~ Photo Cafe' ~ Photographers - Episode #56!
~ Chat Transcript ~
6th of April, 2024
"The AI Chronicles: Interview with Jude!"
H.12:00 PM SLT
[The following Episode #57 will be on:
20th of April 2024]
[12:06] Kika Yongho: Hello Photographers!
How are you? All well? We are.. changed! As you can see, Photo Cafe’ had quite the major makeover, and so we are now meeting in a Mediterranean inspired oasis! So please get comfy again on the new seats and pillows and.. let’s start our episode #56!
Before diving deep into the main topic for today (I recall you the episode’s title: “The AI Chronicles: Interview with Jude!”), we have another news for you: a brand new seasonal photo for our “In The Spot” mono-photo expo on the land! If you look over there, under this main tent, where the curtains are.. – yes, there! – you will see our usual easel!
If you click it, you will be able to take the photo details: this new amazing image is titled “Spring” and has been shooting for us by our sweetest photographer Soileღ (HarperRose Starchild), whom we hearty thank for giving us this beautiful photo that will accompany us until June!
🙂
She also wrote a couple words to go with her photo, that I will paste here for you all to read:
“Spring will come and so will happiness. Hold on. Life will get warmer.” ♡
— HarperRose Starchild
Aww… We can’t wait for more happy moments to come, thank you again Rose! ♥
And now… As we promised: Jude Elowyn (JudeInTime Resident) is here and ready to answer all your questions about a still thorny topic: AI! But who is Jude and how is he entitled to explain this to us? He gave us a short bio, that I will report here for you.
Curious?
We asked him to describe who he is on SL vs. his RL AI related experience. Jude replied: “Hmm, I’ll start with SL I think. I’m mainly involved in two organizations in SL. I curate the Artist in Residence galleries at FOCUS magazine. We’ve had the privilege of featuring artists like Kika there. I also present on occasion at The Science Circle, that’s a non-profit that promotes education in SL.”
“In the real world I work as an AI engineer. My job involves developing AI based solutions for industry challenges and functioning as a mentor, my company runs an apprenticeship programme to train local talent in AI. So I often work with models in computer vision or language processing, but, the bulk of my work or interest has centered around AI in medical uses cases.”
So… Involved with SL arts (he didn’t mention his own SL Photography interest, but you can perv his Flickr here: https://www.flickr.com/photos/157674466@N03 ^^) and AI RL engineer. We want to ask him more questions about AI, and for sure you will do too! So, without hesitation I now hand the virtual mic to our wonderful host Josephine to start episode #56: “The AI Chronicles: Interview with Jude!”. I am done for the intro, please let’s start!
[12:11] Josephine Delvalle: Dear photographers and artists, welcome to a new round of our biweekly discussions.
🙂
Since our topic today is very complex and certainly requires a lot of discussion, we only have 5 questions for Jude and then open the discussion.
First of all, thanks to Jude for putting himself in the hot seat, so to speak 😉
[12:12] Jude Elowyn (JudeInTime Resident): My pleasure 🙂
[12:12] Josephine Delvalle: Are you ready, Jude?
[12:12] Jude Elowyn (JudeInTime Resident): Stretches yep!
[12:13] Josephine Delvalle: laugh
[12:13] Josephine Delvalle: Ok, let’s start 🙂
[12:13] Josephine Delvalle: 1) Can you please briefly (and preferably in layman’s terms) define what AI is in one sentence.
[12:13] Jude Elowyn (JudeInTime Resident): Hmm… Right…
Artificial Intelligence is a collection of powerful, mathematically-based techniques that allow computers to learn from data, and use what it’s learned to perform a range of tasks, like making predictions, identifying oddities, or generating new data.
~Done~
[12:14] Josephine Delvalle: 2) In which areas is AI used?
[12:15] Jude Elowyn (JudeInTime Resident): Quite a broad set of areas really… AI can be safely applied in any industry in which reliable and labeled data is available for it to perform the automation in needs to. Drug recommendations or medical image diagnosis are common use cases in health-care.
In the automotive industry, AI can be used to help self-driving cars identify obstacles on the road. It can be used to build chatbots. Many of us are familiar with ChatGPT, but the same technology can be used to specifically build chatbots to function say, as travel agents or teaching assistants or consultants for government services.
