
ah So that's what I meant by reuse. And that's where the data steward comes in, in every organization, which is is about partnerships. A lot of it is about understanding where where's the supply of data, where's the demand of data, and how can we collaborate to make sure that um this data can be used reused responsibly yeah
Positive Uses of AI and the Importance of Quality Data
00:41:25
Speaker
mean i think to be honest i mean i could probably go on a whinge fest with you um you're you're you know you're talking a lot of uh tremendous amount of sense and uh i think this is fantastic
00:41:39
Speaker
um you know, like we could just carry on. I mean, I could complain about AI all day, but I think the other aspect of your talk, which was light relief, and maybe we should bring that around, I think, at the end of this conversation, is that potentially there are some positive impacts from the technology.
00:41:57
Speaker
You know, if we if we, in the same way that people say, know, gamble responsibly, drink responsibly, maybe we use data responsibly as well, you know. So maybe you can talk to that a little bit, Anna, about how we can like be you know how we can use this ah in a more positive manner to get better outcomes for for the public.
00:42:20
Speaker
Yes, I think ah for me, um as far as I understand, considering you know my non-technical background, is that narrower models that use data, that we know what the quality of this data is, where it comes from, can also be used to target at very specific problems.
00:42:37
Speaker
And I think that's where most of the potential is. I mean, we've just had the news last week that the Nobel Prize in Chemistry has been awarded to David Baker from University of Washington, as well as to people from Google. did mine But what they i did is used and developed an AI model to some to solve what it seems was a very old problem, which was our capacity to predict proteins' complex structures.
00:43:01
Speaker
Right. And it seems this discovery has enormous potential in, you know, can neutralize viruses, target cancer cells. um So it I think use is in medicine, in advancing science and hopefully helping also in our environment. I think that's where I see the most potential. But The common denominator, I think, is that these models have very high quality data.
00:43:29
Speaker
We know what data, we know where it comes from, and we can explain you know and the the process and and the model. So I would say these are, yeah. So that's data provenance, I guess, is the the word we usually use to talk about like knowing where the data came from and and its quality and so on.
00:43:48
Speaker
Exactly. One super interesting thing you just said on and this is why I push back so hard against the umbrella term AI. What you just said is like the best uses of quote unquote AI are on very narrow targeted models.
00:44:06
Speaker
So all of the things you just said, I believe large language model sorry large language models cannot do. I mean, they are, as Emily Bender said, stochastic parrots, right? All they do is predict what comes next or, like I said, extract information from natural language.
00:44:25
Speaker
So they they can't do things like, you know, understand protein folding, right? Or understand how to most efficiently generate and distribute energy. They just can't do it.
00:44:38
Speaker
Yeah, and and interestingly, these models are the ones that I think have played a big role in this hype that we were talking about in the beginning because I think they enter the public sphere and everyone's been amazed that some of these systems can help them, I don't know, um answer questions they have at work or use them as ah browsers, as searchers. But the thing is, they are not they they can't really because you can't really trust the answers. so that yeah No, exactly. That's a fundamental problem. Yeah. That you can't, yeah the hallucinations are a fundamental problem.
00:45:11
Speaker
Yeah. Yeah. Anyway, back to optimism. Yeah. Somebody also described them as a party trick, a really, really, really impressive party trick. but That's all it is.
00:45:23
Speaker
Yes. Yes, exactly. So, yeah, I think that is potential, but as you say, we need to differentiate the type of models and how they are, fed what data they are fed with, et cetera. Yeah.
00:45:33
Speaker
Yeah. So Ana, I realize we've been going on for nearly an hour here. And as Ray said, like, I wish we could do an eight hour interview with you and just talk all day, you know, but I think, uh,
00:45:45
Speaker
Both you and our listeners may ah you know find
Technologist Responsibility and Podcast Conclusion
00:45:48
Speaker
that a bit tiring. So if we if we kind of move toward the end, I really want to give you the last word. And and you know if you can speak directly to technologists, which I hope you're doing on this podcast, I think are our audience would largely be people who are working in ah the technology sector.
00:46:07
Speaker
If you could have them just take one lesson from this interview or one lesson from your work, like what do you want us as tech workers to do? I don't know if I'll be able to submit in one thing, but let me try. i think I think it's that there is power in all of us making sure that we use data and AI systems in constructive ways that a line of code might seem unoffensive, but you know actually it can embed ah prejudices in the data. It can embed opinions or to frame it positively, it can maybe do the opposite, enable more positive, constructive and collective um outcomes and and ways of engaging. So, yeah, I hope that people don't feel like ah pessimistic and apathetical about, you know, the discussion, but on the contrary, that, yeah, there's power now in every one of us because we live in such a connected and complex world that the only way forward for me is is collaboration and all of us playing a role in our little kind of
00:47:08
Speaker
bubbles, but yet, you know, having to be connected Wow. Yeah. I love that. Yeah. I love that. Yeah. So yeah, Ana, again, thank you so much for your time ah today. I really enjoyed this conversation. I'm sure that listeners are also going to be, you you know, both educated and also hopefully motivated to to take this action that you're asking of us. So um thank you so much again. And i hope to speak to you again. You're always welcome to talk to us about whatever you want.
00:47:39
Speaker
Thank you so much. I had a really good time. And it's very important also for people like me in the social sciences to be able to to converse with different disciplines. So I really, really enjoyed that. Thanks. Likewise.
00:47:50
Speaker
All right. Have a wonderful day, Ana, Ray, and everyone listening. If you've enjoyed this slight departure from our normal programming here on Deafin and want to learn more about AI, Ray and I have got you covered with a brand new podcast called Politics.
00:48:06
Speaker
Season one is all about AI. What is it? What should we be excited about? What should we be worried about? You can find the link to subscribe to the feed right here in the show notes or head over to politicspod.com or just search for politics and all of the normal places you find podcasts.
00:48:27
Speaker
Hope to see you there.
00:48:32
Speaker
Thank you for listening to this episode of Deaf Anne. And the awesome vegetarian music on the track is Melon Hamburger by Pizzeri. And the show's audio is mixed by Wouter Dullert.
00:48:43
Speaker
I'm pretty sure I butchered his name. um Maybe you should insert your own name here, Dullert. Wouter. If you'd like to support us, please do check out our Patreon page. And you can show your appreciation to all the hard work or the lack of hard work that we're doing.
00:48:59
Speaker
And um you can also catch up with either Ray with me for some unexplainable reason. ah You want interact with us, then do check us out on Slack, Closure in Slack or Closureverse or on Zulip or just at us at deafenpodcast on Twitter.
00:49:18
Speaker
Enjoy your day and see you in the next episode.