Top six data and AI skills in demand for 2023

Data science is a field that continues to grow at pace in 2023.

As organisations across industries increasingly seek to harness and operationalise the data they hold, those with data science skills will continue to be in high demand. With this in mind, it’s essential that data scientists continue to develop their skills to ensure their purchase in an increasingly competitive job market.

With so much opportunity out there, it is crucial that data scientists are able to stand out with the skills that will benefit organizations most in the year ahead. Here are six skills to keep in mind.

  1. Try a data-centric approach

For organizations seeking to increase efficiency with data, there is an emerging trend of moving from a model-centric to a data-centric approach. A model-centric approach focuses on improving model performance by altering the code and the model’s architecture. Put simply, this approach focuses on model improvement through experimentation. The problem with this is that poor-quality data is often cited as the source of operational mishaps and inaccurate analytics that can impact major decisions and business strategies, leading to substantial consequences for companies.

A data-centric approach seeks to improve the quality of training data. This often involves more laborious activities, such as data gathering, wrangling and labelling, but an eye for this detail is becoming increasingly important. In 2023, it’s not just experimentation and the art of the possible that will deliver value. As businesses continue to contend with a challenging economic climate and look to do more with less, data is precisely where immediate value can be extracted. Data scientists should therefore develop their ability to create data pipelines that ingest the raw data available and create gold-quality datasets that can be used for model development.

  1. Stay bookish

As new ML research is often published as papers, an important and often underrated skill is the ability to read a report and reproduce it in code. Reading papers helps data scientists understand developments in the field and to be more effective, as it allows them to implement the latest techniques, validate them, and use them to create value for their organizations with cutting-edge business-driven applications.

It has been estimated that more than 90% of papers on AI applications are model-centric. This is an example of where a focus on research around data-centric models can really pay off for data scientists in 2023 and beyond, as it shows an understanding and appreciation of the fundamental truth that high quality data is the bedrock of good data science.

  1. Assess model fairness 

As knowledge and scrutiny around the outcomes of decision-making algorithms increases, organizations are becoming more mindful of bias creeping into their models. If a model does not use sensitive attributes, it can still be biased. There are many variables that can plague models and lead to unintended bias, which then leads to unfair decisions. Algorithms that are unfair can then be propagated as training data, which our models again learn and evolve from.

Data scientists who can demonstrate an aptitude for assessing and mitigating bias in ML models to ensure they do not deliver unfair outcomes will be particularly sought after in highly-regulated industries, such as financial services and healthcare, in which models are used to make potentially life-changing decisions.

  1. Explain AI

With awareness around privacy and the influence of algorithms growing and the continued evolution of privacy-specific regulations and legal frameworks, there is an increasing requirement for organizations to “show their workings” when it comes to their models.

A vital aspect of this can be developing auditable documentation for ML algorithms to explain the rationale behind algorithmic decisions and serve as a defense or explanation when faced with charges of bias, discrimination and errors. Building on Google’s work with Model Cards, data scientists should now put their shoulders to the wheel when it comes to explaining each step of model development to eliminate “black box” solutions. Model Cards or other forms of robust documentation should also be easily understood by non-technical colleagues, who may need to explain decisions or data used from a specific legal or data privacy perspective.

  1. Be an explorer

The automation of laborious and costly manual processes has always been the core promise of AI and data science is the key to unlocking this promise. Data scientists are, in many instances, an untapped resource in this regard, as data science is often conducted in isolation from other departments or lines of business within organisations. This is where individuals with the ability to identify where data science can deliver quick wins across will become invaluable.

In accordance with key agile principles, a “fail fast” mindset should always be applied to developing use cases for data models—and any new tool, platform or technology. With so many coming online almost daily, it’s important to be able to distinguish between what might be a little overhyped—I’m looking at you, ChatGPT—and that which will deliver truly transformative solutions. Those who come to their roles with this mindset are an asset to any organisation.

  1. Develop soft skills

Data science is a team sport. Full stack data scientists who can deliver an entire data project or build a product end-to-end are a rare breed. Typically, a data science team is comprised of a number of disciplines, from machine learning engineers and researchers, to data platform architects and data engineers to software engineers. Working in a team—and also within organisations in which data science may at times seem like arcane magic—requires a number of soft skills. Chief among them is the ability to collaborate and communicate effectively with peers and colleagues, which means being able to challenge ideas and to take and impart constructive criticism. It’s also important that the value of data projects is communicated effectively to non-technical leaders and, when appropriate, customers, too.

About the Author

Adam Lieberman is Head of Artificial Intelligence & Machine Learning at Finastra. At Finastra our purpose is to unlock the power of finance for everyone & redefine finance for good. We’re the orchestrator of open finance; building and delivering innovative, next-generation technology on our open Fusion software architecture and cloud ecosystem. We’re one of the world’s largest FinTechs, working with over 9,000 customers including 90 of the top 100 banks globally.

Featured image: ©DW