Just what is machine learning? AI experts talk definitions and uses in UpTech series

Editor’s note: WRAL TechWire is teaming with YourLocalStudio.com, a 10-year-old self-described video agency, to launch today what we believe is a riveting, deep, deep dive into the world of Artificial Intelligence. Alexander Ferguson, the CEO and founder of YourLocalStudio, and his team have done a series of in-depth video interviews with thought leaders in the AI field – many of whom are based in the Triangle. Here is the latest in the weekly series which focuses on machine learning – what it is and its potential to help businesses.

UpTech

Welcome to UpTech Report. In our last video, we looked at one simple question. What is artificial intelligence or AI?

One item we found was it can be seen as an all encompassing term. That there really many subsets or facets of it like natural language processing, computer visioning, machine learning and deep learning.

In this second topical episode, we take a closer look at what is machine learning? How can it be applied in business and what does its future look like?

To help answer these questions, we’ve interviewed a whole panel of AI experts, business leaders and entrepreneurs.

Interviewed were:

Robbie Allen, CEO InfiniaML
Richard Boyd, CEO Tanjo
Mason Levy, CEO Swivl
Brandon Andersen, CEO at Ceralytics
Alicia Klinefelter, Research Scientist at NVIDIA
Chris Hazard, CTO Diveplane
Rett Crocker, CEO Udu
Jeff LeRose, CEO Research Triangle Software
Bjorn Nordwall, CTO Research Triangle Software

To start us off, we did find that machine learning was actually a term first coined in 1959 by scientist Arthur Samuel, who worked at IBM. But a lot has happened since then. Let’s go over to our experts now and ask them.

What is machine learning?

Allen: So machine learning at sort of a fundamental definition is automating something and learning patterns with data. So essentially, you start off with the data set and it’s essentially learning. The software’s learning patterns in that data so that when you give it new data in the future, it can make a prediction or it can tell you how it’s similar to what it’s seen in the past.

Hazard: I would describe machine learning as showing a computer something and having it be able to mimic you and generalize it to new situations.

Nordwall: Being able to read a lot of data and create a map of that data and understand the relationship between that data that you have is what the machine learning is all about.

Andersen: And I’m totally ripping this off from MIT. By the way, I’m gonna make a plug right now. MIT has an open course for machine learning but I love that they point this out. They say that in traditional programming, in traditional ways of solving problems, you build a program first and then you bring in data and when you get the program running with the data, then you have an output. Machine learning is you have data and instead of a program, you have outputs. So I know that I want this thing coming out. So I have data and outputs. And what machine learning gives you is the program. It says based off of this information, here’s how you get from that data to this output. It’s this program. That’s really what machine learning is and it’s all based in statistics.

Allen: Machine learning really acts much like humans do in terms of humans learn off of their experiences. They learn off of data. Humans are probabilistic, they do not get everything right, they do not necessarily repeat the same result given an input and machine learning’s the same way.

Technology is always evolving. So we asked our experts, what are some of the challenges with machine learning?

Allen: By far, the biggest challenge with applying machine learning is just getting the data ready. And if the data has garbage in it or if it’s very noisy or there’s problems, especially ones that you don’t anticipate or you can’t see yourself, then the algorithm’s gonna learn that noise or it’s gonna start making misjudgments because it’s learned patterns that were incorrect.

Levy: Just having your data structured properly is a huge challenge for a lot of businesses just to overcome and understand. And having it all together. There are entire businesses that are customer data platforms that their whole job is to take five or six different datas to stores and try to create a unified view of their in customer. I think that would be the advice that I’d give the business owners is starting thinking about how you’re capturing data, how you’re tagging that data within the data system and ultimately what you want to achieve from it.

Boyd: How do we get enough data? And that’s the real challenge today. How can I find enough data that’s relevant that can train a system reliably on whatever it is I’m trying to teach it? The organizational knowledge of a law firm or how to drive a car. How to navigate to Mars. Those sorts of things. And that’s the issue today is how do we get enough data, get it into the right shape so that machines can drive meaning from it.

It’s clear that data is one of the main issues, but what about accuracy and the difference between consumer and business use-cases.

Allen: Machine learning’s probabilistic, meaning that you can’t count on a specific answer every time. The accuracy level that you get with the machine learning algorithm may be in the 80% or 90% range.

Crocker: In business you only have to get to 80 but in consumer you gotta get closer to 99, which is why people diss Siri because it’s like it didn’t understand me that one time, where as with business, it’s a little bit more pragmatic. You can be like it didn’t work 100% of the time but eight out of 10 times, I didn’t have to do anything.

So what are some of the things that are up and coming on the machine learning front? We ask the experts.

Klinefelter: I think there’s usually two categories I think about with this. There’s kind of the things we interact with in every day life that sometimes a lot of people don’t realize have machine learning in the back end that are really interesting. And I think it’s kind of obscured ’cause we don’t really talk about it. And then there’s really the kind of new and up and coming applications that people are working on right now. I think things like image classification in our every day life that exists on Facebook or word recognition and classification that exists on a Google search or even the ability for your phone to be able to learn on the fly characteristics of your voice, that it can identify something you’re saying better compared to somebody else.

LeRose: It’s very, very impressive. Especially in the medical field, how it can help with diagnostics. It can access the full internet for every single symptom that ever occurred and match it up against the symptoms of the particular patient.

Levy: A real opportunity in this abstraction layer is is what does the design of building these algorithms look like? And can we make it a Squarespace for AI? Can we make it so that it’s so easy that you’re dragging and dropping and now you got a really cool functioning prototype that’s only gonna learn and get smarter over time.

Klinefelter: But I think a lot of the interesting applications that are coming up are in this class of unsupervised learning algorithms, that’s kind of this cutting edge class of algorithms where you actually are telling a machine to learn on its own. You’re not really giving it examples of data to learn from, you’re giving it this kind of goal to meet in terms of accuracy and then you say learn on your own.

Last, we wanted to know how a business leader could actually apply machine learning in their own business? Should you do it on your own or hire someone? Is it very costly and difficult to apply?

Allen: It’s very difficult to go from basic regression to deep learning without having more formal training or a lot more experience because there’s more things that could go wrong. And if you don’t have an awareness or an ability to understand where things went wrong or why they went wrong, then you won’t be able to understand or interpret the results that you get. So I would say it’s not too much of a stretch to go to basic machine learning techniques but once you want to get to something like deep learning, that requires more of an understanding.

Levy: A question that you kinda have to ask yourself, every business, when they’re looking at software or services to add on to their core competencies, it’s a do we build it? Do we partner with somebody or do we buy it? And I think that that’s kinda where companies should be looking is how do we take these frameworks? Are there services out there that allow us to enable these things?

Boyd: We’re always scanning the internet. Going to conferences like SIGGRAPH and others and seeing what people are doing and seeing what we can borrow from. I don’t want to try to create some groundbreaking, new research in my company because, frankly, it’s not a good use a money when there’s so many other smart people spending lots and lots of money like people at Google and IBM and elsewhere. Once they solve the problem, we go awesome. Now we’ll come up with a way to integrate some of those ideas and do that six month 10X return on investment.

What questions to do you have about AI? Let us know in the comments [at the YouTube video]. We’ll find answers to your questions and cover it in future videos. Don’t forget to subscribe to get notified when new videos go live. I’m Alexander and this is UpTech Report.