Many Different Worlds
Have you ever wondered what it would be like to be someone else?
Week 6 post-MAS. (Previous weeks in my newsletter).
Back to my usual number of conversations. The usual mix of old and new friends. But I realized how different each person’s world was. A pattern that was present in prior weeks, but I was blind to it.
I realized it when writing a post about Matt Shumer’s viral post on AI eating all jobs and that you are either early or doomed. Now, I agree with his observations on the dramatic shifts in AI, and I think many things will change. But not how his experience in the world of startups translates to all the many different worlds.
And this point on many different worlds also relates to the field of AI.
But let me walk through the different worlds I saw first.
The familiar. An ex-colleague who is now head of policy for a blockchain intelligence firm. Another now leading policy at a global payments company. A partner at a consultant who I have worked for and reached out for advice at career crossroads. A PhD scholar who I spoke to virtually, back home for the Chinese New Year. Someone who was leading AI governance that I crossed paths with years ago. Who now wanted to do a podcast interview with me on AI trust. Someone I met multiple times in MAS from a banking association. And someone I met in week 4, the one running a community at the intersection of data and AI.
New connections. A pair of experienced tech consultants from the UK building a startup for governance tools. A consulting partner who has spent years shaping how the region’s largest organizations think about risk. Another partner who has been at the forefront of finance and digital transformation. The president of a major financial association. A former banker now running his own trading company, whose dad has known my family for decades. A boutique consultant who was the chief analytics officer for a major bank, doing his own thing for a while now, helping transform analytics and business decisions. A seasoned engineer who pivoted to data and analytics. And someone who has spent years building risk screening and financial crime detection systems, now figuring out what’s next. And someone from Australia building an AI governance platform.
The Bubbles We Don’t See
Interestingly, I can see threads connecting their worlds. But most of them will probably never meet each other.
I feel like this is what it’s like on LinkedIn. We’re all on it. We’re all connected somehow, even if not directly, across 6 degrees. But the algorithm shows us different worlds. Your feed and mine look nothing alike.
Same platform, parallel realities. Even for close friends, we may have not seen each other’s posts for a while. Because the algorithm has created these little bubbles.
The blockchain policy lead and the consulting partners are both navigating a world being reshaped by technology. But their feeds share almost nothing. The president of the financial association and the person I know from the banking association are doing almost the same thing. But have never crossed paths. The boutique consultant transforming analytics and the CEO of a trading company have both pivoted from banking. But I suspect they could be seated next to each other in a coffee shop and not know their similar paths.
Matt Shumer’s take on the crazy leaps in AI capabilities is certainly true. And we should be worried. I am. But he was extrapolating from startup-land, a world where you describe an app and it appears, where errors are tolerable.
This week reminded me why that framing breaks down.
The PhD student isn’t going to graduate by just vibe coding. Research taste, judgment need to be learnt. The consulting partners I met. Will their years of experience across worlds be replaced by a LLM? I doubt so. Even the CEO of a trading company, who is enthusiastic about AI and has seen its benefits, is clear eyed about its limits. They don’t live in the same world as Matt Shumer. His world is but one of many.
AI’s Own Parallel Universes
This shouldn’t surprise me. I lived it during my PhD.
We say “AI” like it’s one field. We say “the AI race” like it’s just one race. It’s not. AI is many fields that happen to share a name.
My research sat at the intersection of graph neural networks, time series and multimodal learning. I applied it to financial networks and, oddly enough, mobile user interfaces. That meant I was publishing across communities that barely talked to each other. A paper at ACL, a computational linguistics conference where researchers care about language. Papers at IUI, an intelligent user interfaces conference where researchers care about design semantics and how humans interact with systems.
The world of AI is made of many many different worlds. Connected, but also as separate as can be.
Consider just the major conferences. NLP researchers gather at ACL, EMNLP. Computer vision at CVPR, ICCV. General machine learning at NeurIPS, ICLR. Data mining at KDD. Each with its own culture, its own networks, its own sense of what matters. Within these worlds, there are also silos. NLP researchers working on financial tasks are not the same as those working on recommendations. The time series community has its own methods and benchmarks. For tabular data, which is the bread and butter of finance, deep learning still struggles to consistently beat decades-old machine learning models. There’s a humbling reality in these worlds that barely registers in the world of foundation models and the hype we hear in the news.
Even within what everyone calls Generative AI, the thing people talk about as if it’s one thing, there are also different worlds. Researchers working on reasoning or alignment may not be the same ones working on multimodal models. There are some people who focus on architectures, others on training methods.
And then there’s the world that existed before the current hype. Machine learning was genuinely useful long before ChatGPT. They didn’t suddenly become “AI” in 2022. They always were. But the spotlight moved, and now when people say “AI” they usually mean LLMs.
I can feel the frustration in some of these communities.
The AI race isn’t one race. It’s dozens of races, on different tracks, with different finish lines. And most of the runners can’t see each other.
6 weeks. Close to a hundred conversations. I now see these invisible walls between worlds.
And more and more I think the value isn’t in any single world.
The value is in the crossings.
#DifferentWorlds #AI #AIRiskManagement #Reflections #Transitions