AI and ML without data? What are you going to do?
Intelligence on AI and learnings from ML (part 1 of 2)
At this point, I’m comfortable saying the world of enterprise cloud has been consumed by artificial intelligence and machine learning. At least my world. Before they consume yours, I’d like to share with you some research and insights to keep the change in perspective.
Without data, what do machines learn?
Atos partnered with Google Cloud to commission a MarketPulse survey on AI and ML in the enterprise space (more on the research here). One of the first findings in the report: the top motivator for pursuing an AI/ML project is “using data to improve existing products or services.” It’s a wonderful lead because it shows that enterprises are going into AI and ML with the right intentions.
Data is the heart of AI and ML but, in my experience, it’s probably one of the areas that people forget most. We talk about AI and machine learning. We talk about Cloud ML and Dialogflow and TensorFlow. People get so enamored with the tech, they forget that the heart of every AI or ML analytics solution is data. Remember that. AI and ML without data? What are you going to do? What are the machines going to learn? That’s where the patterns are. The data drives the result set.
Keeping data in your thoughts will ground your AI and ML strategies firmly in reality. That’s really where you want to be because, in reality, the use cases for AI and ML are some of the most compelling I’ve seen.
The perspective I try to communicate is that there are really three things enterprises are doing with AI and ML:
- One is predictive analytics. It’s a form of ML but it’s very focused on a fixed data set, a fixed problem and a fixed outcome that I’m trying to get a solution to.
- The second is modeling to develop trending-type data to help me make better decisions around product development or markets, things like that.
- The last one is real time. This is different from the other two. Using data for immediate quality control in a manufacturing process, for example. More on this next.
3 use cases
I like three industries today that can go crazy with AI and ML: manufacturing, insurance and healthcare.
For manufacturing, it’s around quality control, costs and customer satisfaction. A manufacturer can use AI and ML to automatically update production based on a photo of a broken product in the field. They ingest the photo, compare it to a photo of an optimal device and trace the device back through every step of its manufacturing lifecycle. They find the point where the flaw is introduced and insert a quality checkpoint. The production process is immediately updated to ensure within a certain deviation that new devices look exactly like the optimal device.
Without delay, the manufacturer is less likely to introduce a device into the field that’s going to fail, say, in a car or a hospital. In addition to preventing potential safety issues, this solution reduces management issues around warranties and replacements and improves production efficiencies. That’s a real-time use for AI. It’s is a little different than predictive analytics or trying to build models of market trends. Not to downplay predictive analytics, as you’ll see in the next use case.
A great example in a hospital is when you think about everything that happens in an emergency room. A patient arrives in urgent need of diagnostics. If a medical device is going to fail due to use, wouldn’t it be great to know that ahead of time? Predictive analytics plays an important role here so devices don’t fail at 2 a.m. Saturday when the ER is at its busiest.
Now to the admissions process. Patient care records are easily available. With just that data, plus AI and ML, the hospital could automatically know a lot about me before I get there. A quick review of symptoms and it can auto-diagnose me and pre-order tests. When I get there, I’m already on the schedule and the devices are in good working order. They’re waiting for me, to help me reach the best possible outcome. It’s a far better patient experience than me waiting around for a bunch of random tests to figure out what might be wrong with me.
AI and ML lead to customized care plans. You could argue that medical care is by nature customized. But when you got your literature at the end of your last doctor appointment, how much of that information really applied to you? AI fixes that, too.
My final use case for this blog is a great one for insurance. It’ll be the sole focus of my next post.