The Great Debate: Dan Kirsche and Robert Mowery on the Pros and Cons of Machine Learning

We’re excited to host The Chicago Tech Debates event on Machine Learning on January 24th. To preview the event, we’ve put together a Q&A with Dan Kirsche, Head of Software Engineering at Enova International, and Robert Mowery, Google Cloud SA-AGBG — both experts will be participating at the Chicago Tech Debates event to discuss the pros and cons of machine learning.

• Let’s kick off this Q&A with a quick look at machine learning. For readers who aren’t familiar, what is machine learning in layman’s terms?

Kirsche: Machine learning is a type of artificial intelligence; it’s the concept that machines can learn from data sets and make decisions off of what they’ve learned. Here at Enova, we rely heavily on machine learning during our loan underwriting process where funding of loans is determined. We use data from many sources including bank statements and credit history and use that to determine whether or not to fund a loan. Historically, this decision was done by a growing ruleset built by our business team and manually coded by our software, but that is slow and over time the rules become too many to manage.

With machine learning, the system can build its own set of rules by looking at the vast historical data including loan performance. The model then continues to retrain itself over time as more data becomes available. These machine learning models are far more accurate than the manual ones created by the business.

Mowery: Basically, you’re taking scientifically based algorithms and modeling a process off of them in the real-world. Using data, machine learning has a number of different applications, whether it’s monitoring a crack in the road or identifying a particular item or person in a photo. It’s based on repetition and datasets; think of how you teach a child to recognize certain processes — that’s similar to what we’re teaching machines.

• How can founders, entrepreneurs, and startups best leverage machine learning to support their objectives?

Kirsche: The best use cases are the parts of your business where you’re making similar decisions repeatedly and have a high impact. Once you find those areas, you can apply machine learning so that you can optimize those processes. A great example is the very common task among businesses of generating customer call lists for their sales reps.  Machine learning could build a model to prioritize each sales rep’s call list by using past call logs.

Mowery: For startups, you want to figure out rote processes, along with unique features, and break them down into mathematical processes. For example, if I was working in the retail space, I could use machine learning to build customer profiles. I could look at customers and the purchases that they’re making to build a model that can identify customers who recently had a baby and send them targeted ads based on that. Instead of having a human sit there and figure that out, I can have a machine do the work.

• Even though we’re not always aware of it, does machine learning affect everything we do in our daily lives?

Mowery: I would say it certainly does — particularly because of how much AI and machine learning technologies have evolved. One area is the banking industry; fraud detection is a really good example. You certainly don’t deal with it everyday, but it’s happening behind the scenes. When certain types of transactions are performed, you develop a pattern of purchase. A machine can take a look at that pattern over time and provide banks with a good idea of what your spending habits look like. So let’s say that all of a sudden, your spending habits drastically change, the algorithms behind machine learning will see that there’s been a vast change and anomaly to your purchasing pattern which likely resulted from fraud.Another example is inventory; a lot of large retailers get their items need to get their items from the warehouse to the store shelves. Machine learning helps them keep track of their inventory and stock items when necessary or when a weather pattern changes that might affect shopping habits. Now as a customer, you probably don’t think about all of that, but when you make a purchase, there’s a machine learning center somewhere that’s tracking and leveraging data about what you buy and how you’re buying it.

Kirsche: I think you see it in our newsfeed everyday. Whether you use Apple News or Google News, the headlines that you see are customized to what you’ve been reading and how you’ve been browsing the news. Additionally, you see it in things that we all use daily, like ride-sharing. When you use an Uber or a Lyft, it’s all machine learning; all of it is a prediction from how long your ride is going to take to when the driver will arrive at your location. So machine learning really is present in our everyday lives even if we don’t always notice it.

• Lastly, you’ll be taking part in the upcoming ‘Chicago Tech Debates’ event on January 24th here at 1871. What are some of the key debates and issues when it comes to machine learning?

Mowery: The credibility of the data that’s being processed by the algorithms is always something that comes into question — remember, machine learning models are built by humans and humans are still fallible. One of the big areas and issues that we’re asking ourselves is ‘what’s the quality of the data?’ You want to make sure your inputting the right information.

The other debate is that algorithms are getting so smart that our privacy is getting impacted. Today, technology is moving faster than laws. You’ve got GDPR and HIPAA, but you’ve still got people giving away data for free. I don’t think machine learning is the culprit — I think it’s more so how it’s being used.

Lastly, I think one of the largest debates is whether or not machine learning and AI is going to take away human jobs. I personally don’t think that’s going to happen, but I do think that we’re going to see an augmentation of humans and machine learning. I think machine learning really shines when it comes to resolving mundane and repetitive tasks. For me, personally, I have a small farm and I’m using machine learning to self-identify plant diseases and help with my food supply. It’s very difficult to find enough people to go through plant-by-plant and identify the sick ones — so I think the machine learning aspect is more of a positive thing.

Kirsche: There’s two topics I see very common in discussions; technology & data privacy. With technology, this field operates so rapidly and the changes are so great new methods are constantly being developed. What are the best tools today?

Then, there’s the data privacy component; data biases, misleading data, bad data — historical data doesn’t mean that it will translate to future data. Collecting and storing that data is also very sensitive, where is the line when you’re collecting data to customize and personalize options for a customer?

To RSVP for the Chicago Tech Debates event on Machine Learning, follow this link.