Machine learning doesn't have to be for the big guys. Even small companies can take advantage of AI.
Mention artificial intelligence and people immediately think of Hollywood. Movies like Terminator or this year’s Ex Machina have created an association between computer learning and intelligence, and the end of humanity. At a recent event where Google’s Peter Norvig, director of research, spoke he accused the movie industry of typecasting AI as evil.
“I think that’s just typecasting. Look, when you have two actors try out for a role and one’s a human and one’s a robot, the evil one is always the robot,” Norvig said.
Yet, most people are already using AI in everyday situations and will soon be using it more often. Examples include applications such as Crystal, which attempts to teach users how to craft emails specific to the recipient’s likes and dislikes, and Microsoft’s recent How Old Do I Look service that looked at photos of people and extrapolated their age. There are countless more opportunities to use AI and machine learning tools in our lives. And businesses especially should pay attention.
So I’m offering a short list of applications, resources or services where businesses can add a little AI magic to their existing operations. Doubters who worry about the machine takeover may be surprised how much AI is already at their fingertips—and how helpful it has become.
Build better models with human help: If you’ve ever listed a home on Airbnb, you may be familiar with a slider that helps you set a price for your rental based on the features of the home, the time of year and other factors. Airbnb built that tool using a machine learning package called Aerosolve, which it has put on Github for others to use. The Aerosolve package isn’t a bunch of algorithms, it’s a tool that lets humans input data to influence a package of algorithms so they can “teach” a learning model. Airbnb offers the examples of being able to teach a computer to paint in the pointillism style or produce income prediction charts based on demographic data. This isn’t machine learning for the common man, but it brings it a step closer out of the realm of the true data scientists.
Amazon makes AI easy: In April Amazon launched a cloud service that lets companies build predictive models using their existing data. The service lets people who work with spreadsheets or huge data sets take advantage of statistical modeling to make predictions. Amazon’s service essentially uses your data to train its algorithms to help deliver the result a company is looking for. Currently it is limited to the following use cases:
- Predicting one of two possible outcomes such as, “Is the shipping address an apartment complex?”
- Predict one of three or more possible outcomes and the likelihood of each one. Amazon gives the example of, “Is this product a book, a movie, or clothing?”
- Predicting a number through regression such as, “How much red lipstick should we stock? What should we charge for the lipstick?”
Finding an Ebola vaccine: Some companies aren’t interested in the tools. They want to provide the entire service using their own algorithms. These are the more popular examples of AI that we’ve seen. Atomwise, a company that recently raised $6 million, is using its algorithms to speed up drug research. It takes the side effects of known drugs and looks for those that might help solve a pressing medical issue. By finding new uses for existing drugs and compounds it can speed up drug discovery. It’s first success in this field was coming up with two compounds that could help reduce the transmission of Ebola.
Real-time translation without humans: Google has spent a lot of time developing and improving its translation services using machine learning. The fruits of that labor are available as the Google Translate API, which lets you build dynamic translation services for a fee. Instead of building a database of words that mean other words in other languages, Google Translate uses machine learning to understand what a word means and can even parse idioms. It has done this by learning how words relate to one another.
Facial recognition everywhere: Facebook, Nest (through its Dropcm acquisition) and Microsoft all have built tools that recognize people’s faces. While machine learning can’t identify the person in the photos, it can say that a person in one photo is the same as a person in another photo or video. When you link that with a limited field of people’s names, you get systems that can identify who people are in a home setting or in photographs on a social network. These facial recognition efforts are a subset of computer vision research that is used in self-driving cars or the “How Old am I” program.
As its core, machine learning helps make sense of the troves of digital information we’re generating. It’s not about taking over the world, but about letting computers do more of the work that computers are good at, thereby letting humans focus on the tasks they are good at. When viewed that way, AI might help people destroy the world faster, but it won’t let computers take on a life of their own and terrorize humanity.