This page features computer generated text of the source audio. It may contain errors or omissions, so always listen back to the original media to confirm content. You can search the text using Ctrl-F, and you can also play the audio by clicking on a desired timestamp.
Kia ora. Today I'm talking with Gareth Watkins, the founder and director of Pride NZ. Gareth, can you tell me about the origins of Pride NZ and a bit about what its aims are? Pride NZ started in 2009, so just over 15 years ago, and it has three aims that haven't changed since it started. The first is to be a first hand eyewitness.[00:00:30] community resource to allow as many people to hear rainbow experiences from Aotearoa as possible. The resource, uh, is to be, uh, inclusive, uh, hopefully, uh, promoting health and betterment for, uh, the all rainbow people, um, and, and a community resource that is not for profit. Uh, so it's been running 15 years, 15 plus years. And there are now over 1, 000 audio recordings on the [00:01:00] website and the audio recordings range from interviews to recordings of community events, uh, to, um, to recordings of, of things like, um, the annual AIDS candlelight memorial. You've developed a relationship in recent years with the National Library. I wonder if you can tell me a bit about how the relationship came about and how it's managed. Well, it was, uh, it was just something that [00:01:30] happened quite organically in, uh, mid 2021. I received an email from the Library of Congress in the United States saying we would like to, uh, we would like to to take a capture of the PrideNZ website. So keep a record of the audio and, and also the information around the audio. And I was just blown away. I mean, this was the first time that a major institution had really been interested in [00:02:00] taking a snapshot or capturing the PrideNZ collection. Uh, for posterity, uh, and so we put out a press release, uh, Pride NZ put out a press release and, uh, from that press release, um, I got some approaches from the National Library and the Alexander Turnbull Library, um, wanting to, to have a further discussion about, um, whether a archive in New Zealand, uh, could, or a public institution in New Zealand could, uh, also hold on to the [00:02:30] material. And I just thought that was really, uh, amazing because I think it's so important for, um, uh, our communities to be represented, particularly in state run, uh, organizations and institutions. So I was very keen. And so we, uh, began an ongoing deposit of Pride NZ material with the Alexander Turnbull library. Initially, it was around about 850 [00:03:00] ish recordings, and that was done as bulk, and that actually, I think for, um, any institution receiving that amount of material. Uh, but one of the really nice things now is that, uh, we have ongoing deposits. So every month I'm depositing maybe, you know, three to four recordings, uh, into the Alexander Turnbull Library. So the accruals keep happening. I think it becomes a lot more [00:03:30] manageable for the, um, the institution receiving that material. Because they don't have to have a like a special project to, to catalogue it. Um, and it just is a, a wonderful incentive to, to keep going and doing recordings, knowing that actually the recording you do that month, uh, will actually end up into the, uh, National Library, um, collections, uh, which are there in perpetuity.[00:04:00] And it's a case though of you making sure that you are pushing the material on a fairly regular basis, a monthly basis to the National Library rather than them coming to you every month and saying, what have you got? Oh yeah, absolutely. Absolutely. So, uh, I mean, it's, it's a very, um, easy relationship, I think, uh, in that, uh, I'm not, certainly not being hounded too [00:04:30] produce material, but I'm also, when I'm submitting material, I mean, it's being welcomed. So, uh, I think we've kind of reached a, uh, a nice balance where a good amount of material is going in per month. Um, it's not swamping anybody and, uh, and it's, it's ticking over, um, very much business as usual, uh, or certainly from, from the Pride NZ perspective. So what about, uh, other organizations? What sort of advice would you give [00:05:00] them wanting to establish similar relationships? I think having a Champions within the, uh, main institution or the institution that's collecting, uh, is key. Uh, someone that is a point person that you can go to and say, Hey, would you like this or would you not like that? Or, um, you know, yes. Someone who you can bring up issues if there are issues, um, there actually haven't been any issues, um, it's, it's, it's [00:05:30] been very seamless, but I, I think actually having somebody there that, uh, knows the collection, knows you, uh, is, is very beneficial. Okay, let's talk a bit about AI. When did Pride NZ begin its AI journey? And what were the key drivers to, that started you exploring these tools? AI has been with Pride NZ, or Pride has been with AI, or Pride has, Pride NZ has been using, uh, AI [00:06:00] for a good number of years now. I think probably the first use came in around 2020 when we got the audio recordings transcribed by a platform called Otter and that is probably the first time, uh, We used AI. So prior to about 2020, audio transcription was very manual. And [00:06:30] so there are a number of transcriptions on the website that were done by, um, a person. Uh, but at the time it was about one US dollar per minute of audio transcription, uh, when you're dealing with, you know, 800 plus audio files, um, that is just, uh, prohibitive in terms of cost. for, for Pride NZ, which is a very small organization. Um, and it wasn't until really 2020 that, uh, there [00:07:00] was, uh, an opportunity to actually use AI to automatically create transcriptions. The