Which AI Programming

Languages Should I Learn?

by DeepLearning.AI

Coding is a must-have skill for anyone

building AI products. It enables you to bring
your machine learning ideas to life. Learning
to code is fun and empowering, but it also
requires time and effort. The last thing you
want to do is start learning a language only
to realize weeks or months later that the job
you want actually calls for a different
language.

Simply put: Deciding which language to

start with can be intimidating.

Not to worry! This article will explain the

basics behind the most popular
programming languages used in AI and
help you decide which to learn first. For each
language, we will describe its basic features,
what it does well, where it falls short, and
which sorts of jobs use it most.
A Basic Roadmap to
Programming Languages for AI

The five most important programming

languages in AI are Python, C++, R, MATLAB,
and Java. Before we dive deep into each of
them, let’s explore which to learn first.

For most people, the first programming

language to learn is Python. It’s easy to
learn, extremely adaptable, and has
numerous libraries specifically for machine
learning. For those reasons and more, it is
the de facto coding language in AI. What to
learn next depends on your career goals.
A Basic Roadmap to
Programming Languages for AI

After Python you should learn:

• C++ — If you want to work in robotics, self-

driving cars, or hardware.

• R — If you want to work in academia or

the financial industry.

• MATLAB — If you wind up working at a

company that still uses MATLAB. (You
should convince that company to switch
to Python.)

• Java — If you want to build scalable AI

infrastructure.

One last bit of advice: Don’t try to learn two

languages at once. Focus on getting good
at Python first. After you have reached its
limits, branch out depending on your career
goals.
Python

The best all-around programming language for AI.

What is it? Python is a popular, general

purpose programming language that is
relatively easy to learn. Its simplicity lends
itself to AI development, and the AI
community has adopted Python as its de
facto language.

What does it do well? Python is popular for

several important reasons.

• It’s easy: Compared to other coding

languages, Python has a simple syntax
(the words, symbols, and expressions
you’ll type to create programs). This
means you’ll have more time to devote to
the stuff that matters: Looking at data and
tuning your models.
Python

The best all-around programming language for AI.

• It’s versatile: Your operating system

supports Python, whether you use iOS,
Windows, or Linux. What’s more, you don’t
need to modify a Python program much
to get it to run across platforms.

• It’s open source: Anybody can adapt,

update, or add to the code that underlies
Python. As a result, many members of the
Python community have built frameworks
and libraries that make it adaptable to
nearly any machine learning or data
science task.
Python

The best all-around programming language for AI.

What are the downsides? Python has

limitations when it comes to performing
complex mathematical and statistical
functions. It also runs slowly compared to
languages like C++ and Java (see below).

Who is it for? We’ll say it one more time:

Python is the most popular programming
language in machine learning and data
science. If your job involves building
machine learning models and working with
lots of data, Python is for you. However, you
may want to specialize in a second
language if you work in data analysis, AI
infrastructure, or plan on doing more
intensive programming outside of your core
AI work. Keep reading to learn more.
C++

The best programming language for AI

infrastructure.

What is it? C++ is one of the most popular

languages for general-purpose
applications. It is the backbone of operating
systems like Windows, iOS, and Linux; apps
like Spotify and Photoshop; sites like
YouTube; along with video games, banking
systems, and more. It is also an essential
language for anyone working in self-driving
cars or robotics.

What does it do well? C++ is a compiled

language: It doesn’t require an interpreter
program, which adds processing overhead.
In practical terms, programs written in C++
are fast and efficient.
C++

The best programming language for AI

infrastructure.

What are the downsides? C++ programs

may be efficient to run, but writing them is
complex — with a capital C and two pluses.
Writing a program in C++ takes time,
debugging it often takes even more time,
and rewriting it every time you adjust your
hyperparameters takes more time than you
think it will. C++ is also notoriously difficult to
learn. If your goal is to be a data scientist,
learning C++ is like learning to fly a
helicopter and then using it only to shop at
the grocery store.

Who is it for? C++ isn’t well suited for most

data scientists or machine learning
engineers, but it is essential for some
disciplines within AI.
C++

The best programming language for AI

infrastructure.

• If your role involves building or

maintaining AI infrastructure — the core
software libraries that others use to
deploy models or analyze data — then
you should absolutely learn C++.

• Most teams working on robotics and self-

driving cars use C++ for their production
code due to its speed and efficiency.

• Most jobs in the chip and hardware

industry call for C++.

It’s important to learn C++ if you want to get

involved in the open source community.
Many of the most popular Python
frameworks and toolkits are programmed in
C++.
R

The best programming language for data

analysis.

What is it? R was built specifically for

statistical analysis.

What does it do well? R was developed by

statisticians, for statisticians. It excels
at finding patterns in data and deriving
insights from model outputs. For obvious
reasons, R also appeals to machine learning
engineers and data scientists who use it for
statistical analysis, data visualizations, and
similar projects. Like Python, it is open-
source, and the community has created a
number of frameworks and libraries for AI
tasks.

What are the downsides? In terms of complexity,

beginners will find R more difficult to learn than Python. R
has more built-in features for crunching numbers than
Python, but it also tends to lag when processing projects
that use too much data at once. R? More like Arghhh!
R

The best programming language for data

analysis.

What are the downsides? In terms of

complexity, beginners will find R more
difficult to learn than Python. R has more
built-in features for crunching numbers
than Python, but it also tends to lag when
processing projects that use too much data
at once. R? More like Arghhh!

Who is it for? R is a great tool for data

analysis, data science, and adjacent
professions, but it’s often used by
academics. You might also be required to
learn R if you get a job working in finance,
and for teams that use it in their legacy
software.
MATLAB

A once-popular AI programming language that

has been mostly eclipsed by Python.

