More is written about artificial intelligence today than ever before. We read about milestones the technology has passed, like DeepMind’s AlphaGo beating the world’s best Go player or Facebook’s facial recognition performing better than humans at tagging photos. And to be fair, coverage is starting to focus on the technological advancements without sensationalist intrusions, but many are still wondering how AI is actually applicable to their work.
For most of us, figuring out how AI fits into our workplace begins with defining a problem that needs to be solved. (You don’t want to start with machine learning and make up a business problem to suit.) Only after you have clearly outlined the problem can you start to look at what solutions AI might offer.
Cognitively simple tasks
Take a look at the picture and tweet below. Can you tell what’s in the picture? Can you determine the sentiment behind the tweet? These are tasks that humans can do pretty easily, but we can use AI to do them at scale and with great success. This enables us to offload basic cognitive tasks and focus on the harder, more creative parts of the problem.
Cognitively complex tasks
AI can also help with complex cognitive tasks. Suppose we want to predict the selling price of a house, given some information about it (number of rooms, location, and more). We can’t solve the problem by looking at descriptive tables or graphs because the task requires complex cognition. An experienced realtor might be able to process this information and estimate the price, but this approach has two limitations. First, you need relevant experience in order to use this information accurately. Second, even if you did have the experience, you can’t scale the process.
With the right data, machine learning models can understand patterns in the data and give you good predictions. You can easily scale this to predict the price of millions of houses in seconds.
With the right data and training, AI can also be used to handle cognitively simple tasks (like seeing things in an image) in more complex scenarios — for example, using computer vision to find malignant tumor sites or language processing for translation.
Customized customer experience
When it comes to applying these capabilities in a business setting, the first example is always personalized customer experience. You can use a customer’s actions and interests to tailor their experience while they are using your product. For example, Netflix uses movies you previously watched, along with other data, to recommend movies. It also uses this information to decide what new content to create.
We absolutely need machine learning for this kind of service, because it’s very difficult to customize recommendations for each of your different users, with their varied interests, backgrounds, and so on, and doing this at scale is substantially more difficult. But don’t just offer recommendations because you can. Ask yourself what customization would actually be of value to your users, while respecting their privacy.
Optimize internal processes
Another avenue for applying AI is to optimize internal processes. This can assume many different forms, but look at what your teams are doing day-to-day. What are some of the tedious, repetitive tasks using data from Excel or images, text, or audio? These tasks typically don’t involve creativity or complex problem-solving, but they are time-consuming for your team.
Going back to our housing price example, let’s say your team of realtors is inundated with requests from customers who want to know the selling price of their house. You can use the data you are collecting to build a machine learning model to solve this problem.
This approach offers several advantages. First, you free up valuable time for your team to work on other things, like actually selling properties. Second, with the right data, AI can make great predictions for a variety of different inputs. Let’s say that for each house we have hundreds of pieces of information; this would be a very difficult task for even an experienced realtor to process, but it is relatively easy for AI to incorporate such data. Third, this solution allows you to respond to customers in seconds instead of making them wait for your human team to come up with a response.
Aid in decision-making
So far, we have seen examples where AI is responsible for the entire decision, but AI can also aid in the decision-making process.
For example, let’s say you are figuring out your company budget for the next year. You can use machine learning to forecast expenses, instead of just making an estimate off of last year’s values. Or let’s say you want to know whether to invest in a new market. You can use models to process tweets about that market and analyze public sentiment. This one signal alone won’t determine your decision, and it shouldn’t, but it can help get you there.
Applying AI in this way is particularly useful when the decision is very sensitive or complex. Going back to our tumor detection model, doctors shouldn’t trust the AI to report directly to their patients. But they can use the model’s results in their diagnosis, especially when it picks up things they’ve missed.
Virtually every business can benefit from these types of applications; however, there are few things to keep in mind. First, you should always start with the business problem and never find yourself saying things like “We need to put our data to good use.” Don’t predict something just because you can. Ask yourself if you can solve your problem using simpler methods (tables, graphs, simple code, etc.) and if not, then think about using AI.
Goku Mohandas is an AI researcher in Silicon Valley with a focus on using deep learning for natural language processing. He also works on democratizing practical AI for business and strategizing scalable machine learning solutions at ExposeAI.