Machine learning engineers are an increasingly valuable resource as companies continue to turn to artificial intelligence (AI) technologies to help reduce costs, increase efficiency, and gain value from their data. According to a recent RELX survey, 63% of companies reported that AI positively impacted their business’ ability to stay resilient during the pandemic, and approximately 7 in 10 companies increased their AI technology investments over the last year.
But as with any new innovation, transforming promised value into a reality relies heavily on having the right people in place to support it. And top AI talent is in short supply—nearly two in five companies cite a lack of technical expertise as a major roadblock to using AI technologies.
Demand for AI skills outpaces availability
One of the most significant problems currently facing companies is that the demand to fill AI positions is much higher than the number of qualified candidates. For example, data from Indeed.com shows there are three times more job postings for AI-related roles than job searches, and postings for AI jobs have grown as much as 12 times faster than searches for AI positions from 2016 to 2018.
The interest in machine learning as a career has also skyrocketed. U.S. Google searches from the last five years for ML engineering jobs hit an all-time high in January 2020.
Other search terms like “how to become a machine learning engineer” and “machine learning engineer salaries” have experienced over 5,000% growth for the same period.
However, despite interest and demand, there is a shortage of skills, knowledge, and relevant experience. Deloitte’s State of AI in the Enterprise report estimates there are less than 40,000 top-tier AI experts on LinkedIn and roughly 300,000 AI practitioners worldwide. It’s no surprise that AI developers and engineers remain the most sought-after role at companies implementing AI, regardless of whether organizations are in the early stages of adoption or more experienced.
Given the global talent pool is small, what are the key qualities that distinguish an exceptional ML engineer from a pretty good one? And if you are able to hire these talented folks, what can you do to help them be successful?
How to find the best ML candidates for the job
As you might imagine at Google, we know a thing or two about hiring remarkable machine learning talent, but we face competition from startups and institutions on the forefront of AI research. We asked some of our engineering managers what makes the best so good. Here are the top qualities Google ML engineers all share, and what you should look for when hiring them:
- They have a keen understanding of distributed systems architecture. Machine learning workloads can be massive, requiring large datasets, parallel data processing, and distributed training. Knowing how to effectively harness storage, network, and compute resources can significantly speed up the process and reduce costs.
- They decompose ML solutions into a modular architecture. Similar to software development, modularity is also important when deploying machine learning systems. It not only enables team members to work efficiently on individual parts, it also encourages reusability for other teams or future projects.
- They recognize the value of testing. A top-tier ML engineer will emphasize rigorous testing at each stage of the ML production process, from validating input data to model performance to integration code.
- They always keep security top of mind. AI and ML introduce a unique set of security risks due to automation, massive volumes of complex data, and managing and processing workloads in the cloud. ML engineers need to consistently apply a security mindset to ML development processes, ensuring that all training data, tools, and communication channels are secured and managed properly.
- They are great communicators. Machine learning engineers work at the intersection of multiple disciplines, so effective communication is essential. They will need to explain status, risks, trade-offs, and much more to different audiences, including data scientists, developers, managers, and business users.
Great ML engineers honor the complex reality of advanced technologies. Having a passion for diversity of input and fostering a culture of support distinguishes average from excellent ML engineers.
- They understand what “good enough” is. There are so many things that can be improved and automated in an ML system, but it’s important to be able to recognize when the effort is exceeding the value. Great ML engineers need to maintain focus on the goals and objectives of a project—and they are pragmatic about when it’s time to stop. Often they will find greater return on investment by moving onto the next model versus spending the time necessary to fit the current model perfectly.
- They voice their needs clearly. Likewise, an ML engineer should speak up when they need internally-focused tools and productivity enhancements. The business always wants more features as soon as possible, but often fails to provide the tools or processes required to make it happen. The majority of models take 1-11 weeks to deploy a single ML model, and 26% of practitioners believe that delays can be caused by lack of executive buy-in. ML engineers may be required to make tough asks for investments in areas that won’t pay off right away, but will ultimately allow them to be more productive in the long-term.
- They are flexible. Machine learning projects can hit all kinds of roadblocks, such as getting access to sufficient data or building models that aren’t accurate enough to meet business needs. Being able to easily change tactics to overcome obstacles without getting frustrated or losing sight of the final goal is key for being able to deliver projects. Demonstrated flexibility with respect to tooling is also an asset. Experience with more than one framework, such as TensorFlow, PyTorch, and scikit-learn, is another positive signal in top talent.
- They are curious and creative problem solvers. It is inevitable that things will go wrong, and the best ML engineers will need to creatively look at machine learning, data, and software elements of problems. Sometimes, an issue might appear to be a data science problem (false positives) when it’s actually a subtle issue hidden further upstream in the ML pipeline that is leading to poor results. A strong ML engineer has to be comfortable with investigating across a spectrum of possible root causes and have the perseverance to keep asking questions.
- They are strong mentors. As there are so many opportunities for growth in this new field, an exceptional ML engineer acts as a mentor for others by sharing their unique perspective and knowledge. Their experience working with complex systems and diverse stakeholders makes them a great source of wisdom within the organization.
- They have a humble approach. AI is changing rapidly and our world is not static. We try to consciously remember that we are always learning, and while it’s an unreachable goal to certify a product as perfect, we can always improve. ML practitioners must make difficult decisions over time, such as when Google removed gender labels from our Cloud Vision API. These changes can be hard to explain, both internally and externally, but great ML engineers honor the complex reality of advanced technologies. Having a passion for diversity of input and fostering a culture of support distinguishes average from excellent ML engineers.
We’ve found that even among engineering communities steeped in data science, machine learning is often a new skill for many. A recent survey from Kaggle, the world’s largest community of data scientists and ML practitioners, reveals slightly more than 55% of data scientists in the community have less than three years of ML experience. And less than 6% of professional data scientists have been using ML for a decade or more.
With such a small global talent pool, recruiting and hiring exceptional ML talent is challenging—but it’s not impossible. While learning how to identify potential and seeking out the traits above can help companies ensure they are bringing on high-quality ML engineers, companies also have choices in the platforms and tooling that they select.
Tooling such as Google’s Cloud AI Platform can help solve many of the scalability, security, and developer velocity challenges. Combining great talent with great tooling will maximize the ability for your employees to thrive and bring value to their machine learning projects.
Keep in mind… AI is a team sport!
AI success is not exclusively predicated on having the best ML engineers. According to a 2021 Rackspace survey, a mere 17% of respondents report mature AI and ML capabilities with a model factory framework in place. In addition, the majority of respondents (82%) said they are still exploring how to implement AI or struggling to operationalize AI and ML models.
Successful AI projects require input beyond engineering. Most AI projects (VentureBeat cites 87%) never make it into production because they lack a shared vision outside of engineering. Engineering expertise will always play a role in AI, but it’s vital that businesses establish workflows that empower everyone—technical and non-technical alike—to play a role in order to move projects from test to deployed AI.