The Machines are learning… What are we teaching them?!

When machine learning gets talked about, many people picture a futuristic world with flying cars, but it was actually introduced in the 1980s and has been advancing rapidly ever since. Machine Learning is a subset of AI where statistics are used to enable machines to learn from experience. Where AI might involve pre-programmed instructions, Machine Learning allows a system to learn from its actions and make it’s own decisions.

Machine Learning: definition, types and practical applications - Iberdrola

Already lost? Here’s an example with everyone’s favorite thing… Ads

Before Machine Learning, advertisers would decide what to show people based on what they know. Kids like toys, so we should provide toy ads to anyone between the ages of 4-10. Now with Machine Learning, the system is smart enought to figure out what to do without those instructions. It will analyze who clicks on what ad, and use that feedback to refine what ads are shown. It does this loop continuously until it reaches the most clicks possible.

Create an AI feedback loop with Continuous Relevancy Training in Watson  Discovery – IBM Developer

This is why you wonder, is Alexa listening to me? No, it just knows that people like you have a high chance of buying weighted blankets!

Sounds scary… is it dangerous?

Danger: Machine Learning" Sticker by eigenmagic | Redbubble

Although long debated by technologists, the short answer is not really. Right now the biggest threat Machine learning poses to you is it takes your job by automating it. There are long term valid concerns though. When we allow complex algorithms to make decisions, without knowing why, it opens up lots of risk. A big issue that has become a topic recently is bias. Because Machine Learning does as we do, not as we say, sometimes we don’t realize what we’re teaching it until it’s too late. For example you may want to create a ML model to determine who should get a home loan. Sounds innocent enough right? It will look at historical data of who got loans in the past, and use that to make a decision about new customers. If there was any racial bias in that data set however, it could learn that pattern, internalize & amplify that bias. This is just one small example, but you can imagine the possible implications.

So what do we do?

AI is not just learning our biases; it is amplifying them. | by Laura  Douglas | Medium

As it was first introduced, big companies were throwing ML models anywhere they could to see what would stick. As the industry has matured, people have become much more thoughtful to how and why these systems are being built, and what potential outcomes and consequences it could have. It’s more important than ever that we understand these concepts as they will continue to shape the world around us.

5 comments

  1. Great article! I think the Group C Ted Talk video really hammered home your potential concerns with machine learning, and how we can ensure we reach expected outcomes with its application. While there will always be edge cases to consider, which will allow biases and other factors to skew the effectiveness of ML, the onus really needs to lie within the company itself to ensure that the employees implementing and maintaining ML are growing and learning from it as much as the ML grows and learns from the users. If it is to be the future heart of a technological revolution, it will definitely require a very symbiotic relationship between user and machine.

  2. One of the things that I really get caught on is the loss of job with machine learning. I feel like many times in the low income/minimum wage jobs that get impacted the most from machine learning. But I also see value in that it can analyze data much faster for businesses to make data based decisions.

  3. I like that you highlighted advertisement to explain machine learning. I just had a guest speaker last week in Consumer Behavior that was talking about the timing of ‘Next Best Action’ when it comes to marketing and I referenced it as “just in time marketing for the individual” in my notes. Specifically because as a consumers begins their consumer journey there is a small window of opportunity to influence their decision making based on what they know about the person and machine learning is doing that for marketing companies. He also was talking about how this is the future of marketing and companies that have switched over to this new line of thinking have seen huge increases on their KPIs.

  4. The last image you have here ties the entire blog together in one piece. Avoid Bias! ML can be great in most cases, but if it’s learning from a model that neglects bias, then what it’s really doing is perpetuating the problem. Humans need to be much more hands on in ML than ever, especially in a time where society is noticing more of the real world issues we struggle to identify. Humans aren’t always the problem anymore, machines are too.

  5. Great post that will likely be a discussion topic in the years ahead as ML becomes more prevalent in our lives. With the rapid innovation within ML and deep learning, do you think we, as humans, can keep up with the complexities to address bias? I think it’s an important step to be discussing the dangers and implications of bias in ML, and I hope that talk turns into action.

Leave a Reply

Please log in using one of these methods to post your comment:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: