8 AI Predictions for 2020: Business Leaders & Researchers Weigh In

The first industrial revolution was powered by coal, the second by oil and gas, and the third by nuclear power. The fourth — AI — is fueled by an abundance of data and breakthroughs in compute power. While this abundance has allowed us to make significant progress in recent years, there is still much to be done for AI to be the positive life-changing force that many hope it will be. We asked thought leaders at the forefront of AI and machine learning technology to contribute some insight into what they think will transpire in 2020. Their predictions center around hardware, the human impact of AI, the public’s understanding of AI, and its limitations.

In 2020, people will accrue a larger understanding of the difference between pop-culture portrayals of AI and where we are in the industry. This will result in a more nuanced understanding of the role AI can play in government, enterprise, and consumer applications. There will be a broader awareness of AI’s current pitfalls, thereby increasing consumer willingness to slow the development process to ensure that AI is being built in a way that is ethical, thoughtful, and good for people. We believe we can push the industry forward while maintaining what’s best for humans as our central mission. It’s not just “do no evil” and instead it’s “let’s do good.”

-Ben Lamm, CEO, Hypergiant Industries

The past ten years have been all about collecting data and big data. The next ten will be about collecting less data and getting smarter about using it. We’ve seen what big data gets us: large distributed systems that have millions of points of failure, dirty training data, and, now with CCPA coming up, legal headaches about collection. Companies will need to get smarter and getting by with collecting just enough to be useful, not too much to be overhead.

–Vicki Boykis, Senior Manager, Data Science and Engineering, CapTech Consulting

I predict 2020 will be a year dominated by the human impact of AI/ML. People will wrestle with social and ethical issues surrounding AI/ML such as privacy, bias, and privilege. We’ll still see improvements in technology with new libraries and hardware capabilities including increased use of GPUs and improved workflows. Yet, people will be at the center of predictions and decision making as well as they will be the social conscience of ML/AI’s impact on others.

-Carol Willing, Steering Council, Python & Project Jupyter

My prediction for AI/ML in 2020 is that people will realize the limitations of the ever-growing supply of automated machine learning tools. A large part of machine learning and AI applications is in formulating the problem, collecting data and defining metrics. Finding a good model and good hyper-parameters is the easiest part of the process. It’s worth automating, but removing this step from your workflow will not drastically alter or simplify how people do AI & ML, and it will not reduce the need for domain experts.

–Andreas Mueller, Data Scientist Institute, Columbia University

Based off of the negative news cycles that have been happening with the large social media companies in the past couple of years, I believe there will be more of a focus on explainability of AI/ML in 2020. Specifically, more explainability around how algorithms on social media are surfacing content. People understand that AI/ML are what curate their newsfeed, but they don’t yet understand how. Hopefully this will provide transparency for users as to why they see certain content and their friends and families see other content.

-Annie Klomhaus, COO, Ribbon

I predict that there will be new algorithmic developments in AI and not just superficial application of deep learning. Reducing dependence on labeled data will be a key focus, and methods like self-training will mature. Simulations and domain knowledge will play a bigger role, and simulations will become more realistic, including on NVIDIA platforms such as DriveSIM and Isaac. These developments will impact challenging domains such as autonomous systems and “AI for science”, e.g in physics, chemistry, material sciences and biology. In addition, there will be a focus on robustness, interpretability and fairness as AI systems get deployed in an increasing number of applications.

–Animashree Anandkumar, Bren Professor of Computing, California Institute of Technology; Director of Machine Learning Research, NVIDIA

The problem of “common sense” will loom large in the 2020 AI landscape. I predict that all of the big tech companies (Google, Facebook, Microsoft, Apple, etc.) will make large investments in research on giving common sense to AI systems — that is, giving these systems the broad (and mostly invisible) knowledge that humans use to act robustly and reliably in the world. The lack of such humanlike common sense will come to be seen as the most serious impediment to deploying AI systems—such as self-driving cars — in complex real-world situations.

–Melanie Mitchell, Professor of Computer Science, Portland State University & Santa Fe Institute

I predict that in 2020, hardware choices for AI workloads will increase dramatically. After several years of mostly uncontested GPU-dominance in deep learning from one vendor, others will be working hard to take some of that market share. New GPU architectures will come on the market. Even more interesting, the market for specialized AI accelerators will really heat up, with more offerings from cloud vendors and chip makers. Data scientists may find the number of choices overwhelming. I think we’ll see an increased need for benchmarking of standard workloads to understand the pros and cons of each option. Most of all, each of these hardware technologies will need to pay attention to their Python user story in order to drive data scientist adoption.

-Stanley Seibert, Sr. Director of Community Innovation, Anaconda