What’s It Like to Work in Applied AI?

Author(s): Branden Lisk

Originally published on Towards AI.

Understanding the role of a Machine Learning Engineer from an early-career perspective

Introduction

Disclosure and Accuracy

This article is written from the perspective of an early-career stage: 2+ years in the tech space, 1+ years of applied AI/ML experience, with an (expected) bachelor’s level degree in engineering. The bulk of experience lies in start-up, small, and medium-sized environments, which are inherently less structured than established organizations.

The author strives to maintain accurate information that is useful to a general audience. However, note that information on this topic evolves quickly, and practices can be wildly different from organization to organization, so there is an abundance of conflicting opinions from various sources.

Intended Audience

This article is intended for other early-career engineers in the field of applied AI, software engineers interested in a career transition, and anyone else with interest in this topic.

What is Applied AI?

There are plenty of existing resources to dig deeper into this, but in summary, applied AI is:

“…the branch of artificial intelligence that brings it out of the lab and into the real world…” [1]

In essence, applied AI is taking scientific research and applying it to business problems.

What is a Machine Learning Engineer?

It would be helpful to start with a definition of machine learning. It’s defined as:

“…the process of using mathematical models of data to help a computer learn without direct instruction. This enables a computer system to continue learning and improving on its own, based on experience.” [2]

Essentially, “AI” is the idea of an “intelligent” machine. Machine learning is the process used to develop that “intelligence”. So in this context, a Machine Learning (ML) Engineer designs and builds the infrastructure needed to enable the process of machine learning.

We’ll discuss more on the particular responsibilities of an ML Engineer later.

Expectation vs. Reality

How does this differ from how we traditionally understand the role of an engineer who works in the field of “Machine Learning” and “Artificial Intelligence”? It’s likely we think of someone who does the following:

• a lot of research and reading into published papers,
• approaching new and novel problems,
• planning and conducting experiments to invent new methods to solve these problems,
• publishing new papers.

These responsibilities are actually fulfilled by someone we’d call a “Research Scientist” or even a “Data Scientist”. We’re falsely attributing theory to practice.

In reality, this is actually what we can expect to see in practice:

We can see that staying informed of current research and approaches to problems by reading papers is unchanged. However, each other point has an analogous responsibility more reflective of a machine learning engineer’s day-to-day.

We’ll talk more about these distinctions a bit later. First, we have to discuss some industry jargon that is thrown around in the “Applied AI” space.

Convoluted Separation of Responsibilities

Job Titles

This is the area where you’ll experience the most conflicting opinions. The separation of responsibilities according to different job titles is not consistent between organizations or even sub-teams within the same organization. Trying to understand different descriptions and convoluted diagrams is largely a waste of time and will make you even more confused about the distinctions. Trust me, when I tried for this article, I started to question what my own job even is. You’ll hear some subset of the following titles:

• Data Engineer
• Data Scientist
• Applied Data Scientist
• Data Analyst
• Business Analyst
• ML Engineer
• MLOps
• Data Infrastructure Engineer
• DevOps

Organizations describing job postings, as well as individuals describing past experiences, use many of these terms interchangeably, so it’s very hard to find consistent data.

At the risk of adding more confusion to this argument, I’ll define my own generic titles that have no major significance outside of this article, according to how they fit into the data science process.

Data Science Process

We can see five distinct “roles” in the above diagram:

• Business: This stage involves understanding and formulating the business problem and defining the goal of the project. This role would typically be called a “Business (and/or Data) Analyst”.
• Data: This stage involves iterating until the desired amount and quality of data is achieved for feature engineering. These include understanding potential data sources that could be used to solve the business problem (assessing key factors such as availability, usability, quality, and cost), collecting said data, and cleaning (mitigating outliers, resampling, etc.) the collected data. This would be assigned to a combination of the “Data Engineer” and “Data Labeler”. Building the data pipeline would typically be assigned to the “Data Infrastructure Engineer”.
• Model: This stage involves two parts: feature engineering and modeling. The collected data is used to perform feature engineering. A model is selected, built, and trained. The model is then evaluated against the metrics set in the “Business” stage. If it does not achieve the required metrics, we have to assess the problem and return to a previous step. This role has the most inconsistency in terms of job title. It can be any of: “Data Scientist” (the most incorrect), “Applied Data Scientist,” or “Data Analyst”.
• Operations: This stage involves deploying, serving, and monitoring the completed model. Monitoring is a continuous action for the lifetime of the model. This role would traditionally be called “DevOps” (software development and IT operations) but is typically called “MLOps” (machine learning operations) for machine learning projects.
• Maintenance: Finally, if monitoring detects the model is no longer achieving the desired metrics, we must revert to a previous stage in the process.

