How does machine learning work? An interview with Appen CEO
We recently grabbed some time with our CEO, Mark Brayan and asked him to answer the question how does machine learning work? We also got his thoughts about the latest trends in artificial intelligence and the high-quality data on which both machine learning and artificial intelligence depend. We also discussed some of the ways we’re seeing businesses—particularly eCommerce companies—put these new technologies to work.
Q: Artificial intelligence is a phrase that once belonged to the realm of science fiction. Now it’s become an everyday term. Can you explain what it is, and how it’s becoming a bigger part of our lives?
Sure. Artificial intelligence, or AI, is a term that describes computer systems that mimic human thought and/or actions. These systems can be used to perform simple tasks, like answering questions from customers. As the technology has improved in the last decade, we’re seeing more potential for it to improve many aspects of business.
Q: We often say that AI systems “learn.” How does machine learning work?
AI systems “learn” using a technique called machine learning. That means that we feed the system lots of data or examples that “teach” it to predict the right response. It’s the same way we learn: through processing repetitive, consistent information.
For example, to build an AI system that “hears” or recognizes speech, we must feed it a lot of sounds, and associate the right words with those sounds (transcribed speech data) to help the system connect the sounds to the correct words. You have to record lots of people saying the same word with different voice tones, accents and background noise, and teach the system that all those slightly different sounds mean the same thing.
Another example is an AI system that “sees” or recognizes images. To teach that, we would feed it thousands of images with some form of descriptor that helps it detect those shapes or things in the images we’re asking it to understand.
“Machine learning” is the process of feeding information to these systems. The more data you feed the machine, the more it learns and the better your AI system will perform.
Q: What are the key components in successful machine learning?
Great question! There are three: compute power, algorithms and the right data (and lots of it). And the data is where Appen comes in.
For machine learning to work properly the training data that the machine learns from has to be collected, prepared and delivered into the technology. We’ve been doing it for 21 years, so we’ve been a major force of improvement in the industry, helping to evolve training data, machine learning and AI to where they are today.
In some cases, we work with our clients to use or “decorate” their data to improve their systems, as is the case with our search engine projects. In other cases, Appen captures and prepares the data for our clients. For example, language and speech data is especially important when you’re training AI to interact with people—and we’ve worked in over 130 countries, in over 180 different languages. We’re really an established expert in the field.
Q: The eCommerce industry is rapidly adopting AI, as you just mentioned, which is exciting in part because that means regular people now have the opportunity to encounter it in their everyday lives. Tell me about some of the places that’s happening today.
The eCommerce industry is using AI to improve the shopping experience in a number of ways. We’re helping our customers improve their search, so their customers get more relevant results, and can find what they want easier and faster. AI also powers the recommendations you see on retail websites, where the site recommends products based on your past behavior, or the behavior of other people who’ve bought the product you’re looking at.
Chat bots are also generating a lot of buzz. These are the software programs that can literally chat with people via instant messaging, answering simple questions. They may help you place an order, solve a problem, or return something. It can be hard to tell if you’re interacting with a person or a bot, in part because bots have gotten so good, and in part because bots might answer simple questions, and then transfer the interaction to a human if it gets too complicated.
Q: AI is sending text messages. Doesn’t it also power voice recognition?
Absolutely. Voice is becoming a more common way to interface with technology. We’ve gotten used to the idea with our phones: asking the built-in virtual assistant for directions, to search for things, to play music. Now more people are using voice recognition technology with at-home assistant devices (like Amazon Echo and Google Home) that can answer questions and help you with simple tasks, including basic online shopping.
Because they’re all based on machine learning, the more we use these services, the better they get. The more they learn, and the better able they are to recognize the voices of the people who use them. They store information about what questions you’ve asked in the past, what you’ve bought. As they improve, they’ll become more useful and a bigger part of our daily lives.
Q: AI can be very useful. It can also be intimidating for people who aren’t familiar with it. How can businesses implement it right, and make sure they get the most out of it?
At Appen, we’ve found that the key is actually human-generated and human-annotated (or decorated) data. To do a good job of interacting with people, the machine learning algorithms that power the search and speech-based solutions, AI needs human-generated data, and lots of it. You just can’t account for nuances in speech, accents and conversation styles without it.
Mark Brayan appeared on Worldwide Business with kathy ireleand® on May 14th to discuss similar topics. To see the full interview, click here.