transcriptions aren't perfect by, by any means, but they give an insight into the content and, uh, And for Pride NZ it was all about really unlocking the door to the audio collections to make them more accessible, to make them more discover by things like Google or Bing or [00:07:30] whoever's indexing the site. Um, that was the key to, to, so the, yeah, the transcriptions, um, are, are always, um, uh, footnoted with saying, you know, please make sure you. Uh, go back to the original audio to confirm what is written, in fact, is what has been said, uh, because they are not perfect, but they are a very good way of, of opening up the collections for, um, indexing. Well, speaking of perfection, uh, or [00:08:00] not, and rethinking about the way that AI has developed at such a pace in recent months, even, let alone recent years, um, to what extent can they accurately transcribe, say, audio featuring Australian and New Zealand accents, or even te reo Maori? They're getting better. They really are getting better. I mean, I, uh, fortunately saved some, uh, transcriptions from, uh, the first time we, we ran that in 2020. Um, [00:08:30] and just last year, um, Christmas last year I ran the entire, uh, collection through another transcription software, uh, and I could compare the two and you could see within three or four years the, the, the, the absolute leaps and Um, ahead, uh, that, that the, uh, software is now currently making, um, or the AI models are making. Um, in terms of Teo, uh, it's still [00:09:00] not good. Uh, I think the best I've seen so far has been a, um. piece of software called Descript, uh, but, um, uh, not being a Trello speaker myself, um, but you can kind of follow along the audio with the text and see, you know, well, the semi accuracy of it. Um, It's a really interesting question about do we feed our content into AI [00:09:30] models so that they learn from us, or do we say no, we're keeping that, that, that material, um, and don't learn from us. I mean, I think for the Pride NZ material, and this is not necessarily talking about the, um, you know, Torreo aspect, but for generally for Pride NZ material, I Like the idea of feeding it into the AI models because then it's, then the AI models are learning, um, about rainbow [00:10:00] heritage in New Zealand, and I would rather have the AI model talking with some accuracy than speculation. So I think, uh, for the Pride NZ collection, which has always been public, so I mean the, the, uh, the, the, one of their biggest, The big ideas with Pride NZ is that, uh, we record material to be public. Uh, we're not recording oral histories that will be locked away in an archive for 10, 20, 30 years. Uh, it's, it's about kind of an immediate publication of [00:10:30] thoughts and ideas and events. So, um, I don't see it as a, as a, as a big stretch to feed that material into AI to actually make it public. AI better for everyone this speaks really speaks to accessibility doesn't it so tell us about how you're using say ChatGPT to further enhance accessibility of Pride NZ. I mean ChatGPT has been a It [00:11:00] has had an amazing growth in the last year. So I got my first ChatGPT subscription in November 2023. And what it could do then to what it can do now is absolutely amazing. When I got that subscription, I realized that actually, AI offered up to small to medium sized organizations like PrideNZ or any [00:11:30] smaller archives the ability to really, uh, open up the catalogs to, to really create like summaries, um, or, um, keywords for searching. Uh, and so the first thing, uh, we did with AI was, We, uh, created a new set of audio transcriptions of all the audio files. We then said, let's feed those audio transcriptions into ChatGPT. In fact, it was [00:12:00] actually OpenAI, who run ChatGPT, OpenAI have a, um, another way of accessing the AI models through their API. Um, and it allows you to do things in bulk. Um, and so using, uh, OpenAI, we basically went through and we fed in each one Audio transcription and got back a 400 word summary of the content that could never have been achieved [00:12:30] by Myself doing it myself. I mean it would have it would have been a lifetime's work I mean potentially you could have had a lot of volunteers doing it, but actually, you know, those summaries were coming back every 10 15 seconds per per recording. And when you're talking about recordings and summarizing recordings that are two to three hours long, like, where do you begin? Where do you begin? So the idea was that we would summarize all these, um, uh, all [00:13:00] these recordings, create summaries, um, and then that then feeds into things like, um, creating an AI chatbot, wouldn't it be amazing to be able to actually have a discussion in conversation with a, a chatbot about, uh, Rainbow New Zealand? And rather than, um, Diving into all of those transcriptions, which is very kind of, um, resource [00:13:30] hungry for the AI. Um, the AI can actually use the summaries that have already been generated to formulate its answers. And so, every step of the way, you're, you're opening up the opportunity of AI to, um, take, um, what you're thinking a step further. So this opens up a whole world of trust and questions of authenticity, doesn't it really? So, um, how do you identify AI [00:14:00] generated content for users of your website? I think the key is to be very open and transparent and clear so at the bottom of each summary on the website there is a sentence that says this has been generated by generative AI and here's a link to the prompt that we fed the AI model. So you can actually see how, [00:14:30] um, the information we asked the AI to, to consider and, and how to, um, how to phrase things. I think the big thing for the Pride NZ website is when it relates to pronouns. So we don't, we don't have