Data Scientist Spotlight: Atalia Horenshtien
The Top Three Challenges in AI Preventing Your Organization from Getting to the Next Level (and How to Solve Them)
It’s no secret that many organizations are using AI to make crucial business decisions. But the secret for getting an actual benefit from AI is not as simple as developing some models or purchasing an AI platform. Using AI is like using your treadmill (I’m a runner, so I’m a bit biased here) or any other sports equipment. It’s not enough to buy it. You only get results if you use it, and then you quickly can become addicted to it.
As a Customer-Facing Data Scientist and an Evangelist at DataRobot, I would like to share with you a success story at Steward Health Care, the largest for-profit private hospital operator in the United States. Steward uses machine learning to make big decisions about staff and patients, reduce costs, and improve patient outcomes and experiences, and they have already started achieving their goal of decreasing costs. A 1% reduction in registered nurses’ hours paid per patient day netted $2 million in savings per year for eight of the 38 hospitals in Steward’s network.
My role allows me to learn about the AI market every day: what’s new, what’s hot, and what’s possible, all while staying informed on the latest AI trends. And by speaking with customers in different industries and conferences, I am collecting data on organizations’ current state and challenges (well, I’m still a data scientist, data is my second name :) ). I work with different industries—from those that are more mature in AI, like financial services and healthcare, to those who are earlier in their adoption, like retail, media, and sports. And you might be surprised to hear that they all share the same challenges that prevent them from getting to the next level.
So, let’s talk about those challenges more deeply and understand why they are important and what we can do to solve them.
The first challenge is getting models into production. In many cases, many models in the pipeline do not ever make it to production, and among those that do, controlling and managing them in different environments can be difficult. Also, degrading models over time can pose a significant risk to the business.
I recently gave a talk on this topic at the AI Summit Silicon Valley. At DataRobot, we call it the inefficient machine learning lifecycle: stuck in the lab, disconnected teams, technology mismatch, lack of stakeholder buy-in, and hidden technical debt. With DataRobot MLOps, you can manage, monitor, and govern your deployed models (regardless of where they were created or been deployed). You can also check which models are stale at a glance and automatically take action with challenger models and retraining policies with Continuous AI.
Why is this challenge so important? Because the speed with which you can deploy and iterate on models in production and unlock the ROI from that backlog of models that are ready to be deployed gives your business a distinct competitive advantage.
The second challenge, which is the rising challenge in my opinion, is ethical AI: how to make sure AI’s actions have a net good effect for society and how to make sure that those actions avoid entrenching historical disadvantages and prevent discriminating on sensitive features.
In order to make informed decisions that reinforce an organization’s ethical code, we must disclose sufficient information to an AI’s stakeholders. Lastly, governance is something no organization can ignore when data is involved, especially with the regulations in place today. Where there is a risk, organizations must apply high governance standards over AI’s design, training, deployment, and operation.
At DataRobot, we take this topic very seriously. We have a dedicated Trusted AI group working to make sure that the platform supports organizations with this challenge by protecting sensitive features for bias and fairness in development and production, bias mitigation, managing humility rules, access control, workflow process, and auto-generated documentation.
The third challenge centers around the AI workforce. The global demand for machine learning (ML) and AI solutions greatly exceeds the production capacity of all data scientists globally, and this gap is growing exponentially. Even when a business has the staff, there are usually so many other priorities that small teams might be overwhelmed. As a result, sometimes the team doesn’t have all the necessary skills due to the rapid evolution of data science and ML technologies.
How can you address this challenge? Outsource! Purchase a trusted AI platform, thereby increasing your capacity to solve pressing problems, and where possible, let the “machine” do the “dirty work” for you. This approach enables employees to be far more innovative and influential and allows them to focus on the more complex business problems that will drive positive business outcomes
I think we can all agree that the demand for informed decisions from data keeps increasing, and you don’t need a predictive model to know that more challenges will arise. It is simply a matter of asking yourself what you are doing about it and if you have the right tools in place to handle what comes next.
Are there other challenges you might want to share? I’m always happy to learn, discuss, and brainstorm together with you. Feel free to connect on LinkedIn.