What is it? MATLAB is more than a

programming language, it’s a five-part
system that consists of a language,
development environment, graphics
visualizer, math library, and interface for
writing programs in other languages.
MATLAB focuses on matrix computation. If
you aren’t familiar, matrices are arrays of
numbers, and the ability to compute them
well is central to many machine learning
and data science applications.
MATLAB

A once-popular AI programming language that

has been mostly eclipsed by Python.

What does it do well? MATLAB is excellent for

working with numerical arrays. It is also
excellent for many other mathematical
operations, and it has built-in features for
implementing machine learning models.
These features give it a leg up over Python,
which requires add-on toolkits and
frameworks for both mathematical
functions and model implementation. In
terms of speed, it is fast and easily outpaces
Python in many operations.

What are the downsides? The biggest

drawback to MATLAB is the cost: You have to
pay a fee to access the system and possibly
for additional functionality depending on
your goal.
MATLAB

A once-popular AI programming language that

has been mostly eclipsed by Python.

This barrier to entry has partly contributed

to MATLAB’s declining popularity among AI
builders; Python, after all, is free and open-
source. Lastly, MATLAB’s syntax is difficult to
learn compared to Python’s.

Who is it for? Some employers and AI teams

prefer MATLAB either because they are
legacy users or their goals require more
mathematical oomph. Fun fact: Andrew Ng’s
original Machine Learning course
was taught using MATLAB. When Andrew
and his teams at DeepLearning.AI and
Stanford University modernized the Machine
Learning Specialization in 2022, one of the
key upgrades was switching to Python.
Java

A fast, versatile programming language that is

useful for building scalable AI infrastructure.

What is it? Java is similar to Python in many

ways: It is popular, open-source, and has
many frameworks and toolkits specifically
for machine learning and data science.
Java is much older, and therefore it has
legacy buy-in from many organizations. It is
also more technically complex than Python.
This means it is more difficult to learn than
Python and R, but it can execute programs
much more efficiently.

What does it do well? Everything Python can

do, Java can do just as well — maybe better,
in some cases. For instance, it has
frameworks for data science, classification,
deep learning, and more. Java has more
rules than Python, which makes it more
difficult to break or misuse the code.
Java

A fast, versatile programming language that is

useful for building scalable AI infrastructure.

It is useful for building full stack, back-end,

large scale infrastructure for deploying
machine learning models.

What are the downsides? Java has a

steeper learning curve than Python (though
not as steep as C++). Writing programs in
Java also takes more time than Python, and
those programs often require more
debugging due to their complexity. These
are speed bumps that slow down rapid
prototyping of machine learning models.
Finally, Java’s community isn’t as active as
Python’s when it comes to developing AI-
focused tools, and as a result Java is useful
for a narrower range of machine learning
and data science tasks.
Java

A fast, versatile programming language that is

useful for building scalable AI infrastructure.

Who is it for? If not for its complexity, Java

might be the dominant AI coding language.
However, you probably only need to learn it
if you are targeting a role building AI
infrastructure or deploying machine
learning products. For beginners, Python is a
safer bet. In the meantime, you can still
enjoy java in your mug each morning.
Tips from Andrew Ng on Learning
to Code

One of the best ways to learn to code is by

taking on a coding project. Start small,
Andrew Ng advises, with a project you can
finish over a week or two in your spare time.
The goal isn’t to build a world-changing
app, it’s to put your knowledge into practice
and learn from your mistakes.

When working on projects, it’s common for

coders to use Google or Stack Overflow to
find pre-written lines of code that suit their
needs. This is an efficient way to work, and
you should feel free to do the same.
However, don’t just copy and paste what
you find, Andrew Ng said. Instead, retype it
yourself. The physical act of doing so will
build muscle memory for your brain, helping
you to internalize the concept and syntax.
Keep doing this, and you’ll be the one
posting bits of code that other people copy!
Tips from Andrew Ng on Learning
to Code

In a 2020 letter to readers of his

newsletter, The Batch, Andrew Ng wrote:

When you’re trying to master a

programming technique, consider these
practices:

• Read a line of code, then type it out

yourself. (Bonus points for doing it without
looking at the reference code while
typing.)

• Learn about an algorithm, then try to

implement it yourself.

• Read a research paper and try to

replicate the published result.
Learn a piece of math or a theorem and try
to derive it yourself starting with a blank
piece of paper.
Conclusion

Coding is an essential skill for AI builders. In

fact, Andrew Ng has compared coding to
literacy:

Code is the deepest form of human-to-

machine communication. As machines
become more central to daily life, that
communication becomes ever more
important.

So, which language should you learn? For

most machine learning engineers and data
scientists early in their careers, the best
choice is Python. It is easy to learn, quick to
implement, and has a ton of add-ons that
are tailor-made for AI. You may be tempted
to learn a bit of Python, then learn a bit of R,
a bit of Java, and so on in order to be more
versatile. We recommend against this.
Focus on getting good at Python before you
switch things up.
Ready to Start?

Are you a Python beginner?

Consider AI For Everyone or Generative AI for

Everyone!

Neither assume a background in any

particular programming language.

Already familiar with Python?

Try the Machine Learning Specialization to

learn the fundamentals of AI.

Are you an experienced coder looking to get

up to speed on large language models?

Check out Generative AI with LLMs to learn

about the latest developments and use
cases.
Ready to Start?

Not sure where to start?

Try one of our short courses!

• Pair Programming with a Large Language

Model

• Building Generative AI Applications with

Gradio

• Building Systems with the ChatGPT API

• LangChain for LLM Application

Development

• Large Language Models with Semantic

Search

• ChatGPT Prompt Engineering for

Developers

• And more!