a list of. all the people on the website and their pronouns, uh, and their pronouns may have changed over time and may change over time. So, uh, rather than [00:15:00] saying he or she or here in, in the description, uh, we've basically gone for a default of they, them. Um, and then. We can, um, change the summaries if people come to us, and some people have, saying, actually, um, these are my pronouns, uh, please update the summary, and, and very happy to do that. But I think that was the big thing for us, was, um, yeah, to, you know, be very, very respectful in, in, [00:15:30] in, in, in that regard. Um, so that's very obvious in terms of, um, the, the prompt that we've used, um, on other pages. So we also do a summary say of newspaper articles, um, very clearly says that, um, it is a summary generated by generative AI, um, and that may contain errors as may the transcription contain errors, but I. I would rather have the transcription with errors [00:16:00] online than have nothing online. Because I think you can always go back to the original audio which is also on the page and double confirm. And it should be up to the human to then go back and listen to the audio and go, Okay, that person did say that. Um, and, uh, yeah. So Rudy, you're using this as a finding aid rather than to tell the whole story necessarily? Yeah, absolutely. It's a, it's a, it's a great finding aid, but the really interesting thing and [00:16:30] the really exciting thing is that, can it be the whole story? Like the, the current developments in AI, for instance, with, uh, Google's Notebook LM, where you can feed in three or four transcripts and get it to auto generate a audio podcast of talking about the, the, um, the, the recordings. And so [00:17:00] at the moment it's a finding aid, but, but potentially it could be a way of creating content as well. What kind of audience feedback have you received so far? Very little, to be honest. Very little direct feedback in terms of anything to do with the AI. I mean, I think the feedback has been, please correct the pronouns. That's really the only thing that I've received. I know, for instance, if you do Google searches [00:17:30] on content or people that are mentioned in Pride NZ material, often Pride NZ is now at the highest level in terms of hits. Obviously, the text based version of the audio recordings is having a dramatic impact. If you didn't have the text based version, Google wouldn't even see those audio recordings. So I think it's certainly made things a lot more discoverable. Um, another [00:18:00] thing that has made things way more discoverable is having the audio recordings. content in Alexander Turnbull Library and on the National Library website, uh, because I am now getting requests from people that have discovered stuff, um, via the National Library website that may never have come to Pride NZ. So, uh, I think dependent, you know, um, the National Library has given, um, a new pathway for people to come to the Pride NZ [00:18:30] collection. So really it's, discovery is being enhanced from two directions, isn't it really? Through the relationship with it being archived elsewhere, but also through the efforts that you're putting in with artificial intelligence to make people be able Yeah, and it just comes back to that whole point of the original idea behind Pride NZ being that, um, we want to be visible. Um, you know, for too long our, [00:19:00] uh, you know, our stories have either been hidden or, um, obliterated, um, wiped out by either, um, uh, families or the state. And, um, this is a chance to actually say, well, actually, um, We are a very proud group of people. Uh, we've got amazing stories to tell. We've got amazing people within communities who show, uh, not only, um, resilience and [00:19:30] advocacy for themselves, but, but, but, but also for their communities. And, um, it's just such a, a rich, uh, vein of, um, amazing lives to be celebrated. Okay, so in addition to managing the Pride NZ website, you're also currently the Collections Data Manager at Papatongarewa. How have you been able to apply what you've learnt through the experiences with [00:20:00] Pride NZ and the website there to the context of the museum's collection? Really interesting, uh, really interesting question. Uh, I think first of all I would say that, um, Let's define what the collections are first of all to give it a bit of context. So the Pride NZ collection, very new collection, a very new collection. Uh, it's been around say 15 years. The collection in its entirety is meant to be public. It's, it's public facing. Um, the [00:20:30] vast majority of the content, uh, is either created or commissioned by Pride NZ. So, um, Pride NZ has a direct relationship with it. with those items. Whereas Te Papa, uh, which has been collecting for 150 plus years, um, not everything in Te Papa's collections is Or should be available to everyone. Um, there are good reasons why some things are restricted. [00:21:00] Sometimes we don't know the provenance of a particular items in the collection. And so it's a, it's, it's quite a different kettle of fish in terms of, um, Thinking about, kind of, rights, and ownership, and, uh, you know, is this a good thing to be putting into AI? So, I think the big thing from the experiences of the Pride NZ AI work, it's an [00:21:30] example of what AI can do with a heritage collection. I don't know if it necessarily translates into work that I would do directly at Te Papa, but I think it's a really good conversation starter, and I think it's a really good way of showing how AI can be used in a positive way on heritage collections. And what advice would you give individuals or [00:22:00] GLAM sector wanting to explore using AI tools themselves in association with their own collections? I think the big thing is just to