Of course, we’re seeing more and more applications of AI emerge as computational capacities increase with newer systems and more and more data is collected to train models. It’s exciting because, by virtue of that, the fields to which the technology can be applied are endless.
~Done~
[12:17] Josephine Delvalle: TY – a wide range and an open end …
3) Where do you think Ai is most helpful/useful?
[12:18] Jude Elowyn (JudeInTime Resident): Oooh… I think, AI is the most helpful or useful, when it’s possible to ensure that the model that is used is robust in its context. Ensuring this is a multi-faceted challenge, but I think I’ll try to explain this by providing an example, I mentioned autonomous vehicles so let’s use that.
Suppose you were building an AI model to help a self-driving car identify things on the road. The eventual goal of course is to ensure it doesn’t hit anything. In order to train this model reliably, you need to have a lot of examples of pictures on the road, and in those pictures you need to have labels, something like rectangles that you draw around these objects. In this situation, the criteria for whether or not AI is helpful would be whether or not there’s a sufficiently large and diverse labeled data pool to train and evaluate it. It’s most helpful when the stuff the car’s cameras will see in the future will be similar to the data it has already seen in training.
(In short… Is my training data enough? Yes? Then my AI model will be strong and useful)
Likewise, we could also ask if the model fits the use case well. Is it fast enough to predict in the midst of driving? Can it operate well if the on-board computer on the car isn’t powerful? So helpfulness varies on a case-by-case basis.
~Done~
[12:20] Josephine Delvalle: and now the opposite 😉
4) Where do you think the use of AI can become problematic/dangerous?
[12:21] Jude Elowyn (JudeInTime Resident): I can go on and on about this haha, bear with what I have please.
ERHEM! In engineering, if you build a bridge, you have to ensure that the bridge can take the expected weight of whatever will go across it, otherwise it’s dangerous. Likewise, if someone builds an AI model, they have to ensure that the inputs to the model are expected. If I train a model to recommend a set of drugs for a patient’s conditions, I have to make sure that the model has seen the patients conditions before in my training data so it knows what to predict.
If the model hasn’t seen those conditions, the AI model has to be able to tell the user it is not confident in its outputs. It needs to be interpretable for this to happen. The user should also critique the outputs. The model becomes dangerous when it gives out an output that is bad, and the user is unable to discern that the output is bad. In such a case, a patient may be prescribed a drug that is useless or even harmful for their conditions. A model like ChatGPT may hallucinate and give you a false answer, even if it doesn’t actually have the knowledge to answer a question you give it about, say, history. If you can’t identify that and you use the output, you can get yourself in trouble.
AI models that are trained on problematic datasets can also be heavily biased. Suppose you train a model to look through resumes for future hiring in your company. The resumes in your training data come from your current pool of employees. Let’s say however, that your current employees are sadly 90% male. There is a chance that your AI model will give an application a poor score if it sees things like “women’s college” in someone’s education history. What was the problem here? The data used to build the model was not curated well, and so the existing biases in the data were exaggerated. Some of us will know that this is an actual problem that happened with Amazon’s recruiting engine.
These are some examples!
~Done~
[12:25] Josephine Delvalle: nods, TY
Now we are most interested in the use of AI in imaging.
5) Where does normal photoshopping (now) end and AI begin: with filters, subsequent embedding of (fake) light sources, automatic tonal value corrections, generating complete images on the computer without using a photo taken in RL or an image drawn in whatever way?
[12:25] Jude Elowyn (JudeInTime Resident): Hmmm, this is a complex question.
think that, AI really begins when the amount of user intervention in something reduces, when automation starts to increase.
I guess with traditional camera imaging and image editing, there’s a lot of user input that goes into how the image was captured, what exposure or aperture or angle was used etc… In a program like photoshop, you decide what filter to use, what lighting effects to apply and how to apply them. You might even use things like healing to cover blemishes yourself, but at the end of the day a lot of the input comes from you as the user.
