写一篇AI & Philosophy article，准备post to medium. 主题是关于在AI时代，model都基本上被LLM取代，那么industry 中 applied scientist这个role到底该如何定义。

Peter Norvig, author of "Artificial Intelligence: A Modern Approach", 根据他写这本textbook都演变，将AI发展归结为3个时代（reference：https://www.youtube.com/watch?v=G5BRhZIAl_Q) :

Algorithm-centered: 我还记得20年前，是AI algorithm百花齐放的时代，AI/ML课程需要学习各种algorithm，比如SVM，Decision Trees等等
Data-centered：自从Deep Neural Network以及GPU开始普及起来，基本上就不需要更多的algorithm了，基本上就是根据已经有的DNN架构，调整一下network layers即一些hyperparameters就可以了，最重要的是如何找到更好的data。algorithm这时候已经不是特别重要了，因为基本上converge到DNN
Human- centered：自从LLM开始普及以来，连Data都不需要了，因为LLM都是pre-trained, 写code也完全由AI来取代，这个时候最重要的是“decide what needs to be built" (另见Andrew Ng "Product Management Bottleneck" https://www.linkedin.com/posts/andrewyng_grok-raises-questions-meta-poaches-talent-activity-7351623555542896640-cTJT).

我在AWS Agentic AI工作，做的是human-centered AI相关的工作，这个工作经常需要深入了解customer issues, 从而理解应该如何开发AI产品来帮助customer解决一些pain points，也会经常做一下eval，metrics，dashboard相关的工作。我也已经有一段时间没有参与model的开发。经常会有人问我，你做的这些工作（比如work with customers，eval，开发dashboard这些）别的software engineer也都会做，也并不是model development，那么applied scientist同software engineer有什么区别？

我觉得这种将applied scientist等同于build models是一种错误的narrative，这就像说software engineer就是写code一样，而software engineer更重要的是作出high available，secure，scalable，efficient的系统。

那么applied scientist的role到底是什么呢？我的看法是，applied scienist是要research the unknown，answer questions that no one has answers to. 这就包括define meaningful research questions, 根据经验和直觉来形成hypothesis，通过严谨的方法验证hypothesis，形成一套体系，来回答从来没有人知道答案的问题。这才叫做applied scientist。相反，利用已经有的方法来train/build model，虽然是做modeling相关的工作，但那更像是ML engineering而不是applied scientist的工作。

AI-assisted Edit Version: 2026-04-10 19:52:20 America/Detroit

Instructions: prepare a post to Medium. Use English. Find an engaging title (and give recommendations on different options)

Title Options:

Option 1 (Recommended): "Beyond Model Building: Redefining the Applied Scientist in the Age of LLMs"

Option 2: "What Do Applied Scientists Actually Do When AI Writes the Code?"

Option 3: "From Algorithms to Answers: The Evolution of Applied Science in the AI Era"

Option 4: "The Applied Scientist's Identity Crisis: Why We're More Than Model Builders"

Beyond Model Building: Redefining the Applied Scientist in the Age of LLMs

As someone working in Agentic AI at AWS, I've been asked countless times: "If you're not building models, what makes you an applied scientist rather than a software engineer?" It's a fair question that reflects a common misconception about what applied scientists actually do—especially in an era where large language models (LLMs) have fundamentally transformed our field.

The Three Eras of AI

Peter Norvig, co-author of the seminal textbook "Artificial Intelligence: A Modern Approach," has articulated a compelling framework for understanding AI's evolution through three distinct eras (reference):

1. Algorithm-Centered Era

I still remember 20 years ago when AI was in its algorithm-centric golden age. AI and machine learning courses required mastering a diverse toolkit: Support Vector Machines, Decision Trees, Random Forests, and countless other techniques. Each problem demanded careful consideration of which algorithm to apply.

2. Data-Centered Era

With the rise of Deep Neural Networks and GPU computing, the algorithmic landscape consolidated dramatically. Instead of choosing between fundamentally different approaches, practitioners focused on adjusting network architectures, tuning hyperparameters, and—most importantly—curating better datasets. Algorithms converged toward variations of deep learning, and data became king.

3. Human-Centered Era

Today, with the proliferation of LLMs, even the data-centric paradigm is shifting. Pre-trained models handle much of the heavy lifting, and AI can even write code. As Andrew Ng has noted in his discussion of the "Product Management Bottleneck", the critical question has become: "What needs to be built?"

The Identity Question

In my current role working on human-centered AI, I spend significant time understanding customer pain points, developing evaluation frameworks, designing metrics, and building dashboards. I haven't been directly involved in model development for quite some time. This naturally raises the question: if these activities—customer engagement, evaluation, dashboard development—are things software engineers also do, and if I'm not building models, what distinguishes an applied scientist from a software engineer?

The Flawed Narrative

The assumption that "applied scientist = model builder" is as reductive as saying "software engineer = code writer." Just as software engineers are fundamentally concerned with building highly available, secure, scalable, and efficient systems—not merely writing code—applied scientists have a deeper purpose that transcends model development.

Redefining the Applied Scientist

The core mission of an applied scientist is to research the unknown and answer questions that no one has answered before.

This involves:

Defining meaningful research questions that address real-world problems
Forming hypotheses based on experience, intuition, and domain knowledge
Rigorously testing these hypotheses through systematic experimentation
Building frameworks and methodologies that provide answers to previously unanswerable questions

Training or building models using established methods, while certainly involving modeling work, is more accurately characterized as ML engineering. Applied science, by contrast, is about venturing into uncharted territory—creating new knowledge, not just applying existing techniques.

The Path Forward

In the LLM era, applied scientists have an opportunity to focus on what we do best: asking the right questions, designing rigorous experiments, and generating insights that push the boundaries of what's possible. The tools may have changed—from hand-crafted algorithms to pre-trained models—but the fundamental mission remains the same.

We are not defined by the models we build, but by the unknowns we illuminate.

What are your thoughts on the evolving role of applied scientists? How has your work changed in the age of LLMs? I'd love to hear your perspectives in the comments.