What is and isn’t AI
First, let’s be very clear about what artificial intelligence work is and isn’t. Feel free to skip this section if you think you’ve got a handle on it.
Artificial intelligence is an incredibly broad term — it involves an aspirational drive to replicate human learning and behavior in machines. How do we cut through the hype?
Let’s talk first about a specific element of artificial intelligence that is actionable and well-compensated: machine learning. Machine learning is a subset of artificial intelligence that involves using certain rules and algorithms to try to generalize insights from one dataset to a broader one.
You might take labeled data that is human-classified and work on extending the logic with machine learning or let the computer go through unlabelled data and figure things out for you. You might work in a form of reinforcement learning that is akin to deep learning: a specific set of machine learning approaches that use layers of reinforcement learning to get to the desired outcome.
You’ll be working with pipelines of data if you choose to get into machine learning — the skill of having machines make predictions and labels for new sets of data after absorbing certain rules from similar datasets.
Machine learning is a set of programming tools to work with data and deep learning or reinforcement learning is a subset within that. While there are models programmed with pre-set rules to take data and process it a certain way — for example, a linear regression model that can tell you how much a dependent variable is affected by an independent variable (rent dependent on the number of rooms in an apartment, for example) — deep learning approaches tend to use semi-structured models to evaluate data at scale that works a bit like the human brain.
The critical distinction is that deep learning will operate through multiple layers of feedback. A neural network-like deep learning model will self-correct and optimize toward a certain outcome, tuning itself so that its output gradually matches its input through the self-modification of weights within the model.
This is perhaps best illustrated with the simplest deep learning model: the perceptron, illustrated above. In this case, from a series of inputs, the layer of hidden calculations performed between the input and the output self-modifies until it arrives at the desired output.
Why does this matter? It forms the basis of all sorts of exciting AI innovations you’ve heard of, from self-driving cars to video/image recognition. By creating increasingly efficient models that help machines manage the complexity of data patterns that can stretch into trillions of possibilities, humanity can benefit from automated processing of ever-larger data — gaining richer insights on datasets that can grow larger and larger. Those insights can allow a social network like Facebook to automatically classify the photos on its network, or allow somebody to pattern-match and predict your behavior based on your search history.
However, despite all the hype, deep learning approaches are nowhere close to how scientists think the human mind actually operates.
Let’s take a step back and define all of these terms so we know exactly what we’re talking about:
Data science involves using statistics and theory to treat large datasets so you can get a business answer or prediction based on the underlying dataset.
Artificial intelligence is the broad aspiration of granting machines human-like learning and reasoning. Much of it is a theory at this point, rather than something practical and implementable.
Machine learning is a way of creating predictive models that learn without needing to be explicitly programmed to do so, an actionable subset of artificial intelligence. You can think of machine learning models as semi-structured objective functions, wherein a data scientist will train a model for a certain outcome, without having to explicitly plug in all of the variables and interactions required. The model understands it is trying to minimize a certain amount of error and adjusts accordingly
Deep learning is a subset of machine learning and specifically refers to models like convolutional neural networks that reconcile an input and output with dense hidden layers that perform self-correcting levels of calculations in order to come to the desired outcome. In practice, in production, the number of layers and calculations performed is exponentially high.
There are traditionally two fundamental splits here when you’re working with datasets and artificial intelligence/machine learning:
Data scientists, who help tailor the business logic of the models that are being created. Basically, data scientists help communicate findings from data models to business decision-makers and they help tune and tailor models that help businesses ask the right questions of their data.
Machine learning engineers build the data plumbing that allows for data scientists to process and work with huge reams of data that continually updates. In practice, they’re responsible for feeding the models defined by data scientists with the data they need to perform well, and they’re often responsible for taking theoretical data science models and helping scale them out to production-level models that can handle the day-to-day of companies that generate terabytes of data.
I’ll break it down into more detail, but as a rule of thumb, even if the two broad roles share some overlap, a data scientist is often going to be working with the theory behind the data science of artificial intelligence, while machine learning engineers will implement models in practice. Data scientists tend to have a stronger theoretical foundation in machine learning, statistics, and mathematics, while machine learning engineers typically have a stronger software engineering background.
You’re going to have to take one of these broad roles if you’re going to work with artificial intelligence models.
Artificial intelligence is highly scientific. After all, mimicking the human brain using machines is a very tough problem to solve, much less master. The skills that you will need to pursue AI as a career are varied, but all of them require a great deal of education, training and focus.
That said, there is a wide variety of career types available in AI and machine learning, and they range from higher-level research to low-level programming and implementation.
For example, researchers use their breadth of knowledge in theory and study to reveal new types of systems and capabilities. Researchers hypothesize new or different ways for machines to think and test their research for real-world feasibility.
Algorithm developers take AI research and transform that research into repeatable processes through mathematical formulas that can be implemented using hardware and software.
Software developers and computer scientists use those algorithms to write sophisticated pieces of software that analyze, interpret and make decisions.
Hardware technicians build pieces of equipment (like robots) to interact with the world. Robots use its internal software to move and operate.
Most careers in artificial intelligence require coursework and experience in a variety of math and science-related topics like:
- Math: statistics, probability, predictions, calculus, algebra, Bayesian algorithms and logic
- Science: physics, mechanics, cognitive learning theory, language processing
- Computer science: data structures, programming, logic and efficiency
Want a career in AI? Then read. A lot.
Read papers and case studies. Experiment with technologies like Map-Reduce, PHP, MySQL, Postgres and “Big Data”, especially if you are targeting a computer science-related career in AI. Expose yourself to as many technologies as you can.
Pro tip: Browse through AI job opportunities. Read the job descriptions and especially the requirements to get a feel for specific qualifications that you need for that job.
For example, some might need experience in low-level programming languages like Python or MatLab. Others, especially in the healthcare industry, need expertise in data services like Spark and Blockchain.
Regardless of the type of job that you’re after in artificial intelligence, there is no better way to figure out the exact skills you need than to read job requisitions and stay as up-to-date in the industry as possible.
Use the Job Search tool here on The Ladders to find AI and machine learning jobs.
The Best Non-Tech Skills for AI
Though the types of careers in the AI industry are varied, most professionals in AI possess five key skills and capabilities, regardless of their individual roles.
Most AI professionals:
Are highly critical thinkers. They take nothing at face value and are naturally curious. They believe in trial and error and must test and experiment before making a concrete decision.
Like to push the envelope. AI is all about pushing the boundaries. Pegging the capabilities of hardware and software to their max, always looking for more. More ways to improve existing systems. More ideas for inventing new ways to live.
Live naturally-curious lives. Always wanting to know more, artificial intelligence pros want to know how things work. They don’t just look. They observe. They don’t hear. They listen.
Don’t get easily overwhelmed. They understand that artificial intelligence is highly technical, but also realize that venturing into uncharted waters is difficult and mysterious. They enjoy the process rather than getting frustrated by it.
Love math and science. AI is highly technical and it’s a natural good fit for those who are gifted and interested in hard sciences and mathematics.
Artificial intelligence is not just about replacing the human component of the industry. It’s also about making it easier to make decisions based on observable patterns, use logic and reasoning to form conclusions and build pathways to boost efficiency and production.
It is not an easy discipline, but that’s also why salaries in the AI industry are much higher than average. It takes the right type of person with the right skill set to excel.
Are you the type of person who’s right for a career in AI? If you have many of these skill sets, then you just might be.