AI algorithms involved in editing make a fair amount of decisions based on what they see in the original image, to apply things like say, style transfer, which can be considered an advanced form of filtering (look up CycleGAN if you’re interested). They can introduce light sources by studying the original content of the image and make automatic adjustments to remove things like blemishes or noise. Of course here, the decision making and therefore the creativity in the editing process isn’t derived so much from the user, but from the model itself, so you can say that traditional editing has taken a bit of back seat.
With generating synthetic images, we are talking about photography or painting and the sort taking a back seat. In this instance the creative input from the user is not as well defined, because a lot of the input also comes from existing data and art used to build AI models. We also see that the technical skill in editing and photography isn’t required as much. But that said, the ability to manipulate the model, by means of the text or image prompt (if it’s user generated) in itself is somewhat a skill and an art. This surge in the use of generative AI only began quite recently, after the inception of the attention mechanism and transformer models in 2017 I believe. So we are still trying to come up with proper boundaries and definitions for AI generated art.
~Done~
[12:30] Josephine Delvalle: Thank you Jude, you gave us a lot of information in a short time and also showed perspectives and questions that I’m sure none of us have seen before in that way.
And now, dear photographers and artists, it’s your turn
Let me just remind you: (discussion rules … you know)
[12:31] Discussion Tracker 3.6: Today’s discussion will be run in the C&Q format.
Please type a c in local chat if you have a comment.
Type a q in local chat if you have a question.
If you wish to withdraw a comment or question, type r.
To post a anonomously, add /1 before your comment.
[12:31] Josephine Delvalle: and pls.
give me a ‘done’ when done
[12:31] Josephine Delvalle: We will start with Igi, Igi pls
[12:32] Igi (Igielka Resident): Thank you
Don’t you think AI take personality from image ? I ena I think imperfections of images give something special to image, AI is perfect but can be empty in some way, WE not put hard work , just use AI make it for us ?
[12:33] Josephine Delvalle: Done?
[12:33] Igi (Igielka Resident): Yes
[12:34] Jude Elowyn (JudeInTime Resident): Pardon is it okay for me to go straight ahead and answer?
[12:34] Josephine Delvalle: yes, sure, sorry
[12:37] Jude Elowyn (JudeInTime Resident): Awesome, sorry haha wasn’t well versed with the protocol. I’d say… Yes, for me as a gallery curator, I agree that sometimes, the imperfections in photos do add a tinge of personality. And the drive for perfection with technology sometimes diminishes the effect the human hand has on the art. I will say that… I think it’s really a subjective matter. Different people look for different things in art. Some of us love the imperfections, others look for beauty, others a certain style. It’s really up to you, how to appreciate AI or traditional art 🙂
~Done~
[12:38] Josephine Delvalle: nods, ty
next is Ava, pls
[12:39] A. (Ava Darkheart): You talk about data usage, but we know well that in art, for example, we go beyond mere data usage, especially in image creation and now more broadly in the reproduction of existing people’s voices, faces, etc. It’s about consent and authorization, intellectual property rights, and simply transparency. And we can clearly see that this is absolutely not accounted for in AI usage. How can we have control over these “data” which increasingly resembles expropriation or even theft? Done.
[12:43] Jude Elowyn (JudeInTime Resident): Hmmm… You raise a really complex question here… The notion of consent with the advent of generative AI over the past few years has become a complex problem in the context of AI governance. I think Kika and I once discussed the issues of the misappropriation of existing art, or photography and such… There certainly are algorithms that work to confuse AI models. They introduce TIIIINY changes to the pixels in the image that the human eye can’t see, but changes that will completely mess with the AI model’s ability to learn features from the images
[12:44] A. (Ava Darkheart) shows her tag proudly :p
[12:44] Jude Elowyn (JudeInTime Resident): So measures like that may have to be introduced more widely,when you upload photos onto social media platforms or flickr or the sort. That’s one solution
[12:44] A. (Ava Darkheart): yes but omg… it is us that’s need to protect our art from AI, is it so not right…
[12:45] Jude Elowyn (JudeInTime Resident): The other of course, involves surveillance…Using technology (well AI models themselves) to identify when misappropriation has occurred
BUT… The fact of the matter is… Whether we like it or not. We’re in an age where this technology exists, where people have access to data whenever you post it on the net. For good or bad, we’ve found ourselves in a situation where it’s easy for people to develop generative models using our art, our pictures, our data.
