Sven Christian [SC]: What were you working on at the time of your first residency? Was it Credo?

Carla Busuttil [CB]: So, I had a solo exhibition at the end of it. I produced a series of paintings and alongside that we were making videos. The idea for the videos came about as we reacted to the space. One thing we discovered was an old _________ on the farm up at the Koppie. We looked up its history and found that it was apparently Lord Milner’s outpost.

Gary Charles [GC]: We had studied Lord Milner back in school but couldn’t remember the depth of it, so we went and relooked. At that time the #RhodesMustFall movement was happening. We were based in Oxford where there was also action taking place, which we joined. It was interesting to observe that Lord Milner was the heir (an ideological heir, at that) and executor of his estate. And then we found out that Lord Milner had written a Credo which spoke of Britain being superior and more powerful and thus needing to civilise the world. The ideology is written explicitly. At the time, Carla was making papermâché masks which end in these hideous faces, so we recast Lord Milner in one of these masks and reperformed the Credo, almost like a reenactment.

CB: I was also doing a lot of satirical paintings of historical figures at that time…

GC: Polly just pointed out that Carla was making masks out of found shin pads.

CB: So, Polly actually was in one of our films that we did; one where we had come to the farm to work on the project that was a mosquito lightning where we did fictional private security adverts based on real life found adverts. In the video, we were all dressed in masks that were sort of cricket pads. [Polly was around 3years old.]

[There is interval conversation with Polly reminiscing the video shoot experience and living with the props at home thereafter and discussing art as a career]

[0:06:04] SC: So from Credo Mosquito, Lightning there’s this undertone of ‘empire sensibility’ but skewed in a way as a result of using material like cricket pads and masks. How did the mask come about?...Why was it a motif?

[0:06:20] Carla: I think it was a way to make the figures more neutral, to eliminate it from being about gender or race (this was for mosquito lightning). And then for Credo it was a way to create something grotesque- I wanted it to be a relationship between the paintings and the videos, creating a sense of connection.

[0:06:42] Gary: The original plan for the films was to make it look like they’ve jumped out of the paintings and it evolved a bit from there.

[0:06:50] SC: How did the collaboration come about?... when you guys decided to incorporate these different aspects of your work; Gary, you are from a sound/music background…At what point did that start becoming about AI and machine learning?

[0:07:04] Gary: So most of our work has been thinking about systematic or institutional injustices, and so a lot of the works, from; Credo, Mosquito Lightning and we did a bit of work relating to The Paradise Paper and (indistinctive) ([0:07:33] Carla: As well as landownership in the UK) …So, essentially the role of wealth and equality on both a global and a micro scale. 

I did my masters at Oxford Brookes producing mixed media works, installations, films and performance. But I originally come from a sound focused background. I ended up finding a supervisor in Birmingham who was really interesting, I then decided to work around the role of essentially the largest monopolies that have crept up in a short period of time. And the forefront of everything they were doing was driving this neuro-network based AI as the computation of the future. And so, a few of the large Tech firms also had projects in music. So, it was a perfect thing for me to try to understand and get involved in. Then overtime, we started finding avenues that could…in a way the AI projects in music are uninteresting, I would argue that that area is hype based and vacuous than the term AI in general which I think is more a marketing term than a description of the reality of what these projects can do. And so, we started playing with other areas and organically started working on some of the same things. During lockdown, Carla started working with more digital tools and so these ideas sprung up naturally out of working with those and thinking about the shortcomings of them and what they mean in the world.

[0:09:34] Carla: I think its also the way we are into taking these academic papers that you have been reading with really interesting ideas and trying to think of ways we can visualize them, so they become accessible for everybody to understand.

[0:09:52] SC: The one thing I wanted to ask you was around; I have a vision of what AI is, which is locked into minority report and that kind of film. But when you talked about AI and music being boring in comparison…what is that relationship? What is AI in music?

[0:10:15] Gary: So, most of the popular culture references and people see AI as robots, whereas, in the last 10 years when people are talking about developments in AI they are really talking about neuro-network based algorithms that have been around for a very long time and now fueled with our increased computation and data collection -capabilities have become much better at certain functions. Things like object recognition. And the way that that works is really because you have such a large database of images that can be scraped from the internet, you will have someone being paid very little to label pictures of cats for example. And that forms the data set by which patterns can be recognized that conform to a cat in an image.

