The Art of AI

The Art of AI

The Art of AI by Mark A. Taylor

Each week I write a brand new article to support members of our three wonderful art groups on Facebook, The Artists Exchange, The Artists Directory, and The Artist Hangout. This week we take a deep dive into the world of Artificial Intelligence and the world of art.

Artificial Intelligence (AI). It’s the buzzword of the decade and probably the century and it seems that everyone is clambering to be the next AI kingpin or whatever, or artist. Yes, AI it seems has made its way into the art world but more on that in a moment.

Wherever you go everything today relies on at least an element of artificial intelligence. Have an online chat and chances are that you won’t be talking to a real person, you will be talking to a bot, some of them so convincing that it’s really difficult to know whether there is a human in the equation or not. The lines of communicating are more blurred than ever.

AI chat bots were originally marketed as a way to augment a live interaction but most of the time they don’t, they simply replace it. 

When I contact an organisation to complain about poor service or I contact an organisation to say well done (which I do more than I complain), I really don’t want to have that conversation with a bot. For me, and given the fact that I am deeply involved in IT and digital art, right now there is no place for bots to perform customer service for absolutely everything.

Good customer service needs a human interaction and this can be supplemented with AI, but I’m not convinced it can ever be completely replaced. Technically we’re still worlds away from being anywhere close to connecting with people in anywhere close to the same way as we would do when we interact as humans face to face or even over the telephone.

That’s a weird thing for me to say given that algorithms are something that I get paid to create and I have been involved with them for as long as I can remember. But there are certain things in life where algorithms and AI simply don’t have a place or at least don’t as yet have a place. 

I think the biggest issue I have with AI and algorithms is that they are frequently misused by organisations who have begun to rely on them as safe guarders of their bottom line, and they put their faith in the algorithm front and centre and way above any human interaction. But here’s the thing, I am less inclined to do business with a bot than I am with a human and I think the reality and I am guessing here, is that most people would probably agree with me.

AI does have a place, research, engineering, weather prediction, science, medical investigations, forecasting earthquakes, sorting out funny cat memes on social media, the list goes on and there is no denying that for some people with mobility issues, the emergence of AI into things like home automation and medical science has had a deep and positive impact. But when it comes to things that are just better face to face and when human interaction is needed, call me old fashioned but AI just doesn’t give people the same kind of connection, it often just winds them up.

There’s another problem too. If the world is connected digitally, what happens when it isn’t? An internet connection goes down or servers fall over so there has to be a plan B, the problem is that lots of organisations don’t have enough of a plan B to get things back online as quickly as they need to. Plan B’s are expensive too because for the most part your plan b will always only be that, and if you are running a small business then having a might need is an expensive addition to the do need. Sometimes it pays to make sure your plan A has enough resilience built in.

AI and chatbots are also misunderstood when it comes to using them as an answer to certain problems. Banks offer financial advice all of the time and they are increasingly turning to AI to alert you via automated messages if you are about to go overdrawn. It’s a great reminder that there is way too much month left at the end of the money, but AI saying that you can’t afford to eat today because your rent is due tomorrow, isn’t addressing the problem of spending behaviour.

If you overspent on tickets to a rock concert knowing that you still needed to eat and your rent is due and you couldn’t do all of that with the money you had, AI doesn’t address behavioural issues like this and there’s possibly even something that suggests it could even be driving some bad choices because people rely on the safeguards too much. I’m sure there is a research paper in there somewhere!

When you engage with a chatbot resolving issues can also take much longer. Last week I tried to communicate with one when I wanted to upgrade my internet package. Question after question and many “did you mean this” responses finally stopped after fifteen minutes when the bot gave up and decided to route my enquiry to a human. 30-seconds later I was speaking to someone who understood exactly what I wanted. Chatbots might have fifteen minutes to waste, I don’t. 

Chatbots are designed to only solve problems and they are relentless in doing that, but they just don’t seem to understand that they are not equipped to deal with the simplest of things like a customer wanting to upgrade a line. They just go through an epic and endless loop of responses in the hope that one of the responses that someone has programmed in will prevent you speaking to a human. Big businesses love them, customers really don’t.

