Data Science: Influencers review 2018 and share their 2019 predictions
Data science was one of the hot topics of 2018, and it’s likely to dominate again in 2019.
The IBM Big Data and Analytics Hub asked five data science influencers — Jennifer Shin, adjunct professor at New York University Stern, product director at NBC Universal, and founder of 8 Path Solutions; Mike Tamir, head of data science at Uber ATG & UC Berkeley faculty; Bob E. Hayes, Ph.D., founder of Business over Broadway; Ronald van Loon, director of Adversitement; and Christopher Penn, co-founder of Trust Insights — to take a look back at 2018 and look ahead at what's to come in 2019.
IBM Big Data and Analytics Hub: What was the most surprising data science trend of 2018?
Jennifer Shin: The duality of trends within data science was unexpected, especially at the start of 2018. The year started with a sense of urgency around just about anything and everything having to do with data, including regulatory restrictions, demand for cloud-only solutions, and a need to scale edge computing and IoT (Internet of Things). As the demand for technology grew in 2018, these trends have been overshadowed by more moderate trends. For the most part, businesses have adjusted faster to the changing technology landscape and entering 2019 with a more human-centered focus on using technology to enhance human capabilities rather than replacing it. Businesses will move away from expecting completely automated solutions, such as chat bots, and find solutions that can exist as complements, such as workplace AI.
Ronald van Loon: Edge analytics gained traction this year, bringing real-time analytics to sensor data right at the source where data is generated, which reduces latency, minimizes issues associated with transferring massive amounts of data to central locations and increases security. In IIoT (Industrial IoT), we’re seeing edge analytics used in predictive maintenance for industrial organizations to identify and address maintenance needs in advance and improve a manufacturer’s response time to defects, quality control and customer needs.
Mike Tamir: For the past three years, with the growth of so many deep learning (DL) development ecosystems, we have seen a trend toward using deep learning more and more in production systems. While any solution should ultimately depend on the data, looking back, we may point to 2018 as the year when using DL transitioned from the exception to the rule for modern technology companies with sufficient data. For specific algorithms, I've been especially excited by the resurgence in graph-based algorithms like graph sage, as well as the improvements in contextualized word embeddings that Google and OpenAI have been working on.
BDAH: Which data science trends do you expect to dominate 2019?
Bob Hayes: In 2019, while AI and machine learning will continue to be the focus, I expect to see more of an emphasis on non-technological aspects of data science come to the fore. Specifically, I see topics related to ethics to dominate in 2019. These include issues around privacy, security and even fake news. Rather than focusing on showing that machine learning works, we need to understand how it works (how it made decisions). Also, US companies will focus efforts on how they use consumers’ personal data. California adopted the California Consumer Privacy Act (goes into effect in January 2020) and I expect (hope) that other states will follow their lead. I fear a growth in the problems around fake news. AI and machine learning are being used to create videos via deep fakes that are looking more and more realistic, showing people saying things they haven’t said or acting in ways that they didn’t act. As Max Tegmark says, being cognizant of how AI can be bad is not fear mongering, it’s simply “safety engineering”.
Ronald van Loon: In 2019, we’ll be seeing better infrastructures for deep learning with more integration of software and hardware stacks that will help promote mainstream deep learning applications. Growing customer demands and expectations requires the capabilities to provide varied offerings through AI and deep learning applications that can improve the overall customer experience. Also, end-to-end AI lifecycle management platforms will enable companies to use all data sources and support digital transformation as they progress in analytics maturity.
Jennifer Shin: In 2019, we will see more scrutiny of models and algorithms that have been popular over the past few years, as well as greater demand for transparency in how well a model works and whether the results were validated.
Mike Tamir: I expect the turning point we've seen with the trend in 2018 to continue into 2019. Deep learning should continue to dominate in 2019 in addition to the important advances coupling traditional approaches with reinforcement learning.
BDAH: Which data science technologies do you expect to gain traction in 2019?
Mike Tamir: I'd like to see deeper work in AutoML. In 2017, we saw a lot of attention in AutoML, but I don't think that we have begun to fully take advantage of the potential here.
Jennifer Shin: In 2019, I expect data science will continue to be more human-focused. Rather than creating a new role within the organization (such as chief data officer), more businesses will consider modifying or adapting existing data processes into existing operations.
Ronald van Loon: The optimization of combined hardware and software stacks that will accelerate AI innovation and provide a competitive edge for businesses so that they’re able to meet the growing demands of customer expectations. This will also ultimately make it easier for companies to prepare for and transition to AI. Software that’s simple to use and deploys quickly and unifies with hardware allows data preparation to be augmented for deep learning distribution.
BDAH: Which data science technologies do you expect to decline in importance in 2019?
Jennifer Shin: Standalone solutions that lack integration with other technologies, as well as niche solutions with a very limited application or scope, will be too outdated and expensive to appeal to data science teams.
Mike Tamir: I'd like to see efforts like ONNX start to make it easier to transition from one DL development ecosystem to another. This may not mean that technologies decline, but will help for more specialization, as we saw for example with Hadoop and Spark.
Ronald van Loon: Standalone data science “silo” infrastructures will be replaced by enterprise end-to-end solutions that give organizations the ability to handle the growing surge in data and new technologies and tools that require AI and deep learning capabilities. Older legacy systems and silo infrastructures simply can’t keep pace with the evolving nature and demands of new technologies that are designed for the AI era. These older systems have data architectures that predate the recent digital revolutions in AI, big data and new algorithms, and thus can’t support the agile IT systems, performance capabilities and scalability required for AI.
BDAH: What will the data science landscape look like five years from now?
Ronald van Loon: In the coming five years, there will be a “battle of the platforms”, as this becomes the key ingredient in both digital business models and digital strategies across all industries. With leading brands offering enterprise-level digital platforms, companies can take advantage of optimized solutions for AI, deep learning, big data, security, cloud computing, IoT, edge computing, 5G and machine learning. This creates a digital ecosystem and intelligent enterprise that plays a vital role in business success as companies invest in the technologies and capabilities that accelerate digital innovation and transformation.
Mike Tamir: On that time scale, we could see a lot of change. While it's hard to speculate what kind of algorithms or techniques will be in vogue by then, I am optimistic that current trends which allow us to abstract and improve the ease of development will mean that in 5 years one will be able to easily implement techniques that by today’s standards take quite a lot of time. (But this is a safe bet for any five-year period.)
Jennifer Shin: In the next five years, new startups and businesses will find it difficult to enter the market and compete against the larger corporation while industry leaders will find ways differentiate the products, services and the overall value that is being offered to customers.
Christopher Penn wrapped all of his predictions up and shared them in this video.
Interested in learning how to solve complex business issues with data science? See how IBM clients have done it.