In music, most of the software that is composing by AI – so what we would think as a more general Artificial Intelligence or actual creativity, where we say you can press a button and compose a whole piece of music. Generally, those pieces of software using media as a language to transcribe notes and rhythms in the music. And media is a language that is used in most of the music production and composition software that we used today, and in synthesizes, and so on. And so, it is basically reducing or transcribing music to simple notation, and then building up a huge database, then finding patterns in those. So, if you put in a basic melody, it creates a probability of what comes next based on huge training sept.  The results are not that dissimilar to the cassio tone, keyboards with a band in the box that were available in the 70s…But the question is; if we keep building up the training sets, will that become more sophisticated? will it become better at composing music? My argument is that it probably won’t, but that it’s also kind of meaningless – because for that to work, you must treat music as a settled object; so, something that is not changing anymore, and that the categories are set in stone. And even if we did have powerful enough computers and large enough data set then, we could turn out billion continuous streams of acceptable music. But you still need somebody to spend the time to listen to it, so it almost becomes a moot/mute point. 

[0:12:51] Carla: …And some cases play it as well. We have that AI Nirvana song, don’t know if you’ve heard that? So, they composed an original nirvana song, but they actually had to have someone play it in order … (indistinctive)

[0:13:14] Gary: …Those were absurd, because the thing that is difficult to replicate- which is you could argue then is a human thing. The midi for all the nirvana songs/ notation, and then all are fed into the neuro-network, and it generates a bunch of melodies and core progressions that are reminiscent. But then you still need someone to voice them or play them on a guitar or plug the media into your sequencer into garage band or logic or (indistinctive) whatever you are using. And you have a got a nirvana backing track, theoretically. So, it is this illusion of everything being automated whereas actually there is a huge amount of human intervention and curation that goes into it.

[0:14:11] SC: What I’m hearing is almost like; if everything is being fed into it, you are dealing with the masses then reduced or essentially filtered to arrive at something that is the most consistent throughout a particular pattern. But then its always like a flat line. I was interested in the idea of; on the one hand time that you were talking about, I think in relation to spotify and the kind of recommendations etc. But I guess its also a flattening of place, time and culture…

[0:14:37] Gary: A number of theorists have constantly spoken about this slow cancellation of the future. So, the idea of the future has stagnated, and people use different analogies. But essentially if you were a teenager in the early 20th century, you had a vision of what the year 2000 was and there was a future to lookout for. Something that was aspired to and slowly eroded. Mark Fischer talks a lot about the cultural remanence of that, and so he talks about this continuous present- Jeremy Gilbert calls it the long 90s where you see ‘friends’ Tshirts in stores etc.

[0:15:32] SC: But this is to the idea ‘history repeats itself’, this is more of ‘nothing changes’?...

[0:15:34] Gary: There is no cultural progression. And one of the things in music that was pinned is the complete access to the archive. So, Mark Fischer often spoke about immergence of jungle drum and base and UK garage as the last time that there was this feeling of future shock. That is going back to the early 90s, and yet today (so I teach music and music production) I still have a lot of students that want to learn how to do drum and bass. So, it has just been subsumed into the ossified categories of what music is. But South Africa (and maybe Africa at large), have a few outcrops that I think almost deny that i.e the immergence of Gqom and AmaPiano now. With amapiano, the thing that lets it stand out is not the notes that it picked, it’s that base sound. So, it’s a (indistinctive), as opposed to a note, its an atmosphere. Those are the things that are much more difficult to capture or to replicate through big data and certainly through media/midi. I love the fact that a lot of the lyrics are switching between languages and usually the lyrical content is really directly related to the live experience- sometimes in a hyper localized space. And none of that will appear in the training sets of the systems that are being used as neutral, and global, and universal

[0:17:11] SC: The aspect of bass that you mentioned, is that because the quality of base is the physical thing that hits you in the chest or is it because there is another quality to base…?

[0:17:17] Gary: I don’t know, there’s very distinct sound and I think it’s a FM synthesis bass sound that has been used extensively within amapinao. But its evolving, because that came out from; kwaito, barcadi house, deep house. But its got its own flavour and that will keep evolving I’m sure. It is interesting because suddenly you have hundreds of thousands of youtube videos on how to get that bass in amapiano tracks. So its like, light shaker bass percussion, loose groove with this intermittent digital sounding bass liner and it just kind of works. So, the whole genre is around very simple tomber (sp*), which seems simple because people are using the same sound. But, it’s the most complicated thing to explain and everybody is searching for it. 

[0:18:12] SC: I’m thinking, again back to the nature of your collaboration and how it becomes interesting in terms of the form of it. This is because for improvisation to work you need two kinds of things, in this case two humans (different perspectives and the way of doing things). I guess that also feeds to your other work being collaborative as well as the two collaborations you have done here (Nirox) for the performance you working on

[0:18:41] Carla: Actually, there was 4 artists all together in collaboration. So, we approached Napo Masheane (who’s a poet and screen writer), Thulisile (indistinctive surname) (who’s a dancer). We wanted to have a multimedia approach and use the tools in every way possible and expose the culture biases on as many level as we could, and also make it an immersive experience for the audience so its not one language.

[0:19:09] SC: So, you guys started with Thulisile’s dance. She came up, met you guys in studio and it played out from that moment?...