That might be really good news for small businesses who can’t afford the technology or who don’t have the scale to get the benefit that AI can bring. As an artist, that means that you can start to really differentiate yourself from the big box stores that sell off the peg prints by marketing your personal up close and human customer experience. But some artists are doing much more than that with AI, they are creating whole new bodies of work that has been generated by an algorithm.

The Art of AI…

Back in October 2018, Christie’s one of my all-time favourite auction houses sold a piece of work produced by AI for $432,000, beating the original estimate of $10,000. The artwork was produced by Obvious, a Paris based art collective of artists, machine learning researchers, and friends interested in AI for Art. You can see the Obvious website right here. 

The work, “Le Comte de Belamy”, is an artwork that was produced by a machine learning algorithm, built to enable a computer to create portraits of the 18^th Century in the 21^st Century. GANs or Generative Adversarial Networks were created in 2014 by Ian Goodfellow, a researcher in machine learning but there are stories that even the AIs code was borrowed from Robbie Barrat after placing it on GitHub or at least elements of Robbie’s code was borrowed for the project. Appropriation and controversy it seems really are rampant everywhere in the art world. You can see Robbie’s extraordinary work right here. 

How these systems work isn’t exactly rocket science, I think it is reasonably safe to say that at a deeper level it’s far more complicated than that. The principles of how these systems work though is relatively simple. Algorithms are given various data sets to learn from and ultimately they generate a response based on what they have learned, along with instructions on how and what to apply. The rest of it is down to the AI alone.

It’s a little more involved in that the generator will create new images by mimicking the characteristics of images in a data set and it will then try to fool a discriminator into thinking the images are real. The generator is typically a deconvolutional neural network, and the discriminator is a convolutional neural network. The training continues until the point that the discriminator can’t tell the difference between what’s real and what’s not. In short, it’s a bit like politics and fake news. Think of it as two AIs fighting to the death or at least until one of them says I give up, you win.

Image by Mark A. Taylor

Haven’t we seen this stuff before?

Well, yes. The reason that this latest artwork made the news was that it was the first time that a work produced using AI had been offered through an auction house. AI has had links with art for around half a century, the portrait from Obvious is just part of a new wave of AI. It was also a genius marketing move that managed to peak people’s interest and divide opinion on whether what the AI produced could be called art at all, but AI, neural networks, and many variations of the same thing have been used for many years to produce and influence art.

My first foray into the world of AI generated art was around five or six years ago as an experiment, and over the past four or five years, even devices such as the iPad have been able to generate neural images with a ninety-nine cent app. I dabbled with it but was never completely satisfied with the results, they certainly weren’t good enough to sell. The difference between the Obvious created production and apps though is that the images produced by apps have more limitations which might be because they have fewer training data sets to call from and the resolution of the final images is to be frank, pretty useless for anything other than having a bit of fun.

The apps are best described as a click one button to produce a magical effect affair, there’s no real skill needed other than picking an original picture to disrupt, you don’t have to have a creative bone in your body to produce something that will make someone say wow. Some of these apps will go a bit further and allow some other changes to be made before the final picture is produced, but the work from Obvious was produced by something that was on a different scale altogether.

It’s all smoke and mirrors though because the art whether you call it art or not, wasn’t produced solely by the algorithm. Someone had to write that algorithm and someone had to select thousands of images and place them in a training data set so that the machine could learn. There was an element of human involvement so forget some of the sensationalist news stories that say this is the end of the art world as we know it and artists will be replaced by AI, because AI just isn’t close to good enough to just go off and generate a brand new work on its own without any human interaction at all. I would be more impressed and more worried if the machine had a new idea and woke up at 3am and frantically scribbled it in a notebook under candlelight and turned up at a new art show by 5am.

We shouldn’t be worried… just yet…

ThisPersonDoesNotExist.com

As artists we shouldn’t be too worried that our roles will become obsolete overnight, but there are some elements of AI that should worry us for the future. AI is becoming the vehicle of choice to create fake visuals and fake writing.

There is a website ThisPersonDoesNotExist.com, that is either the stuff of nightmares or pure genius that should be celebrated depending on how much you like the idea of AI encroaching into your daily lives.