As of now… To me it’s a matter of addressing what exists haha, or trying to do that, whether it’s right or wrong.
~Done~
[12:47] Josephine Delvalle: Nods
and we go with Moki
[12:47] Moki Yuitza: ty Josephine °͜°
From how you defined AI, the term “intelligence” seems not completely correct to me. intelligence presupposes autonomous judgment and a critical spirit, while what you have described is an extremely complex machine that collects, compares and compares data in order to obtain a result compatible with the request made, based mainly if not exclusively on a calculation statistical and probabilistic. My interpretation as a non-expert is probably too simplistic, but from what I understand, an AI that generates images does not create “aesthetically beautiful” images because it deems them as such, but because we ourselves choose between the different outputs that it provides us (all equally meeting the required criteria according to its first analysis) which ones we prefer and we train it to orient itself more towards those. Is my interpretation correct?
and if so, in your opinion is it correct to say that this “machine” is to all intents and purposes a mere tool that still requires human intervention to turn a
simple mathematical output into a “creation” (artistic or otherwise)?
hope Goole helped me enough… done °͜°
[12:49] Josephine Delvalle: laughhhhhh, ty, Moki and google 😉
and we give Jude and google a bit time to answer
😉
[12:50] Jude Elowyn (JudeInTime Resident): Hmmmm….
[12:50] Moki Yuitza: lol i could give im the direct link to my script 😀
[12:51] Josephine Delvalle: nono, you two would start a private discussion without us 😉
[12:51] Moki Yuitza: never 😉
[12:52] Jude Elowyn (JudeInTime Resident): What you have described Moki, when you use the statement “because we ourselves choose between the different outputs that it provides us (all equally meeting the required criteria according to its first analysis) which ones we prefer and we train it to orient itself more towards those.” ….. This is reinforcement learning. It’s where the model produces and output and a human labeler tells the model whether this works or not.
[12:52] Moki Yuitza: exactly
[12:52] Jude Elowyn (JudeInTime Resident): This is…Only one form of learning haha
[12:53] Moki Yuitza: °͜°
[12:53] Jude Elowyn (JudeInTime Resident): And it’s not the most common form, although it’s how ChatGPT is trained
The mathematics that governs the training of the AI model allows the model, using labeled data to, itself, understand where it got the production of a certain image wrong.
This is called the loss function, in a supervised learning scenario.
Intelligence is a complex topic, and I think there are many debates over the notion of… What constitutes intelligence with technology. I would think of it this way. You are right, that the human has a lot of input when providing labeled data. They tell the model what a beautiful image is and what the expected outcomes in production are…
But the intelligence that the model exhibits thereafter, is in its ability to use the patterns learned from the data the human has given it, in the extraction those patterns through the mathematical algorithms that were used in its architecture, and the use of those patterns on unseen data, or unseen prompts, to produce something that meets the outcomes expected. It’s like, you train a lot in the training grounds for a football match, so that you reproduce your skills and knowledge in an actual game and you deliver results.
For folks in the AI field at least, I would think that’s the definition of intelligence
~Done~
[12:59] Josephine Delvalle: TY
[12:59] Moki Yuitza: :))
[13:00] Josephine Delvalle: Next I have Igi on tracker (and over over next too ;-), maybe you can do all in one or decide for one question, so we can give Manuel and Ava and Matt a chance too?
[13:00] Igi (Igielka Resident): hahahaa.. I will try 🙂
[13:01] Jude Elowyn (JudeInTime Resident): You folks have some deep questions… haha… Oh yeah go ahead I have time
[13:01] Josephine Delvalle: 🙂 TY 🙂
[13:03] Igi (Igielka Resident): Should we be so lazy like that ?
[13:03] Igi (Igielka Resident): Done
[13:04] Josephine Delvalle: So lazy to use AI?
[13:04] Igi (Igielka Resident): Yes
[13:04] Josephine Delvalle: Short and good, Igi
[13:04] Josephine Delvalle: Jude … there you go
[13:04] Jude Elowyn (JudeInTime Resident): Hahahaha, okay I don’t have the full context of this question so I’m going to try….
