archives in liquid times
through the bio-scanner. Paul, a robot from the ALGIVIST-5000 series is waiting at the
desk. His emotional module can use an update, I catch myself thinking. I only get a nod
and a metallic "Hello, how can I help you?", so much unlike the newer models that can
really lighten your day with their big smiles and warm voices. I answer the way I am
supposed to do, with a clear question and context: "Hello Paul, I'd like to see all
documents containing discussions on the use of advanced mind models, especially whole
brain simulations, of Facebook users prior to the formation of Mindbook. I also would
like to look at pictures and footage of the meetings that include people from the legal
department, and can you please provide me with additional CV information of these
people? Thank you." Paul knew from prior contact that I would be coming to the archive
myself; otherwise he would have downloaded the interpreted documents, or DOC-
INTERPRETs as they call them here, to my communicator. Now he only sends the
requested CVs and projects an interactive map of the archive a floor below which will
guide me to the right boxes. Since Paul scans and stores all items (including photos and
a shallow semantic analysis of texts), and organizes them in the physical space, he
knows where I have to go. At least, that is what I have to believe since there is no way of
knowing what is in the complete archive. While going downstairs, I sense excitement
from my side on how optimized and effective my routing past all the right boxes, 16 in
total, is. Five more boxes are off limits for me though. It turns out another researcher
has a similar research question in parallel, and his (or her?) combined scientific h-index
and social media coverage is so much higher than mine. Also, according to an analysis of
the planned social activities in our agendas, and our biophysical energy levels in
combination with the predicted moist weather in the next weeks, Paul estimates that I
will not put enough hours in my analysis of the documents and my writing anyway.
Sure... I need to stop eating snacks and boost my metabolism... but come on... who does
Paul think he is? My doctor? According to Paul the overall estimated impact of the other
researcher publishing the material alone is higher when I do not interfere. I have no
other option than to accept, but I don't think it's fair. Archival robots such as Paul are
built to optimize their impact since they too get ranked. Of course, everyone gets ranked,
and so are archival robots. Paul needs to optimize the use and costs of the archive while
at the same time striking a balance between preventing possible negative impact on the
donor organization Mindbook, and stimulating positive impact from researchers and
journalists publishing the right kind of information, again according to Mindbook. Oh
wellthe rest of the day I look at the documents, trying to find what I am looking for.
The surveillance-sensors watch my every move while interacting with the documents,
which helps them to further optimize the archive, so they say. Well... they sure also use
them for the projected advertisements that are appearing on the electronic walls for me.
Heyyes indeedI do need a snackmy hands are tremblingHow did they know?
Oh. never mind.
This scenario may sound like science fiction today, but could be happening in the
near future. The algorithmic archivist Paul, or algivist as I will call it, will be a
natural outcome of the digital age we are only just starting. It is not a matter if all
this will happen, but when. I define the coming archivist singularity as the
moment when all core archivist's activities will be replaced by an algivist. Usually
singularity amounts to general technology (Shanahan, 2015) but here I focus more
specifically on the archivist profession. Just like in autonomous cars, we can talk
about various levels2 of autonomous algivists: some will only maintain digital
archives, some will have a robot body (for physical collections), and some may only
268
martijn van otterlo from intended archivists to intentional algivists.
ethical codes for humans and machines in the archives
function as an assistant of a human archivist. All will, however, be responsible for
selecting, ordering, and documenting archival documents. Introducing algorithms
into our lives, and letting them take over jobs that were exclusively done by humans
will cause profound changes in society and requires considerable thought on how to
do that in a "good" way. A central question in this essay will be about how we can
ensure that algivists will uphold the same moral standards and behavior as their
human counterparts who have been our human gatekeepers to societally important
information for so long.
Worries about the general singularity, when computers will outsmart "us" in every
way possible and may spin out of control relative to our human interests, trigger
existential fears. It reminds of concerns when another technology was in its initial
phases: nuclear technology. Albert Einstein warned President Roosevelt in 1939 in a
letter3 for the consequences if some other nation (Germany) would obtain the
technology for powerful bombs and suggested to start a nuclear program in the
United States. The current explosion in digital technology and algorithms may very
well trigger a similar arms race. But before worrying about superintelligence, we
should study the many ethical challenges of not-yet-fully-superintelligent
algorithms, such as our algivist Paul.
Ethical issues with algorithms arise on a daily basis. For example, Google's search
algorithm tagged4 (photos of) black people as "gorillas", showing either a bias in
data or learning procedures, or errors in the application of the tagging algorithm.
Autonomously driving cars constantly make mistakes5 or are not yet fully capable of
driving in our complex, physical world. A related case is when algorithms are
deliberately used for the wrong purposes, such as the Dieselgate6 case which dealt
with cheating software to fool emission tests. Another example with ethical
dimensions is Facebook's idea to predict7 potential suicides to proactively aid people,
which is in the same direction as Google's recent efforts on depression detection.8
Whether some of such issues may be against current or coming laws, in most cases
we can say they are at least creepy (Tene and Polonetsky, 2014) since people will be
targeted by Google's and Facebook's algorithms as depressed or suicidal: what
consequences will that have? Another creepy example is the Cayla9 doll which can
communicate with children, send their data (voice, things said, possibly video
capture) to the manufacturers' servers, and in addition, it can say anything to a
child through a microphone. Apart from possible hacks, such "connected" dolls are
creepy because they invade (just like smart-TVs and cell phones) the privacy of
intimate family life, without doing anything illegal.
Other recent ethical challenges have to do with the typical gatekeeping role of
algorithms employed by search engines and the like: fake news, Pariser's (2011) filter
bubbles (where algorithms reinforce people's biases), and censorship. As an example,
Facebook's policy to allow or disallow particular content, essentially implementing a
form of censorship10, raises many ethical issues given their 2 billion user-base.
2 https://en.wikipedia.org/wiki/Autonomous_car#Levels_of_driving_automation
3 https://en.wikipedia.org/wiki/Einstein%E2%80%93Szil%C3%A1rd_letter
4 https://www.theverge.com/2015/7/1/8880363/google-apologizes-photos-app-tags-two-black-people-
gorillas
5 https://phys.org/news/2016-09-dutch-police-probe-fatal-tesla.html
6 https://www.cleanenergywire.org/factsheets/dieselgate-timeline-germanys-car-emissions-fraud-scandal
7 https://www.wired.com/2017/03/artificial-intelligence-learning-predict-prevent-suicide/
8 https://www.theguardian.com/commentisfree/2017/aug/25/google-clinical-depression-privacy
9 https://www.nytimes.com/2017/02/17/technology/cayla-talking-doll-hackers.html?mcubz=1
10 http://fortune.com/2017/05/22/facebook-censorship-transparency/
269