Qualia are private and subjective experiences that exist in the human mind. They can be described, but not accurately enough for another individual to know what the experience is like.
Common examples of this are the smell of coffee or the taste of an orange. The best way we can describe these qualia are in reference to other qualia. Like “the smell of coffee is warm, and similar to that of toasted chocolate”, or “this orange tastes like the other times I tasted an orange”. Even then, the specific qualia that an individual experiences cannot be translated to an external source.
I have heard theories that qualia can be translated, but our spoken languages are not advanced enough to do so yet. I believe it was a Vsauce video. I think it is possible that individuals have their own internal language through which they think, after all, language is just a system of signs. This language is then translated into a universal language like English when it needs to be expressed to an external being. Information is lost, as the universal language is not as advanced as the internal language, and must be simplified.
Vsauce covers this at 4:00 in the video below
Where Does Consciousness Reside
A lot of these questions and thought-experiments boil down to whether one believes that consciousness is a physical phenomenon, the arrangements and number of neurons and the nervous system, or if consciousness is intangible.
My outlook on the world used to be nearly entirely rational and I would have agreed that consciousness was entirely physical. After realising that rationalism is an almost useless way of actually living life, I began to have a more empirical outlook. This is not to say that I still don’t believe that consciousness isn’t physical, I still think there’s a good chance that it is. My experience is entirely irrational to me.
This thought experience involves an individual having an exact copy of themselves but are not conscious. They do, however, behave as though they were conscious.
The problem seeks to address whether consciousness is external to just behaving as if one is conscious. Would you feel guilty about mistreating intelligence that behaved and reacted to stimuli as if it were human? Who’s to say that consciousness is anything more than behaviour.
I am currently studying existentialism for my DHT essay alongside this project and have come across some references to consciousness. There is a whole school of thought dedicated to consciousness and experience called phenomenology. This school preceded existentialism and I know very little about it. I believe that phenomenology seeks to understand the world through human senses and experience. The most well-known phenomenologist, Edmund Husserl, famously said: “to the things themselves!”.
Jean-Paul Sartre first heard of phenomenology when his friend, Raymond Aron told him, “if you are a phenomenologist, you can talk about this cocktail and make philosophy out of it!”
Id, Ego and Superego
As I was reading a section on bad faith in Sartre’s “Being and Nothingness”, he refers to Freud’s theory of the id, ego and the superego. Freud theorised that the id is the drive to fulfil basic urges as soon as possible, like eating, drinking and sleeping. The id does not reside in the conscious mind.
The id obviously can’t be satisfied immediately, so the mind will begin the ‘primary process’. The primary process causes the mind to daydream about fulfilling the need until the id can be satisfied.
The ego is responsible for dealing with the id in a pragmatic and socially acceptable way. The website linked above mentions Freud’s metaphor for understanding the id and the ego:
“Freud compared the id to a horse and the ego to the horse’s rider. The horse provides the power and motion, yet the rider provides direction and guidance. Without its rider, the horse may simply wander wherever it wished and do whatever it pleased. The rider instead gives the horse directions and commands to guide it in the direction he or she wishes to go.”
The superego is responsible for the morality of the methods in which the id is fulfilled. These morals are often gained from a person’s upbringing and the society in which they exist. Going against the superego will often result in feelings of guilt or remorse. The superego and ego exist in all conscious states.
Notes from Sarah Bakewell’s “At The Existentialist Cafe”…
Phenomenology, was developed by Edmund Husserl. He set out to “describe phenomena”. Phenomena can be anything, object or event that presents itself to an individual’s experience, rather than trying to denote its objective qualities.
Here is where phenomena begin to resemble qualia. Bakewell states:
“What, then, is a cup of coffee? I might define it in terms of its chemistry and the botany of the coffee plant, and add a summary of how its beans are grown and exported, how they are ground, how hot water is pressed through the powder then poured into a shaped receptacle to be presented to a member of the human species who orally ingests ut, I could analyse the effect of caffeine on the body or discuss the international coffee trade. I could fill an encyclopaedia with these fact, and I would still get no closer to sawying what this particular cup of coffee in front of me is. On the other hand, if I went the other way and conjured up a set of purely personal, sentimental associations – as Marcel Proust does when he dunks his madeleine in this tea and goes on to write seven volumes about it- that would not allow me to understand this cup of coffee as an immediately given phenomenon either.”…
“If I treated all these as purely ‘subjective; elements to be stripped away in order to be ‘objective’ about my coffee, I would find there was nothing left of my cup of coffee as a phenomenon”.
In 1939 Sartre wrote an essay titled “A Fundamental Idea of Husserl’s Phenomenology: Intentionality”. In this essay, he describes what it is to be conscious. He believes that to be conscious of something, is to go “beyond oneself”. His words aren’t definite, and so it is quite hard to decipher what he means. He explains that when he tries to be a tree, he is pushed away from it, he can’t “lose” himself in the tree, no more than than the tree can lose itself in him.
He describes consciousness as something that is trying to leave itself and be ‘other’. He says that consciousness has no ‘inside’ which is interesting. I always think of consciousness as being like a dark room in which I exist and process some sense-data around me. He ends the passage by quoting Husserl: “All consciousness is consciousness of something”.
A neural network is essentially a method in which a machine compares its input to many different parameters to identify the input. It is designed to mimic the process of recognition in the human brain.
This type of intelligence is useful for image and sound recognition, as there are so many parameters, that it becomes extremely inefficient to write an algorithm for it.
This project explores the realm of computer consciousness, including exploring computer vision, artificial intelligence (AI), machine learning (ML); their associated toolsets, philosophical and ethical considerations.