The site created by Philip Wang who is a software engineer for Uber, uses research released last year to create an endless stream of fake portraits. Sometimes the images are obviously fake as there are blemishes where the machine hasn’t quite removed the blemishes that appeared in the training data sets, but there are portraits that could literally be of very real people and you wouldn’t know the difference at all. This my friends is equally brilliant and very scary stuff.

Click on the link for This Person Does Not Exist (opens in a new window), and an image of a person’s face will appear. Refresh the page and a completely new person will appear. Nvidea the famed chip set manufacturer released the research which Wang has taken further, and they have made an Open Source algorithm named StyleGAN available. 

Already we are starting to see that the algorithm has other uses beyond 18^th Century portraits and people who do not exist because we are starting to see anime characters, fonts, and graffiti being generated from the same algorithm.

This is the worrying part for artists. When AI is generating great looking fonts, anime, graffiti, portraits and landscapes, and the lines are increasingly blurring between work produced by a machine and a human, it is probably time to get a little jumpy.

GANs have many uses and some of them are not as well intentioned as creating a portrait. They have been used to paste people’s faces on to videos, often in blackmail scams, which is starting to also blur the lines when it comes to trust. How can we trust anything at all if the results are this good and we can no longer tell the difference?

The real problem will come though when GANs become truly weaponised. Already we have seen instances where videos of famous people including The President of the United States, Donald Trump offered advice to the people of Belgium on the issue of climate change. The video had been created by a Belgian political party and was a fake.

Yet it was only back in 2017 when this fakery or what is known in the business as deep fake, started to really open people’s eyes and it was right then and there that a Reddit user with the username “deepfakes” which is a combination of deep as in deep learning and fake, started to upload digitally altered pornographic videos. He had used Google’s own Open Source machine learning software called TensorFlow to superimpose someone else’s face on to the bodies of the women in the video.

Reddit banned the user but not before he had done the inevitable and released an app that did this stuff for you, finally the old advert of ‘there’s an app for that’ rang true. Reports were written by academics including this one, Deep Fakes: A Looming Challenge for Privacy, Democracy, and National Security which you can read right here. 

No sh1t Sherlock springs to mind. Imagine a video of a Police Chief supporting racially charged riots or the President of the United States supporting, I have no idea, but you kind of get the feeling that this activity in the wrong hands can be a flaming match to any tinderbox no matter where you are on earth. I’m sure if they haven’t already, conspiracy theorists will no doubt even suggest that some of this stuff they keep telling us about is very real and here’s the video that proves it. Oh my, there really are aliens hiding away in Area 51.

It doesn’t take much to weaponise this kind of stuff either. If a synthetic deep fake video is released hours before a polling station opens for voting, there is no time for any damage control to be put in place. Elections are one thing but wars can be started just as easily, sometimes it really is just about the timing.

Blogs, news stories, anything that needs to be written can also be generated by AI. Recently a company backed by Elon Musk refused to release a text generator driven by AI because it was too dangerous. Open AI is a non-profit artificial intelligence research group who you can find right here, said that their GPT-2 software is so good that they worry it can be misused.

The software generates coherent text, and can be prompted to write on certain subjects or in a certain style by feeding its learning engine with paragraphs of source material. The algorithm was trained by 8-million web pages and the results far exceed anything we have seen this side of government doors before, and I wouldn’t be at all surprised if we haven’t seen anything better behind those doors either. This is deepfakes for text generation that is so good that if you knew exactly how someone wrote, you would be hard pressed to tell the difference.

Sometimes that would be useful to write this blog and would mean that I could occasionally get more than 4 or 5 hours of sleep, but it’s worrying too. So how good is AI at generating text for blogs?

Well, I signed up for a free account on one of the many online AI text generators which were not using anywhere near the level of sophistication in Open AI’s latest creation, to generate some sample text using only the keywords, art, advice, visual artists, and here are the three sentences that came back.

With the nature of careers in communication, you are expected to keep up with tendencies and the changes to date. Illustrator: Illustrators utilize computer applications and ways to develop images which nutritional supplement text to offer your audience with a clear comprehension of the subject matter. Content Developer: A content developer will compose text or produce other forms of content, like graphics and audio that are used on an internet site.

