Thought Leaders in Artificial Intelligence: Comet ML CEO Gideon Mendels (Part 4)

Sramana Mitra: There seems to be a very large number of developers who are learning Python and are either practicing or aspiring machine learning engineers. The number is in the millions. What are these people doing?

Gideon Mendels: As in how are they learning?

Sramana Mitra: What is the application? What is the career path? Is the industry active enough to absorb five million ML developers?

Gideon Mendels: Some of the ML problems out there will require that deep sophisticated machine learning knowledge. That will continue. If you think about solving novel problems, that is typically less relevant for someone just getting into the field. When you look around the enterprise setting, there are so many areas where you can apply machine learning. A lot of these problems are not necessarily hard to solve. If you can build a simple model and get significant results, there’s a lot of opportunity for that.

A lot of companies are trying to figure out how to deal with these personas being able to do multiple things. There are people writing code. You can transition them to other roles. A lot of the teams today have that mix of talent. If you are able to get yourself to certain level with your ML knowledge and you have strong software engineering knowledge, you would be 10x more productive than someone with deep theory knowledge but no software engineering background.

Writing and testing code is a big part of the process. If you’re a good developer and you’re picking up ML, you can be very impactful. My view is, ML is an amazing tool, but it’s one tool within the toolbox. For some problems, you’re going to write If-Else statements. For some problems, you’re going to implement ML model. You shouldn’t implement it in areas where an If-Else will suffice.

If I think five to ten years in the future, my view is every software engineer in the world will be building models in some capacity.

Sramana Mitra: Python becomes table steaks for a developer career effectively.

Gideon Mendels: Yes.

Sramana Mitra: Many years ago, probably mid-2000, I had a meeting with the CTO of Azure. He was telling me that they were interested in providing as much abstraction as possible within Azure for people to be able to develop ML applications.

From where you sit, can you talk about these abstraction attempts out there? Google is one. Watson is one.

Gideon Mendels: There are different levels of abstractions. The most common one is, I’ll give you a dataset and you’ll give me a trained model that works. All the cloud providers have some offerings in that space. There are successful companies in that space. There is immense value in that, especially for that audience of more business analysts. In some ways for the easier problems, they provide a lot of value. Data Robot is the leader in the space.

The challenge is that in order to make an ML project successful, there are a couple of components. For an analyst that can code, that has a lot of value. That’s not necessarily the hard problem. The hard problem is aligning the business needs to what machine learning can do. The core of the work is spending time on data. You can’t just throw all the features on the wall. You have to understand the features. What do they represent? How do they behave?

In order for these products to be successful, there is a lot of work that needs to be done that requires ML understanding that they’re not able to abstract. That’s why you’re seeing significant services components. They bring their own data scientists and provide a solution. The fact that you need service providers to do the actual work shows the challenges. I do think there is value to it. I don’t believe that this is where the space is going.

This segment is part 4 in the series : Thought Leaders in Artificial Intelligence: Comet ML CEO Gideon Mendels
1 2 3 4 5