Students will be introduced to methods which allow the computer to see and make sense of its environment, and look at how these can be used to drive realtime applications.
The keyword to consider here is awareness. What exactly is awareness and how do we evidence it? How can we make the computer more aware of its surroundings, build knowledge from this and respond to external stimuli? How can we make the computer seem aware?
Over the period of the project we will look at methods for expanding the computer’s awareness of its environment and itself, but also explore debates surrounding recent developments in the subject area.
Possible technical areas are computer vision, AI and Machine learning. My project should have a strong theoretical base considering the subject is rich in moral and philosophical study.
I may find it difficult to apply my research in an interesting way, however. Machine learning pieces tend to have an obvious reading and they always look like machine learning. I would need to do something with more depth than “This is what a computer thinks this looks like”.
I could also go down the route of computer vision, seeing as I have a webcam and I believe I may have Kinect back home.
There have been a great number of cases where machine learning has been applied with negative consequences.
A common example used is Microsoft’s Twitter chatbot. The aim was to create a bot that would speak like a teenage girl on Twitter but within a day, trolls managed to train the bot to make unsavoury remarks. This isn’t as good an example as it was obvious it would happen from the start and could easily be prevented.
In many cases, people have tried to use AI to eliminate human bias. AI is now often used in job applications and interviews. The AI used by Unilever in its interviews analyses an applicant’s speech. This could prove troublesome as speech is an obvious identifier of class. A candidate with a strong accent or a local dialect will likely be unfavourable. Amazon built an AI that filters through many applications in order to whittle it down to the best 5. This AI would automatically exclude women from having a chance at the job.
Another AI used to identify criminals, produced many false positives when used on congressmen and women. These false positives were twice as likely to occur in people of colour.
Human’s are extremely biased creatures, and although we can learn to consciously overcome bias, it is unlikely we will eliminate it. AI seems to inherit our biases, but I am unsure whether it amplifies or suppresses them.
Gathers the visual data that invokes a strong classifier response from a neural network, then constructs ink sketches from this visual data.
The neural network would see this object as a fan, whereas we would see an abstract series of strokes and lines. However, once you are told this is a fan, you can begin to see an image of a fan. This is a really interesting aspect about both human and machine perception- although very different, we can often alter our own perception to begin to understand how a machine sees images.
I have already encountered Tom White’s work as research for my web APIs project in second year. That project was about how AI may unintentionally censor creative work through the goal of being advertiser-friendly.
Now I am looking at the piece through the lens of perception. It is interesting to see how many of the objects still bear a strong resemblance to the real object in a human mind. I suppose the difference between a human mind and a computer’s mind is the degree of certainty in identifying these objects. We know that these images look like these objects, but a computer is most certain that these images are the objects.
I like the idea of turning the function of image recognition AI inside-out. White is creating images from image recognition.
Mario Kilingemann tends to work with visuals generated with Neural Networks.
Google’s teachable machine is a quick way to have p5 trigger events based on visual, audible, or pose data. After the machine is trained, you are given a link that can be used in p5. Label variables can be accessed and can be used to trigger events at a specific level of certainty.
Here’s a quick p5 sketch that triggers an image depending on what object I hold up.
I have noticed that MPs all seem to have an official government photograph taken with the same lighting and backdrop. I realised, that this may be the perfect image set for Runway image generation.
After researching bias in AI, it would be interesting to ask a machine what it thinks a politician looks like. Perhaps it will reveal a bias in politics.
I assumed that there would be a government data set including all the images of MPs. I couldn’t find it, but I did find a government site in which you can enter a number into the URL and it’ll return the MPs photo. With a few lines of code, I wrote a sketch that cycles through the photos and saves them to my computer in 512x 512px png. I set the limit to 10000 images, but I noticed they stopped at around 5000.
It wasn’t just as simple as that, unfortunately; it never is. The sketch will save duplicates quite often and the data set has hundreds of avatar placeholders. I had to manually delete all duplicates and placeholders. I was left with 926 photos, which should be enough to experiment in Runway.
I began training the algorithm. I started with Flickr faces, as it seemed to encourage me to do so. I was a bit worried because it included images of children which is completely nonexistent in my dataset.
It is interesting to see that even during the training, many women have changed to men, children have turned old and people of colour have turned white.
I saved the little example video it gives you at the end.
As you can see, there is a definite trend in MPs. The computer repeatedly draws older white men. Occasionally it will turn to a woman. I find it interesting to watch the clothing they wear. The tie shifts colour and when it hints at turning to a woman, the collar dissolves into the skin. I was expecting this result, but it still has an impact.
This is a good start, but I will admit, my dataset is flawed. I noticed it contains many MPs that are now resigned. It also contains many other people that I believe are not MPs. There are also 650 MPs in the UK but I had 950 images.
I will try to produce another dataset and increase the steps this time. It would be good if I could sort the MP photos by party to see what the average MP for each party looks like, I am just unsure that I will be able to gain enough photos, considering each party will only have a couple of hundred MPs at most.
I managed to find a way into the XML sheet that contains all of the MPs that are eligible to sit. I can tie this data in with the image database, so I can use there ID number to find the photo. I can also use my processing sketch to create a datafile containing their names, ID and party and link this to their photos.
I began writing a script that goes through all of the XML MP data and retrieves there name, ID and party. I will be able to use this to save an image file for each MP and sort them by party.
I copied my old code but changed it so the URL followed the MPs ID number. For the most part, this worked, but many MPs don’t have images, and the URL seems to break for several MPs. I will need to go in and manually fix these ones.
It took a good while but I a went through each image and replaced the duplicates. I have 650 images in total for 650 MPs, however, it is probably a good idea to go into photoshop and produce more images for training the model.