As AI technology is being implemented across industries, the roles of data scientist and AI architect are merging -- for good reason. As scarce as data scientists are, AI architects are even harder for enterprises to add to their team. Since a robust, effective AI strategy requires skill sets that overlap between data science and coding, it often makes sense for data scientists who are already in place in the enterprise to keep boosting their skill sets.
Fortunately, the starting point in enterprise AI is currently prediction-focused -- researching and anticipating the market, competitor use cases, customer demographics and individual consumer ideology. This narrow and entry-level data science strategy in turn narrows the AI skills that data scientists need to master, but to excel in maintaining enterprise AI architecture, there are a bevy of skills required.
1. Versatility in machine learning platform knowledge
Any competent data scientist is well versed in the nuts and bolts of descriptive analytics, and is equipped to dig into relationships among datasets. However, to be a competitively trained data scientist, strong competence in one or more platforms that automatically process data and create algorithms is preferred.
Machine learning platforms such as Amazon Web Services, IBM Watson, Salesforce Einstein, Apache Spark and Microsoft Azure Machine Learning provide adequate functionality out of the box for designing and deploying easily integrated, functional machine learning for mining customer data and gathering customer analytics. The problem is, using the program as-is just scratches the surface of what analytics can do.
Descriptive analytics -- perhaps segmenting consumer populations for hyperpersonalized marketing -- may be the latest platform offering, but gathering effective and highly-specific predictive analytics requires a data scientist who understands the platforms robustly enough to tinker with their settings.
There are some canned predictive analytics capabilities in each of the platforms above, but expanding machine learning to other enterprise domains almost always means customizing, and that's where data scientists need to broaden their AI skills.
When the platforms aren't giving enterprises gains, or need to be finely tuned to avoid bias or errors, the machine learning algorithm will require custom coding. The data scientist will need to improvise, and it's ultimately faster and more effective for the enterprise if their internal data scientist can handle this personally.
2. Understanding coding languages
There are four major coding languages that any data scientist can wield in order to build and maintain a sound AI architecture.
The leading AI language is Python, and it has many advantages the data scientist can appreciate. Python is easier to learn than C#, yet just as powerful (and more efficient) in the number-crunching game. Due to its popularity, there are abundant libraries available that are particularly useful to a data scientist, from the full ranges of regression analysis methods, to factor analysis, to support vector machines. Getting comfortable with Python allows data scientists to converse with open source platforms and set parameters.
TensorFlow is an open-source machine learning toolkit that includes strong modeling capabilities. Every data scientist builds models, and TensorFlow makes that process faster and simpler, while also providing APIs that make it easy for the data scientist's bench prototyping to interface with enterprise development.
While you don't need to know multivariate calculus to be a data scientist for most enterprise applications, AI more often than not requires the construction of functions with multiple inputs, where the influence of each must be separately established. By having multivariate calculus and Python in your kit, data scientists can both build the function and understand the math behind the outputs.
These four language AI skills alone will put a data scientist squarely into the realm of AI, but machine learning is the leading edge, not the bleeding edge -- and the bleeding edge is already starting to emerge.
3. Understanding neural networks
Machine learning is built on the idea that the designer already knows what's important in the data, and what needs to be analyzed. The relationships that will give insight are either already well-known, or have been discovered in the design process.
But what about situations where that's not the case? Computer vision applications, speech recognition, natural language processing and social network filtering are all cases where it's impossible to explicitly program a machine to mirror the data, because the human brain process is relatively unknown.
Fortunately, we have methods of getting around that, and data scientists with the AI skills to dive into the unknown of deep learning will have a leg up.
For the past couple decades, the neural network has been the go-to means for tackling problems that have no algorithmic solution. A neural network accepts multiple inputs into a system of interconnected nodes, each of which "fires" or doesn't, like a human neuron, based on those inputs. The weighting of each node changes as the network performs a task repeatedly, as outcome feedback scores the network on its performance.
Neural networks are being applied in applications with great success, the downside being that, because their solutions are non-algorithmic, there's no real way to know what's going on inside them: a neural network is truly a black box. Proficiency in neural network methodology is essential to any data scientist who wishes to break into deep learning.
A neural network isn't itself a coded solution, but must be coded as tools. Many out-of-the-box neural network libraries take care of this, so that from-scratch neural networks will probably never be needed; but these networks have a fundamental moving part -- the matrix -- and all manipulation of nodes and weights happens within such a matrix. A good data scientist will need at least a strong understanding of linear algebra and the matrix theory.
It's an exciting time to be a data scientist, and it's poised to only get more exciting from here on out. If you want to be part of the changes that are fast approaching, it's time to hit the books and perfect these widely demanded AI skills.