For the thematic project of the second trimester, i've been working on this film, which i called...

...'serving simulations'...

[what]

...which is an inquiry on the dialog between humans and Artificial Intelligent machines. It contains 5 short filmclips, which are made after 5 iconic science fiction films, that are made in the period from 1968 to 2013. The choise for these films was based on their main characters: as in all these stories there is an actor that is playing the role of a Artificial Intelligent machine.

And so this film is made thanks to

HAL from 2001, a space oddysey
David, from AI
Gerty, from Moon
Ash, from Black Mirror's 'Be right Back'
Samantha, from Her

As science fiction films are generally a source based on human desires, these 5 films all offer a different perspective on artificial intelligence. They present their own depiction of the communication between human and machines.

[how]

So my main focus were the dialogs in these films. And I've been working with the subtitles files mainly, to analyse and select these specific parts from the films. i worked with a piece of software called videogrep to select the parts of the dialogs in the subtitles file that i wanted to use. Videogrep is a command line tool, that makes it possible to then search for this 'higlightcode' which was here 'XXX' and so create a new version of the film.

blurring

I've been blurring the films itself for two reasons: 1. because they were that iconic, specially 2001 which is visually super rich and 2. it would be a distraction when the viewer tries to listen to the spoken text. Next to that, it puts the focus on the voice, the main channel where these machines communicate through.

After that, I tried to name each clip, by describing the certain characteristics of that AI machine.

option 1 : non stop here (Ash)
option 2 : promising politeness (HAL)
option 3 : perfect simulation (David)
option 4 : unlimited trust (Gerty)
option 5 : personalized profit (Samantha)

The film is been put in the format of being an option menu. So at the moment you come home with your newest AI device, these are the options you can select. And so the film end with: your selection is: 2

[why]

this introduction in a book called 'Verbindingen/Jonctions' describes the issue that i wanted to speak about in this film, in a clear way. It's an introduction to a little story about cookbooks, by Femke Snelting (2009).

Whether we operate a computer with the help of a command line interface, or by using buttons, switches and clicks... the exact location of interaction often serves as conduit for mutual knowledge - machines learn about bodies and bodies learn about machines. Dialogues happen at different levels and in various forms: code, hardware, interface, language, gestures, circuits.

Those conversations are sometimes gentle in tone (...) and other times they take us by surprise because of their authoritative and demanding nature: "Put That There".

I consider artificial intelligent machines mainly to be an interface. It's the form of the interface that tries to mimic human to human contact.

And as this contact is mainly going through a linguistic format, namely through dialog,

I was wondering how this these human simulations, INFLUENCE THE DIALOGS itself.

And so I created these dialogs by selecting moments from the films, where the machines start to question, request, or even instruct their user. The dialog is not only a format of communication that goes from human to machine, but certainly also the other way around.

does it bother you?

To emphasize that even more, i added the typographic layer, that speaks to the viewer of the film. It's a mixture of using the sentences of the spoken text, and describing the AI machines.

the one that knows how you'll react

With this layer, I tried to highlight the mutuality of these dialogs.

the questions: who's serving who, & who adapts to who, could be answered by saying that these borders are blurred. It's a constant dialogue between the user that tries to adapt the machine's system in its life and the machine that tries to adapt to you.

It's a bit like: me learning to code, and becoming aware of concepts like - loops - else-if statements - switch case-statements and so slowly learning to understand the computer → in stead of having the computer trying to understand me.

And it's this understanding of a computer, of a machine, of AI system, that is all going through the interface of an human simulation.

An human simulation as:

Siri
Google Now
Cortana
Amazon Echo

The first three are applications that come with specific hardware. The last one is both: a device to put in your living room.

These 4 devices, use the AI interface, as a metaphor. It's trying to be a metaphor for human to human contact, through dialog. And so you would say easily that you 'interact' with your device, as you would do with another human.

But: as Florian Cramer mentioned last Thursday in the Q&A of Nana's talk (in the context of 90s aesthetic interaction): interaction is a social concept.

This is also the argument that Bruce Sterlin is making: - humans have cognition - and artificial intelligent machine are products of computation --> and so it's wrong to mix these two up so easily. They won't converge.

This interest in seeing the interface as a dialog, has been coming back in both the 1st and 2nd trimester. I enjoy to work with linguistic tools, as i feel that this brings me close to the subject of dialogs.

And so i would like to list some tools i've been using :

CMU Sphinx

In the first trimester i've been looking at speech-to-text software for example. And I've been looking at CMU Sphinx. A speech recognition system developed at the Carnegie Mellon University from 1988.

Software that is used for voice-commands for example, (in devices as Siri, and Echo), but is also used to automatically generate subtitles for films and video. and Youtube has been integrating this option in its services since 2014. I was curious about the way a computer interprets human speech, and what elements it needs for that.

ngrams

One of the elements where ngrams:

language model
lists that contain all the *probabilities* that a spoken sentence can contain

[img phonemes]

dictionary file, .dic file
I used it here to demonstrate how voice-commands for Siri are recognized through a set of phonemes, as the phonemes are the 'raw' input of the STT software

Pattern

workshop week Cqrrelations in Brussels, organized by Constant

Pattern is a piece of software for text-mining, Which is a type of dialog, where the computer tries to recognize patterns in text, and classify it on for example: sentiment, gender or age.

It contains a module for - machine learning - train software, to recognize a text being "positive" for example. to do this, you show it examples and classify them as being "positive", negative, or neutral - after a while, the software is able to recognize positiveness by itself

This the annotation process
it's a dialog between human & machine # moment between two truth systems, as Femke called it
the moment that a human classifies a text as positive, contains a certain truth
later on, the algorithm is able to "point out" which text is positive or not
algorithms that are being used to present a certain 'truth' They are used in applications for
- lie-detection in reviews
- and pedofile-profiling on fora's

—> this makes me realize that this could be a critical position to take an argument that makes it very important to reveal the 'behind-the-screen' of software

→ to start to understand the machine in stead of letting it only trying to understand me

User:Manetta/serving-simulations/presentation-text

Contents