Serendipity and Emergence
Weird realization and take. On rojak, serendipitous connections and emergence in AI.
Week 17 of life post-MAS. (Links to past weeks in my newsletter.)
The familiar. An old friend from a cell group when I was studying in the University of Toronto. Now leading work in an organization doing advisory for decision making. A bunch of folks planning an AI bootcamp. An interview with a director of AI governance for my CCAF course.
New connections. Board members from an African insurer that I was conducting training for. Around 50 folks from the risk team of a local bank that I did a talk for. A researcher from the CCAF coming to Singapore soon.
The week ended with two serendipitous connections.
First, I had breakfast with 3 classmates from Dunman High who I had not met for years. The breakfast made me want to craft a joke on what happens when a veteran reporter and lawyer and two PhDs who don’t stay in their lane have kopi and kaya toast. Then after a talk I did for a local bank’s risk folks later that very afternoon, I looked up and saw another classmate from the same class as my breakfast mates standing in front of me.
Second, I was on the way back to my car after the talk, and crossed paths with a retired boss of mine from MAS. He was travelling between the offices of two financial institutions he was on the boards of. A common friend in MAS had just mentioned him to me a week ago about a potential gig. So this chance and short encounter was actually a productive one.
These two encounters made me think about the connections between serendipity in life and emergence in AI. And about my rojak life and introverted nature.
Rojak & Introversion
For the uninitiated, rojak is a Singapore dish that sounds like a mess. Fruit, vegetables, dough fritters, prawn paste, crushed peanuts. Completely different textures, flavours, origins. And yet it works.
I’m rojak in my approach to life. Arts policy. Banking regulation. Quant finance. Investment risk management. Digital art. A mid-life PhD in Computer Science. AI governance & risk management. Watercolour painting. None of it makes sense. But it works for me.
I am also an introvert. If you see me being awkward, that’s natural. I actually don’t like meeting people. Human interactions require recovery afterwards.
But I have been on a tear, meeting and speaking, week after week, for the past 4 months post-MAS. Some instinct told me I needed to get out of my comfort zone, after 18 years at MAS.
The two chance encounters this week reminded me of how these life choices connect to emergence in AI.
Emergence
In complex systems, emergence is when something meaningful arises from the interaction of conditions, not from any single cause, but from all of them together.
In neural networks, emergence at scale requires three things running together. Capacity - the size of the network. Data - the diversity of inputs, how much of the world the network has seen. And connections - how the layers interact, how patterns compound across depth.
Scaling laws, strongly associated with the Chinchilla paper from 2022, showed that emergence doesn’t happen gradually. It happens at compute-optimal thresholds. Below a certain scale the network won’t work well. Then something tips. And suddenly it can.
The Realization
Being rojak. Pushing against discomfort to meet people. These are all ways to build the conditions for emergence to happen in life.
I finally understand what I was doing. Now I am just wondering what’s my threshold.