Not great results but this was a relatively low level and very quick trial of software that is not as sophisticated as the latest offering, but it is kind of sort of, coherent and could probably be used to generate some fake news articles quite quickly. Add those articles onto a monetised web page where ads are generating money based on views and you start to build a picture of why we should be a little concerned about how this stuff is used. You just orchestrate a few moves on a laptop and wait for the money to roll in, good job some of us have at least some ethical standards left.

Sometimes AI can be really great…

I work with algorithms and AI is something I use a lot and while I believe that much of it is misused, there are many practical instances where it is used effectively and it does the job better than any human can, and faster and cheaper.

I automated some of my home, the heating system and lights, which is something that for people who haven’t automated probably seems like a lazy thing to do.  Practically it offers convenience and it also allows me to trigger home routines remotely and I can turn on the lights without having to stop painting. I no longer have to keep the heating on if I am out, I can turn it back on when I am on my way home and save some money on gas and electricity and help the environment. If I stay out later than planned I can remotely turn the lights on and off in different rooms and it provides a little more security. You can even install smart doorbells so you can see who is at the door from your phone.

If you have serious mobility problems then using AI and smart home devices can give you back some independence, and devices such as Amazon’s Alexa devices have been trialed with Dementia and Parkinson’s patients with fantastic results. They can be assistive technologies too, I use mine to set reminders and to make calls and it constantly surprises me when it learns to do new things.

Google Analytics Intelligence

Google Analytics…

If you have a website or you utilise the Pixels sites through Fine Art America then you really should be using Google Analytics with it. I have written about Analytics before but one of the more exciting features to emerge has been the introduction of Analytics Intelligence. Just ask Analytics a question such as how many people visited my website in the last hour will bring up relevant data.

It’s easier too if you have ever tried to generate sets of comparative data. I can tell you that asking for a comparison is so much easier than setting up views and pouring through different reports. It also allows you to set smart goals, smart lists, session quality, and conversion probability. Essentially that latter one is using predictive modelling which means that you will be able to start seeing things like the probability that someone will buy something or follow a call to action or it could alert you to trends such as a lower number of sign-ups for email.

That’s the kind of data that you need to know much sooner than you probably already do because you then have time to get the problem back on track. You can read more about Analytics Intelligence right here, and if you are interested in the other practical uses for AI then visit Google AI right here. 

You already use AI to do lots more than you would think…

If the concept of artificial intelligence scares you then you are not alone. It is scary to think that a machine is capable of thinking or doing something ahead of you in the knowledge that you are likely to do it. Just last week I was freaked out completely when a product ad showed up in my Facebook timeline for a TV I had been looking at in a store, both the exact same model.

There was no way that Facebook is that smart, but it will use cookies from the websites you have visited to show relevant ads unless you change the settings. My experience in the physical store was coincidental but that made me think that it’s not that much of a stretch to realise that this kind of targeting might very well exist in the future, if it doesn’t already.

But AI is all around us, chatbots, Netflix recommendations, and flight prices on online apps, Siri, Google Home, and now it has taken a footing in the art world. There is nowhere left to hide.

Will AI generated art be the next new major art movement? I wouldn’t get too excited about it becoming the next cubism or renaissance movement just yet. We have been conditioned for years to believe that this type of technology will be the future but when you get up close to the work and once you get over the fact that a computer created it, much of what we see is low resolution smoke and mirrors and not particularly original. It is currently only as original as the training data sets it is fed.

But perhaps what we should be considering as artists is embracing it and looking at AI and GANs as yet another tool we can use to produce something even more creative and original. What we have seen to date is a demonstration of the potential, but what we can do in the future is to put it to work to do good, but we do have to be very careful that we don’t continue to create something that is seen as yet just another weapon.

About Mark…

I am an artist and blogger and live in Staffordshire, England. You can purchase my art through my Fine Art America store or my Pixels site here: https://10-mark-taylor.pixels.com  

Any art sold through Fine Art America and Pixels contributes towards to the ongoing costs of running and developing this website. You can also view my portfolio website at https://beechhousemedia.com

You can also follow me on Facebook at: https://facebook.com/beechhousemedia where you will also find regular free reference photos of interesting subjects and places I visit. You can also follow me on Twitter @beechhouseart and on Pinterest at https://pinterest.com/beechhousemedia

If you would like to support the upkeep of this site or maybe just buy me a coffee, you can do so right here.