initially play with AI to um, to feed it information that's already publicly available and see what it comes back with. To increase your, um, AI literacy. Uh, so, uh, that will hopefully, uh, reduce fears, [00:22:30] but it will also bring up challenges in terms of, well, you know, what, what, what are the rights implications? Um, what happens when we put it into this model? Do we want it to be learning from our data? Uh, maybe we do, maybe we don't. Um, so I, I think. Playing with AI and playing with a whole variety of different types of AI is, is incredibly useful. I would suggest learning, um, a free computer language, something like Python, which you can [00:23:00] easily learn online. Because one of the really neat things with AI, and particularly things like, say, ChatGPT, uh, it will help you code, so it will help you interact with the world. the different AI platforms, um, and a little bit of, um, coding knowledge using Python, uh, which is again freely available, um, can make development, um, really, really fast. I'd be wary [00:23:30] about developing In house models, uh, at this stage because I think AI is moving so fast that the large commercial platforms are putting a lot of energy and money into, uh, trying to outdo each other. Um, and I, my fear would be that you would, you would buy an in house model, uh, that would be out of date. You know, within a month [00:24:00] of, of, of having it, I think as an initial step, particularly for small to medium organizations, I would, I would do a subscription to a cloud based, uh, a commercial, uh, based product, say like, um, open AI or, or others say I like Amazon or Google, uh, because, uh, they have got the expertise to, uh, Uh, to, to generate those models, to, to maintain those models. Um, [00:24:30] whereas actually a lot of organizations don't have an IT department. Um, so, uh, yeah, I, I would personally go for a, um, more of a commercial based model to begin with. Um, yeah. Okay, so I'm a very small glam institution. What kind of investment am I looking at in terms of AI and upskilling? Yeah, I think a lot of, well, if you go down the road of, [00:25:00] Doing a commercial subscription to somewhere, so like, um, ChatGPT You know, you're talking 20 or 30 a month US to use to really kind of to get to understand how it might work If you're wanting to go down that kind of programming route where you can actually use Python code to [00:25:30] query, say, OpenAI. Well, we were doing an experiment the other day where we were creating descriptions of collection items for, it was just over one New Zealand cent per 300 word description. Uh, and those descriptions were coming back within 10 seconds. Um, I think A key thing is making sure that your [00:26:00] data, your original data, the metadata around collection items, uh, and say in, in this case with pride nz, um, you know, the record date, the location, the people that are in it, um, is all up to scratch because actually AI is only as good as the material that you feed it. So if you feed it, um, a lot of data with holes in. Um, then it won't be as, as good, but actually if you can, and this, this is a really fantastic thing for my [00:26:30] job, say at Te Papa, which is all about, uh, really trying to, um, encourage people to enter appropriate data into the appropriate fields in our collection management system. Uh, this is a great carrot because we can say, actually, if you put this data into EMU, uh, our collection management system, um, then. There are amazing possibilities with AI that we could, we could try. There are amazing possibilities. It's really exciting when you think about it. And just the speed with [00:27:00] which things are changing is extraordinary. So, in conclusion, what's next for Pride NZ, bringing us back to where we started? Great question. Uh, It's about getting our data in order, uh, first and foremost, because, uh, I think I'm very aware that actually good data will give good results for AI. So making sure that we've got things like production date. Who's speaking, [00:27:30] um, all that kind of core stuff about say collection items that you need, um, making sure that that's there. Um, and then it's just about being open to the AI possibilities that are being presented. Uh, for instance, the other day I was made aware of a notebook, Google notebook LM, where you can feed in. to up to 50 sources of information and it will create timelines based [00:28:00] on those 50 sources or it will create summaries based on those 50 sources so you know exactly where the source of your information is coming from and then it can do all these amazing things. Now for me that Google LM was completely new and who knows what will be around in six months a year. Uh, so that, yeah, that's really exciting. Um, the, the kind of the next biggest thing, uh, we're doing on, uh, Pride NZ [00:28:30] is we've just started summarizing newspaper articles that are on the Papers Past website. So these newspaper articles, uh, released under Creative Commons. Um, and, uh, the, the, uh, OCR, the Optical Character Recognition, um, is a bit variable on on some of the old newspapers, but one of the lovely things that creating summaries does is that it tidies that all up. That actually the [00:29:00] summary that AI replies with doesn't have any of the inaccuracies of the OCR. It's kind of got a good A, a, a gist of, of, of the article. Um, and it makes it a lot easier to read, a lot easier to discover. Um, and so currently, uh, we're going through looking at, um, rainbow related content, uh, from, uh, early New Zealand papers and, uh, trying to, uh, identify and acknowledge, uh, our rainbow history.
This page features computer generated text of the source audio. It may contain errors or omissions, so always listen back to the original media to confirm content.
Tags