If I am delivering a presentation on science and I want some random images to make my presentation/poster, whatever interesting. I am absolutely going to be the lazy bum that I am and consider the use of generative AI for pictures.
[13:06] Igi (Igielka Resident): Yes !
[13:07] Jude Elowyn (JudeInTime Resident): But, when it comes to featuring photos as someone’s own art… Or editing someone’s art… Applying AI and then staying it’s completely their own? Problematic.
But I can BET that if I make a comment on this, somewhere, somehow, there’s a generative AI user posting on flickr that’s going to come to me with an entire litany of arguments justifying AI art haha. So I am going to simply say, I’m not all that certain.
~Done, sorry Josephine had to think that one through~
[13:09] Josephine Delvalle: Thinking is good 😉
Now we go with Manuel, pls Manuel
[13:10] Manuel Caeran: i think part of my question was answered lol but here i go xd
hi jude! my question is do u think in a future the AI will to replace the designers? or the designers will to use it like a tool?
[13:11] Jude Elowyn (JudeInTime Resident): Gosh another good question… Right…
I’ll say this: Manipulating generative AI is a challenge. Using the right prompts to get what you want is a challenge, there’s a whole new field called prompt engineering to address that… So in that sense, one can say that designers may use it like a tool. That said…
[13:13] Manuel Caeran: oh nice!
I sincerely hope people don’t stop designing manually, I really do want the human hand involved in art, and I don’t want the value of that sort of art to diminish with the introduction of generative AI
ty for the answer and done
[13:14] Jude Elowyn (JudeInTime Resident): And I hope that, with galleries like the one I curate and the sort, we’ll keep that value for art that was made manually, even if it’s not as refined, intact…
~Done~
[13:15] Josephine Delvalle: TY Manuel and Jude and we go on with Ava
[13:15] A. (Ava Darkheart): What do you think about the calls for a pause in artificial intelligence development made by figures such as Elon Musk and Steve Wozniak and some more… and what do you think about the “Existential risk” predictions?
[13:15] Josephine Delvalle: Done, Ava?
[13:15] A. (Ava Darkheart): yes sorry
[13:15] A. (Ava Darkheart): am human :p
[13:15] A. (Ava Darkheart): done
[13:16] Josephine Delvalle: ahh, that was the proof, same as Judes ‘thinking’ before 😉
[13:16] Jude Elowyn (JudeInTime Resident): My response to that is… Fat chance ahha
If you have a desktop/laptop with a GPU, even a CPU, you are capable of writing an AI algorithm to do some sort of automation.
[13:17] A. (Ava Darkheart): not me for sure, trust me 😀
[13:17] Jude Elowyn (JudeInTime Resident): The fact that you’re able to enjoy this virtual world means you probably have the computational power to develop in AI
All you need, is a youtube video to teach you how to do it in AI
Sorry, I mean how to do it with code like Python
[13:18] A. (Ava Darkheart): yes
[13:18] A. (Ava Darkheart): am old school, just SQL so far :p
[13:18] Jude Elowyn (JudeInTime Resident): In a sense… The proliferation of this technology, preventing it with policy is going to be soooo difficult, given where we are with it now… ~Done~
SQL’s great!!! You have a good foundation 🙂
[13:19] A. (Ava Darkheart): hahah
[13:19] Josephine Delvalle: And now Matt (it was long time to wait, I know)
[13:21] ᴍʌᴛᴛ (MatthSuogan Resident): Thanks for sharing your thoughts. I personally struggle to see it as genuine AI art… for me, art is rooted in human creativity and emotion, I believe that true artistic expression comes from human experiences, emotions, and imagination, which AI lacks.
and thanks you
[13:22] Josephine Delvalle: TY, Matt
[13:22] Jude Elowyn (JudeInTime Resident): Good point, I agree with that. It’s one reason it’s really difficult for me to respond to someone that comes to FAIR with a request to feature AI art haha.