[0:19:16] Carla: She was amazing!. We hadn’t met her before, we just spoke over zoom and we were both a bit worried about how she was going to interact with the sound (which can be quite abstract, unconventional and sometimes a bit uncontrollable AI).

[0:19:43] Gary: Its really taking all the midi that the AI is splitting, which is usually very basic, in-tune melodies, very on grid and applying processes into it, mangling it. So that what comes out wouldn’t be detected by the algorithm as music, even though if you converted back into media/midi, it wouldn’t read as…

[0:20:07] Carla: And sometimes you have some control over it and sometimes not (you never know what’s coming). So, we needed someone who could work with that and improvise. Thulisile had this experience working at ‘Centre for the less good idea’ and we had seen a few of her performances in an unconventional way, she made of a good person to approach.

 As soon as she arrived, we got straight to work, and we were immensely impressed by even just her warmups. Her rhythm alternated, she just moved; sometimes it was dance and when it wasn’t music she could dance to, she created movements. While she was dancing, we would record everything, and we sent the recording to be translated as a 3D - AI figure which is ‘Radical AI’ (motion captured without having to use suit). What we got back was a bad translation of her amazing movement, it gave an effect of someone drunk stumbling over…

[0:20:32] Gary: The dancing software takes a 2D video based on a training set of arch white standard western movements. So the training set then interpolates/fills in the gaps on a 3D plane, which is impressive technology (if it works), but it couldn’t manage with some of Thuli’s movements (nuances and subtle movements cannot be handled by the program)

[0:22:08] SC: What I preach about that, thinking around what translation is; we often think that translation is what bridges cultures but it is often also a one way absorbative process. For example; outside of English as a hegemonic language, everything gets absorbed into it and English grows and the other languages slowly start to be less needed within this other large system. 

Kylie Coetzee writes about the importance of misunderstanding and mistranslation as way to highlight contextual differences and feels like that applies quite a lot here…

[0:22:51] Gary: …Yes, it certainly applies with Napo’s work (GPT 3). The first thing that we spoke about was the idea of things like ‘artificial intelligence’ and all the other new language that goes with it, i.e neuro network, late in space. All these auto encoders and language only exists in English and these technologies are sold to us or represented as future facing. Essentially, going forward these are going to be the everyday technology meaning they will be everyday language. So, the first thing we did was to try and translate artificial intelligence to Sesotho with Napo. There is no direct translation of the words artificial and intelligence in Sesotho, the best translation that Napo could find was ‘Bohlale baMaka’ which roughly translates to (depending on context) false intelligence and that’s the misunderstanding that resonated the best with us. So, Napo started conversing with GPT3 which is the largest/best known text generation tool in which you will find the most newspaper/media articles talking about that as the closest thing to what we call artificial general intelligence from popular culture that in a singularity becomes more powerful than humans and outthinks/ outsmarts us.

[0:24:24] Carla: It fell over the last hurdle of Napo interacting with it. The initial idea was for her to build a poem with AI, we suggested she starts with a line in Sesotho and see what it comes back with then add another line. In the end we decide to keep the whole dialogue with her talking to AI and asking these questions.

[0:25:05] SC: How did we land up on the national anthem? 

[0:25:11] Gary: So, we were asking GPT3 to write some poetry. We asked it to write poetry about the future and then we tested what it knew about the Sesotho and IsiZulu culture. Napo asked it to write a poem in IsiZulu and it literally came up with the first stanza of the South African national anthem…

[0:25:49] Carla: …and then it became quite nationalistic, putting emphasis / glorifying the nation’. A lot of it spewed out the stereotypes of IsiZulu/ Zulu people (the language and the people) being warriors and fighters, and that came out of the poem as well…

[0:26:05] Gary:...and we asked it if the poem was original. To which it replied “I don’t know if it is the original poem, however, it is beautiful”. We mentioned that it sounds like a national anthem and it agreed . We further asked if it had any idea which anthem and it did not know. So as per request, it can provide something in IsiZulu, however, it is stolen as a result of it being incapable of producing something original

[0:26:40] Carla: So, when we asked it to write a poem in SeSotho, it wrote a poem in English and featured one IsiZulu word in there. By virtue of having to give an answer, it makes stuff up.

