5 More AI & Data Science Predictions for 2020

2019 was undeniably a big year for artificial intelligence (AI), machine learning, and data science. But what will the new year bring to these fields?

In a previous post, we made five predictions about what’s in store for AI and data science in 2020. In case you missed it, you can read it here.

But as we all know, AI and data science encompass quite a bit — there’s a lot going on in these domains. That’s why we’ve decided to make five more predictions!

1) The Hype Cycle Is Shifting Emphasis

A few years ago, you couldn’t read tech news without encountering the term “big data.” Since then, the emphasis has shifted to “data science.” And slowly but surely, “AI” has started taking over more and more headlines and conversations in recent years.

Words are powerful. How we define the above terms and discuss them have a monumental impact on how we understand them and ultimately affect how we implement them. The fact that our lexicon for these fields seems to shift every few years has a substantial effect, especially when you consider that data science and AI development are attracting a colossal number of newcomers.

To simplify this matter, those entrenched in these fields emphasize certain aspects while onboarding newcomers. For instance, it’s now common for data science practitioners at many tech companies to focus on statistics and machine learning while letting other math modeling disciplines like simulation and operations research fall to the wayside.

AI is no exception to this trend. Machine learning, neural networks, and deep learning now dominate most conversations, particularly in the context of natural language processing (NLP) and computer vision. Classical AI niches such as knowledge representation have taken a backseat as a result.

Jumping into AI and data science is overwhelming enough. As the deluge of definitions and terms continue to shift and get emphasized or de-prioritized, it’s important for newcomers to focus on what’s relevant to their objectives in these fields. Finding more content is not as important as finding the appropriate learning materials that will allow you to progress in your own personal roadmap. Basically, keep your eyes on the prize — not what’s getting hyped up at the moment.

2) Automation Efforts Will Increase but Still Be Limited

Unless you’re a well-funded San Francisco development startup that can pay AI experts a pretty penny, you’re probably all too familiar with the employment problem in this industry. Tried-and-tested talent is hard to come by in the fields of AI and data science, so it’s no surprise that automation is seen as an attractive alternative. But the limitations of technology really only allow for highly-specific tasks to be automated.

To highlight these limitations, let’s review the general steps of a data science project focused on model building:

1. Honing in on the goal of the project, putting a team together to make it happen, and securing funds if needed.
2. Defining the problem and the right approach to solve it. For example, is a supervised or unsupervised machine learning system the right way to go for a failure prediction model?
3. Choosing the right data to use.
4. Processing the data to ensure the model develops in the correct manner.
5. Generating hypotheses, or in other words, deciding what ideas are worth the time and effort.
6. Building and optimizing the model.
7. Validating that the model is effective and valuable.
8. Embedding the developed machine learning models into a production system or established business operations.
9. Building out future releases so your model can be seamlessly integrated with other systems.
10. Once the machine learning system is up and running, it’s time to interpret and act on its output.

Out of all those steps, automation is only being used to take care of one: Step 6. While automation undoubtedly has plenty of potential, much of it has not been realized due to technological limitations. Tools like AutoML or DataRobot have as much of a chance at replacing data scientists as website builders like Squarespace and Wix do at putting web developers out of business. Essentially, it’s not happening anytime soon.

3) AI Hardware Competition Will Heat Up

Nvidia currently dominates cloud AI due to an ample headstart in the deep learning hardware market. Google, Amazon, Qualcomm, and many others have tried to attain a significant market share, all to no avail. But if Nvidia wants to maintain its lead, it will have to shift gears.

Attention on AI hardware is shifting from just a chip to complete hardware platforms that are portable not bogged down by vendor lock-in. As a result, Nvidia’s competitors stand a chance at taking over market share if they focus on doing this right. And whether it’s Intel, Facebook, Alibaba, or Huwaei, many are attempting to do so.

4) Edge and Fog Architecture Will Become More Common

Out of a need for practicality and a desire to drive down deployment costs of large models, new architecture patterns like edge and fog computing will take center stage in 2020. The former will especially help in regards to computing and data transfer requirements for real-time video analytics. Advancements in computer vision and the rise of purpose-built hardware like AWS Deeplens are also supporting this shift.

5) Teaching AI and Data Science Will Remain Easier Than Making It Work!

It can be hard to face the facts. But the truth is that it’s still much easier to teach AI and data science and sell tools for those fields than to actually make them work in practice. Creating value with AI systems is a difficult endeavor. Not only is AI still an emerging technology that few understand, but new developments every day only make catching up more complicated.

There’s a lot of attention on doing things right in data science and AI but not enough emphasis on doing the right things. At its core, AI is an optimization machine. And not enough people are asking, “What are we optimizing for?” To make matters more complex, figuring this out is usually obfuscated by politics at many companies.

Therefore, it’s safe to say that 2020 will bring many more tools and learning materials to simplify AI implementation. But the actual implementation will remain elusive for many organizations.

The Era of AI Is Just Getting Started

We hope you’ve enjoyed and learned a thing or two from this 2020 forecast for AI and data science! Which of these predictions do you think will come true? Which do you think are wrong? Let us know your thoughts in the comments below!