It’s true that art in its traditional sense is valuable by virtue of how experiences shape it. There was a gallery, gosh what was it called…LOST by Mihailsk
wrote an introduction for it in the FOCUS magazine, it went like this (sorry just a bit more Josephine)…. In the evocative space of Nitroglobus Gallery, Mihailsk, affectionately known as Miha, unveils “LOST.” This return to the site of his Second Life (SL) artistic debut is marked by a deeply personal exhibit: a monochromatic journey shaped by the tragic loss of his beloved partner. Miha’s minimalist photography transcends mere visual appeal, delving into a narrative of love, of loss, and fortitude.
That’s an example of how experiences shape art. ~Done~
[13:25] Josephine Delvalle: Aaand the last one is …. Igi … again, pls.
[13:25] A. (Ava Darkheart): i have so much more questions like 256 but oke, i bite my tongue and fingers lol
[13:26] Jude Elowyn (JudeInTime Resident): Hehehe I’m sure there’re going to be more opportunites to discuss that in future iterations of this at the Photo Cafe 🙂
[13:26] Josephine Delvalle: I am sure, Jude will have 256KB answers 😉
[13:27] Jude Elowyn (JudeInTime Resident): 256 bytes 😀
[13:27] A. (Ava Darkheart): 😀
[13:27] Igi (Igielka Resident): oh my… I have so many to say and ask still. But yes nice was hear Jude. You agree human bring something which still Ai can’t. And left it as is. I think AI is easy use for lazy people, but who care bring art … still will work manually
[13:28] Jude Elowyn (JudeInTime Resident): You’re right Igi. It is a nice tool for a lazy person haha. I prefer taking my own photos as many of us do.
[13:28] Igi (Igielka Resident): Yes, I saw and it was nice see
[13:29] Jude Elowyn (JudeInTime Resident): Thank you :))) as are yours
~Done~
[13:29] Josephine Delvalle: 🙂
[13:29] Igi (Igielka Resident): Thank you 🙂
[13:29] Josephine Delvalle: Applause for Jude, pls!
[13:29] Kika Yongho: ** APPPPPPPLLLLAAAUUUSSSSEEEEEEE
[13:29] Moki Yuitza: ** APPPPPPPLLLLAAAUUUSSSSEEEEEEE
[13:29] Igi (Igielka Resident): ** APPPPPPPLLLLAAAUUUSSSSEEEEEEE
[13:29] ϮƗ Vɪᴏʟᴇᴛᴛᴇ Mᴏʀɴɪɴɢꜱᴛᴀʀ ƗϮ (Violette Rembrandt): .•♫•♬• Ｃｌａｐｓ ☆ Ｃｌａｐｓ ☆ Ｃｌａｐｓ •♬•♫•.
[13:29] Jude Elowyn (JudeInTime Resident): Hehe thank you so much 🙂
[13:30] Josephine Delvalle: It was great. Tons of information and a very good and fair and open discussion.
[13:30] Igi (Igielka Resident): Yes, it was 🙂
[13:30] Moki Yuitza: we MUST do this again :)))
(sorry Jude) 😀
[13:30] Igi (Igielka Resident): Yes Ai is still hot theme for talk
[13:30] Jude Elowyn (JudeInTime Resident): I really appreciate the opportunity and the thoughts and insights you’ve all shared 🙂
[13:30] Josephine Delvalle: Jude, you hear the calling?
🙂
Thank you to everyone.
[13:31] Jude Elowyn (JudeInTime Resident): Hahaha I think Kika’s got more personalities in mind with more ideas to add to the conversation as well, I’ll be happy to add a bit here and there when needed 🙂
[13:31] Josephine Delvalle: We will see us again 20th of April.
[13:31] Jude Elowyn (JudeInTime Resident): Have a fantastic weekend everyone 🙂 Thank you so much for the opportunity Kika and Josephine 🙂
[13:32] Kika Yongho: Thank you Josephine, Jude and everyone who attended!!!
[13:34] Jude Elowyn (JudeInTime Resident): My pleasure, and privilege too I learned a lot about your thoughts too 🙂
[13:34] Moki Yuitza: i’m afraid we had just scratched the surface… 😀
[13:35] Igi (Igielka Resident): I still have many questions and thought, but maybe will be another talk about it 🙂
[13:35] Josephine Delvalle: Yes, even Jude gave us a big bag filled with infos