[0:26:54] Gary: Eventually we got to a point where Napo asked if we knew all of the responses, or it had just made it up? …to which we brutally-honestly confirmed that it was all made up. When you however, consider that most apps that are being developed in this field are made using that as database/model that will spew out complete fabrications about cultural references that aren’t American or Western/European (where it does have some nuances and something to fall back on), that model being used in apps across the board is also the model upon which Tim and Gabru criticized and was fired from google for. In a world where (right wing) we speak of cancel culture, you have someone who has lost a senior job at one of the second or third largest companies in the world on the AI ethics team (not to mention that it’s a woman of colour in a male dominated environment). Then the answers in that context become foreboding, even though they are (in the moment) absurd and hilarious, especially when reading about the incredible power of the system with an enormous amount of inputs and tokens (indistinctive). So, when we asked about what we know about the SeSotho culture, we said it is rich in tradition, vibrant with very hospital people. IsiZulu culture on the other hand is similarly rich in culture, however, the people are known for a fighting spirit. So, it’s the same thing with a limited input as it is what it knows. Or it resorts to stereotypical divisions and we ask about contemporary culture (as to what is happening now?). It says SeSotho culture people are very traditional, wear traditional clothing and love traditional music. So, constantly seeing that culture is part is of the past and not as participating in the present or the future. Even when we were talking about the future, it has no concept to the future outside of Silicone Valley.

[0:29:45] SC: I guess there is an equivalent in google images (for example) if you were to type in ‘Basotho people’, you would probably get an equivalent imagery? … [Gary agrees]

Are there people working on other films of AI addressing this kind of thing? Or are we dealing with something that wants to shut down, are there other avenues for it to function?

[0:30:19] Gary: With composition, on the music side, most of the systems that are well funded and well known are converting everything to midi. However, there are also some systems that are just creating audio and there are a lot of composers and artists dealing with that. For example (mentions name- indistinct) made work looking at the socio-political implications of being able to recognize the sound of glass shattering. Whose control does that power reside once you have got that algorithm that can contain that, and that’s on purely the audio side. 

My work is more focused on the composition side. I actually have a master’s student who’s has taken some of the same models that I’ve been using, but also looking at the work that has been done with Spotify to essentially quantify what makes the most successful songs and to reverse engineer it. She is a pop music writer and she is using these tools the way they are meant to be used, which is interesting. In my supervision of her studies, I find it interesting as my instinct is to think about these things critically particularly because of my relation to South Africa (as it doesn’t speak to our situation here-UK)

On the visual side we used the Gants (which are the generative, adversarial networks), and once again with those, we liked the aesthetics it creates. So, we used some imagery taken from those networks in our last film ‘Always the Land’ and it worked well. But, there is an aesthetic (when you use it a few times), prompted differently; you get really interesting results. But then you figure out how its working and what the date set is as well as the aesthetic of the output. Whenever someone is talking about automating art making it makes me laugh.

[0:32:50] Carla: And then cultural bias creeps in as well and you get a horrible result where if you type certain religion or (indistinctive)

[0:33:06] Gary: So, Microsoft had a chat bot called Tay (sp*), that they let loose on twitter and very quickly it started becoming ; misogynist, racist…so they had to close it down.  One of the things you can see when using ChatGPT3 is the fail safe they built in. So, the engineers will have us believe that the more data you feed the machines, it actually learns on its own and learns to moderate itself. However, in fact, the fail safes have to be hard coded in.

[0:33:35] SC: Does it start with the first thing you program? Is that always the fall back? For example; if you program that tech with something different, would the fail safe respond accordingly? 

[0:33:57] Carla: Yes, if the data set were to be shrunk. Gabrew (sp*) is essentially arguing the decrease of database and increase of quality as well as having proper human contact to build these databases and have them specialized. 

[0:34:23] Gary: There is also danger in pointing out cultural biases and algorithmic injustices , because of something Cory (indistinct surname) calls Critty-Hype. Looking into how face recognition fails to pick up black faces in comparison to white faces. The solution the boils down to getting more data. The Zimbabwean government, for example, sold their entire drivers license image database to a Chinese government to use in AI training. Then the question is, if every art, music or people can be represented in a giant database…does that make things fair? However, if the power/control of that resided in three monopoly or two states that would be unfavorable for anyone. As a result, everything you get will be decontextualized [0:35:38 Sven : You also make assumptions on what diversity is and how to get a handle on that complexity ]… Absolutely and I think the conversation should be whether we are pursuing these avenues at all? And when you add on the resource demand of training, the environmental cost is huge and for what benefit? Even in the areas where really useful progress has been made, i.e medical diagnostics, we seem to be hitting a wall. It can be used to augment human interaction or professionals essentially. However, it is not a substitution at all. Are we going to progress much from where we are? … my answer to that is ‘I’m not convinced’. When it comes to music, it is strange… so google has the magenta project ([0:36:33]: Not sure how to word this this after) . They have made them free and available; I use them. However, the reason that they allocate capital to music is because there is some kind of understanding that if you can engineer automated composition it demonstrates the capital of cultural cool music that ones approach works really well and you have solved music. So, they have no intention of competing with ‘Able to no logic’ or ‘Native Instruments’ in creating interesting digital instruments. They have an interest in raising the capital value of their research and using music as a foil to transfer cultural capital to the real side of their business. That’s possibly a sinical take, but it kind of stands to reason.