Covering the missing title from above, the responsibilities of an “ML Engineer” are Data, Model, Operations, and Maintenance, typically with an emphasis on Data and Model.

Early-Career Perspective

Now that we understand what “applied AI” is from a high-level and the common roles you’ll hear in the field, what does this mean for you? In this section, we’ll discuss what you can expect as an early-career ML Engineer.

Common Misconceptions

We’ll start by walking through some misconceptions about the field.

Machine learning is heavy in statistics. You need deep statistical knowledge to work in the field.

If you tell people you work in machine learning, you’ll often hear, “wow, you must be really good with statistics and math,” and my response is usually, “Maybe? But it’s not incredibly relevant for my job”.

This is to say that this statement is true to an extent but heavily exaggerated. At its backbone, machine learning is just a new application of old mathematics: statistics, linear algebra, and calculus. It’s important to understand the fundamentals to understand how to apply machine learning truly effectively.

However, this only goes so far. As an ML Engineer, will you ever need to implement backpropagation from scratch? Probably not. Modern ML frameworks abstract all of this fundamental understanding away (for better or for worse). Will you need to understand the math used in research papers (at a high level)? Absolutely. This is required knowledge to adapt their approach to your own business problem effectively.

You need a graduate-level degree to work in the machine learning field.

This is untrue from an early-career perspective in the applied AI field. Many that work in this field do have advanced degrees, but that doesn’t imply that you need an advanced degree. Many that I have worked with prove that this isn’t necessarily true (at least anecdotally).

Practical machine learning engineering is quite a bit more traditional software engineering than you’d think.

Machine learning is all research.

Machine learning engineers are engineers at their core.

“Engineers are individuals who combine knowledge of science and mathematics to solve technical problems that confront society.” [5]

ML engineers utilize scientific research to solve business problems. They care more about how to make the solution work in practice. Most problems are solved problems, and understanding and adapting existing solutions to a business use case is the primary goal.

Where do you fit in?

As an early-career machine learning engineer, where in this process do you fit? This can heavily depend on the organization, but I’ll describe this as generally as possible.

Referring to the “Data Science Process” diagram, I’ll break this down into individual sections.

• Business: Generally, you will not be involved in this stage. You may be approached to discuss technical feasibility and, in some rare cases (specifically start-ups), may be involved in prioritizing business problems.
• Data: You’ll be quite involved in this stage—specifically, understanding data sources and cleaning data. Data collection is often outsourced in larger organizations (if it’s needed at all). In rare cases (again, start-ups), you’ll collect the data yourself. This is generally the most lengthy (and most important) stage.
• Model: You’ll be very involved in this stage. Feature engineering and modeling are generally your full responsibility.
• Operations: You’ll likely not be involved in this stage. Most organizations have a dedicated DevOps team that will handle this. The exception would be configuring model monitoring tools specific to your application.
• Maintenance: You’ll be responsible for future maintenance of your deployed models. I’d suggest you make this process as easy as possible for yourself (if you’re wondering what could go wrong, check out a previous post).

Conclusion

We’ve seen what “Applied AI” is and the job titles of those who work in the field. We also saw the various stages of the data science process and where the “Machine Learning Engineer” fits in. Using this knowledge as a foundation, we explored the key responsibilities of an early-career machine learning engineer and some common misconceptions about the field.

I hope this article paints a realistic picture of the early-career environment of the applied AI field and helps guide you on the next step of your career.

This is just the beginning! If you enjoyed the article, follow me to get notified of the next ones! I appreciate the support U+2764️

Branden Lisk — Medium

Read writing from Branden Lisk on Medium. Machine Learning R&D Engineer U+1F916 U+007C Passionate about building cool and…

medium.com

Sources

[1] “Applied Artificial Intelligence,” Cognizant Technology Solutions. [Online]. Available: https://www.cognizant.com/us/en/glossary/applied-ai. [Accessed: 24-Jul-2022].

[2] “Artificial Intelligence vs. Machine Learning,” Microsoft Azure. [Online]. Available: https://azure.microsoft.com/en-us/solutions/ai/artificial-intelligence-vs-machine-learning/#introduction. [Accessed: 24-Jul-2022].

[3] C. Nantasenamat, “The Data Science Process,” Towards Data Science, 27-Jul-2020.

[4] A. Burkov, Machine Learning Engineering. Québec, Canada: True Positive Inc., 2020.

[5] D. Wang, “The Engineering Profession,” in ECE 190, 2018.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI