Artificial intelligence and machine learning (AI/ML) are becoming increasingly central to any business strategy and this trend has only accelerated since COVID-19. The key to gaining significant advantage over the competition is talent (data practitioners with financial domain knowledge), along with targeted investment and access to the best technology tools.
- Explainable AI is a significant challenge that must be addressed to meet regulatory mandates.
- During 2020, many new roles were created in data science. Data scientists can identify new ways of doing business and create considerable value.
- Firms are using natural language processing (NLP) to unlock the value of unstructured data. Our 2020 Machine Learning survey showed that 17 percent of firms use only unstructured data, an increase from 2 percent in 2018.
For more data-driven insights in your Inbox, subscribe to the Refinitiv Perspectives weekly newsletter.
2021 follows unprecedented business change caused by the onset of the COVID-19 pandemic. Financial firms have had to rethink policies and processes, accelerate digital transformation, and adopt technologies that can drive growth in a rapidly changing environment.
A tougher economic climate has meant that companies are turning to AI/ML for the promise of automation and efficiency gains. This has allowed companies to scale up massively and re-deploy human capital.
The potential of AI and ML has risen exponentially. Some firms see the challenges of change as an opportunity to increase investment and build for the future. Refinitiv’s 2020 AI/ML report, The rise of the data scientist: Machine learning models for the future, shows that 40 percent of firms expect to increase investment in AI/ML as a result of COVID.
As we move forward in 2021, three key AI/ML trends are emerging — AI/ML explainability, the rise of the data scientist, and the criticality of natural language processing (NLP).
Trend #1: The imperative of AI — explainability
Explainable AI — tools and frameworks used to ensure that the results of machine learning models can be interpreted by humans — will be a considerable challenge in 2021. Increasing investment and greater use of deep learning — 75 percent of firms surveyed by Refinitiv are in production with the technology — raise the stakes, and regulators need to know how decisions are made. Machine models don’t tend to provide explanations for their predictions.
As models become more complex and consume larger and more diverse data sets, the challenges increase.
Leading AI practitioners are countering the challenges by building model governance teams, hiring linguistics specialists to interpret decisions, and deploying tools such as LIME (Local Interpretable Model-Agnostic Explanations) to explain AI/ML models.
There is a trade-off between explainability and performance. Models whose predictions are totally transparent tend to be impoverished in their predictive capability, or are inflexible and computationally cumbersome. However, regulation favors the former.
To improve the outcomes of AI/ML, Refinitiv Labs is working with MIT-IBM Watson Lab researchers to build solutions that can differentiate between causation and correlation.
Today, narrow AI that performs specific tasks at a super-human rate struggles to differentiate between actions or states that appear in proximity (correlation), and actions that actually affect each other (causation).
The problem is addressed by applying causal inference to AI solutions. Causal inference is the process of drawing a conclusion about a causal connection based on the conditions of the occurrence of an effect. This creates more robust models and predictions, as well as higher-level causal reasoning and explainability.
Trend #2: The rise of the data scientist
2020 was an undeniably difficult year, but a silver lining was that it helped provide a springboard for data practitioners.
New roles were created in data engineering and data modelling; there was a surge in demand for specialists in data procurement, legal data rights, and AI/ML; and ‘citizen data scientists’ emerged in the business, using Python, the language of AI/ML, to speed up decision-making.
Highly skilled data scientists can translate business requirements into data science, identify new ways of operating, and create value that has the potential to provide a huge productivity boost for the economy. They use data science to help answer critical and complex business questions.
Refinitiv Labs, by way of example, responded to uncertainty caused by the pandemic with COVID-19 Company News Tracker, which uses AI/ML to process news stories and detect, classify and store any risks or opportunities mentioned.
The resulting time-series signal allows firms to focus on material impacts as opposed to general news, saving valuable time and effort.
The need to hire data scientists with a strong understanding of statistics and domain knowledge is higher than ever. Refinitiv’s 2020 survey shows the average number of data scientist roles per company increased by 17 over two years. The number of firms with teams of five or more data scientists rose to 28 percent.
Refinitiv is empowering growing communities of data scientists and practitioners with an open platform and tools such as its data exploration tool, a sophisticated data discovery solution, and CodeBook, a Python development environment that enables rapid build and deployment of AI/ML models.
Trend #3: NLP set to transform finance
The perfect storm of exponential growth in data, increased compute power and advances in NLP technology and AI/ML model design are creating a meaningful opportunity for financial services firms.
While structured data has been largely democratized over the past few years and is declining in value, unstructured data is booming as firms turn to NLP to unlock its value, increase efficiency, scale up massively and deliver a wealth of new signals to the business and revenue to the company.
Refinitiv’s 2020 AI/ML report highlights increased use of unstructured data with 17 percent of firms using only this type of data, up from just 2 percent two years ago.
Refinitiv Labs has been exploring deploying NLP technology for some years, working to enable smarter customers with prototypes such as Sentimine, which surfaces equity performance themes in company transcripts and research reports.
The NLP story is a data story and, as a world-leading provider of data, Refinitiv is unsurprisingly approaching NLP in many guises. Examples include building NLP and AI/ ML into its content, like Machine Readable News, and leveraging NLP to improve internal operations.
Looking ahead, we see two exciting developments in the world of NLP.
The first is non-verbal communication. There is increasing investment (from a time and money perspective) into how machines can not only understand what we say, but also how we say it. This is particularly interesting given our move to virtual working environments and body language/facial expressions being more important than ever.
Separately, Refinitiv Labs is exploring the concept of providing customers with NLP models trained on our own data sets. This will allow our customers to train their models with ours and could dramatically improve performance.
Watch this space, because the word on the street is that words on Wall Street are what’s going to give us the edge in 2021.
