Does AI fail to deliver the 3 Ts?
Every conference seminar I have been to recently includes a deep discussion on AI. My favourite was a presentation given by Tom, who threw out a number of Ts which undermine the future of AI as a useful tool. I am not going to paraphrase his seminar but I am interested in exploring some of the notions within 3 of these Ts – Truth, Trust, Transparency.
Truth?
As a picture researcher, one of the holy grails (of images requested that don’t exist) is the football match between British and German troops during the truce over Christmas in 1914
MidJourney5: WW1 British and German soldiers playing football in no man’s land – Elements of truth but not convincing. One of the few images produced where only one ball was visible.
MidJourney5: WW1 British and German troops playing football against each other in no man’s land on Christmas Day – Christmas Day = snow… and a Santa hat!
MidJourney5: WW1 creased and torn black and white photograph with a white border and mold spots showing British and German troops playing football against each other in no man’s land on Christmas Day. – change up. Lots of balls, cannonballs? Lots of troops.
None could be argued as truth. AI is not there yet. For some time to come, military historians will be able to pick holes in uniform details. Others will know the types of football available in 1914, all man made features can be dated , often determined as feasible within a certain geographic location. Natural records can ascertain truth. Was there snow on the ground over Christmas in 1914, if so where?
The real question is does it matter? Factually yes, of course it’s important. The point of Truth is not whether something, that is AI generated, looks real but if, it is clearly labelled as AI generated. Passing off AI as truth is the largest threat, whether that be imagery, text, or other.
One has to remember that before photography was published we had illustration. Photography was used as a source of copy but inherently the illustrations depicting news were the imaginings of the creator. The viewer was under no illusion that the scene was not real but nevertheless the illustration would provoke an underlying myth that the viewer would consume as truth.
MidJourney5: The Christmas Truce has become one of the most famous and mythologised events of the First World War. But what was the real story behind the truce? Why did it happen and did British and German soldiers really play football in no-man’s land? Late on Christmas Eve 1914, men of the British Expeditionary Force (BEF) heard German troops in the trenches opposite them singing carols and patriotic songs and saw lanterns and small fir trees along their trenches. Messages began to be shouted between the trenches. The following day, British and German soldiers met in no man’s land and exchanged gifts, took photographs and some played impromptu games of football. They also buried casualties and repaired trenches and dugouts. After Boxing Day, meetings in no man’s land dwindled out.. – Prompt from the first paragraph at Imperial War Museum no media suggestion, don’t count those fingers.
Trust ?
Trust can be broken through truth. We have more than the passing off from AI prompter to viewer, we also have the relationship between the AI generator and the AI prompt. How can I trust what comes out of the machine is what I have asked for? This is a question that goes back to another T – Training. What datasets have AI been trained on and are they accurate? what biases do they contain? This is a complete unknown, for all I know the AI generators are harvesting the data of AI generated images and feeding them back to a fulfill completely distorted reality. If I wanted an accurate WWI image I would be more trustful if the dataset was restricted to something like the IWM image database. I am unable to isolate that, there is no trust, only a deception of smoke and mirrors.
MidJourney5: there is no trust, only a deception of smoke and mirrors.
Transparency ?
Linked to all of this is transparency. Again, labelling the AI image for the trust of your viewer adds transparency. What we do not have is transparency from the AI generators. AI fails to list its training datasets. We lose trust in what it delivers, truth gets distorted and the resulting images are skewed further from reality to an accidental inclusion of partial fact and unbridled fantasy.
MidJourney5: truth gets distorted and the resulting images are skewed further from reality to an accidental inclusion of partial fact and unbridled fantasy.