Think about the defining technologies of the last decade. Social media tools like Facebook and Twitter revolutionized communication, cloud technologies disrupted storage and mobile technologies like Airbnb and Uber are changing how we live and commute.
It seems intuitive that these technological tools come with specific, well-defined purposes. However, there is one technology that is incredibly versatile, with ever-expanding boundaries to its uses. That technology is artificial intelligence (AI), a tool that combines statistics and computing to create functions that do everything from identifying trends and predicting the future to automating tasks and much more.
Before AI, all technologies had one thing in common: They would do exactly what you told them to. Applications were built to be deterministic. If a user were to enter some input, they would get an expected output.
Now, with AI, we are able to create programs that output the unexpected, that are probabilistic instead of deterministic. AI intends to uncover the unknown, learning from data much as a human would.
One of the pioneering examples of AI was IBM’s Deep Blue when in 1997 this computer defeated a grandmaster in a game of chess. (Full disclosure: IBM is one of my company’s partners.) Had the moves been expected, the computer could never have won.
Today, AI has expanded far beyond the realms of research and into the hands of billions of people. Even if you’re not aware of it, many of the services you’re using ingest your data, learn from it and use it to improve their features. For example, YouTube uses deep neural networks , a form of AI, to recommend videos to over 2 billion users . Google Search services an even larger user base with AI-driven search results .
But it’s not just about getting more clicks and recommending content online. AI can be applied anywhere there’s an input-output problem and indicative data to train the model on.
One of my ventures gathers AI problem statements from the industry to crowdsource solutions to a pool of data scientists. Some of these problem statements include saving animals’ lives, increasing agricultural yield and speeding up health care claims processing.
For example, we know that certain shelter animals are more likely to be adopted in certain locations than others, and those animals can be relocated to potentially save them from euthanization. However, those are only the trends we’re aware of, and AI can be used to uncover hidden trends about animal adoption patterns.
More specifically, AI models such as natural language processing can be used to analyze the text description of an animal in a shelter, and convolutional neural networks can be used to analyze images of the animals and help determine the probabilities of animals being killed or adopted in different shelters.
That technology, neural networks, is a versatile sub-field of AI that can be applied even to satellite imagery. For instance, agriculture is shifting to precision farming, enabling farmers to reduce their environmental impact. The very first step in precision farming is to identify fields from satellite imagery, which can be done by feeding various images through neural networks, which will output the probabilities of various contour boundaries existing.
It may sound confusing, but again, AI is a probabilistic technology, so it doesn’t tell you that something definitely is the case, it simply outputs statistical probabilities. For instance, AI can be applied to increasing operational efficiency in fields like health care, where initially denied claims are the source of billions of dollars of administrative costs. Here, AI can be used to expedite the process of filing claims and flagging them according to the probability of being rejected.
Clearly, AI is an amazingly useful tool, but that flexibility in its uses can also add to the confusion and, ultimately, fear surrounding AI. Due in part to the rapid growth of AI, there are generally a lot of misconceptions surrounding what AI can and cannot do.
The first stage of the AI journey, therefore, has nothing to do with AI. Before undertaking an AI engineering project, you need to formulate a problem statement that current algorithms are apt at solving, and then find indicative data for the problem. While we often hear about the explosion of “big data,” the reality is that having data that is relevant to the problem at hand is more important than just having a high volume of data.
However, by deeply understanding the problem, you’ll have a better chance of finding relevant data and building creative solutions with AI.