Note: This was the write-up for the last part of my EDBT keynote. The talk itself was of course slightly different.
This concludes
the scientific part of my talk. In the remaining time, I would like to talk
about bananas.
My parents were
visiting me in Copenhagen in August 2024. One day, we were walking in the
Vesterbro neighborhood. And we saw a young well-built man coming toward us in
the other direction who had an empty banana peel in his hands while eating
another banana and holding a third, not yet peeled, banana.
Seeing this
image, I turned to my dad and said, “You have given me one third of a banana my
entire life, and here is this guy eating three whole bananas by himself.”
Why one third of
a banana? My dad has the principle of splitting whatever food he serves into
equal pieces to people he is serving that food to. While I was growing up, we
were three people in the house (my mom, my dad, and I). When we had/have visitors,
the distribution of the food was/is adjusted accordingly.
I get my dad. I
get where he is coming from. It isn’t just because he didn’t grow up rich, and
that is an understatement, but for his generation growing up in Turkey, bananas
weren’t that accessible even if you were rich. (Depending on your country of origin
you may have this relationship with different kinds of fruits.)
Today, bananas
are more accessible for everyone in Turkey because of the many greenhouse
productions. Similar trends exist for many other consumer products whether it
is clothes, food, or … technology.
Since their
release, the cost of using generative AI tools has declined. This is what we
expected, and this is what we wanted. As a result, they are more accessible to
a larger scale of users, and I experience more and more of the following:
A friend tells
me that they use ChatGPT to get ideas for what to cook later.
Colleagues tell
me that they use one GenAI tool for literature search, another GenAI tool for
brainstorming for ideas, yet another GenAI tool for help with writing, and
Claude Code for coding …
Students tell me
the answers they got from GenAI tools for errors, definition of a concept,
setup steps of a library ...
I get why people
embrace these tools so much. Today, it also takes me less time to search for
something if I am using such a tool compared to old-school web search. But I am
also almost embarrassed to say that I don’t have a drive to consult GenAI tools
by default, and I in general avoid them. (I also avoided smart phones for a
long time.)
Technology is
like any other consumer product. The cheaper the product gets, the more
accessible it becomes, which creates higher consumption, which in turn requires
ever growing resource needs to deliver the product. In economics, this is
called Jevon’s paradox.
There is always
a high cost to a cheap product, and high consumption translates to high carbon
footprint.
That is why we
get depressing / dystopian news articles about the energy demands of data
centers and this being driven by the demands of AI. *
How do we achieve
more sustainable progress in the field of AI?
Do we always
need the biggest / latest GPU? Do we always need bigger scale?
We keep
rejecting academic papers for not targeting larger scales.
Do we have to
use a GenAI tool as often?
The banana
pictures I showed were taken by me, not by GenAI, and I then ate those bananas
over a two-week period.
How do we decide
how often to use things? Who decides?
I neither want
these tools to be inaccessible to people nor want to be patronizing and tell
people not to use them.
But how do we incentivize
lower use?
At the end of
the day, all hardware vendors want to sell more hardware, since there is no
economic incentive to do otherwise.
I have a BSc in
computer engineering and PhD in computer science. I am not an economist,
anthropologist, political scientist … I
am qualified to discuss a subset of the above questions but not all. A more
holistic discussion requires reaching across the aisle and talking and
collaborating with people from other disciplines.
* Some of these articles:
The
Obscene Energy Demands of A.I.
Google
plans to put datacentres in space to meet demand for AI
Inside the Dirty, Dystopian World of AI Data Centers
