You Can’t Have Effective Artificial Intelligence Without Human Intelligence: The Importance of Training Data
The following is adapted from Real World AI by Alyssa Simpson Rochwerger and Wilson Pang.
AI is often poorly understood as a black box. Something goes in one end, and magic comes out the other. So when an AI model goes wrong and produces inaccurate or biased results, people blame it on the AI.
But this “black box” thinking is only partially accurate — the box isn’t actually opaque.
With every AI model, inputs are fed in, and the model is taught what the expected answers are. After being fed enough data, the model will eventually be able accept inputs it has never seen before and produce, with some degree of confidence, corresponding outputs. Its output is wholly dependent on the quality of the data input, which was curated by humans.
Essentially, the intelligence of an AI model is almost wholly determined by the human intelligence used to curate the data that trained it. Since good AI models can be a powerful tool for business growth, while bad models can be a waste of time and money, it’s critical to choose training data carefully.
How AI Models Are Trained
Imagine you have a new puppy. You want the dog to sit down every time you say, “Sit.” You repeat the command and give your dog a treat whenever it connects the dots and gets it right. After a while, the dog will infer that it is time to sit whenever it hears the word “sit.”
What happens with AI is very similar. You teach AI how to get it right. The teaching process is called model training, and the only difference is the teaching target: a model instead of a puppy.
Data scientists load a lot of data into a machine, and the machine tries to choose a model to “fit” the data. The model or algorithm can range from a simple equation (like the equation of a line) to a very complex system of logic/math that gets the computer to the best predictions. The training process decides all the parameters (weights and bias) in the equation.
Choosing the right model and finding its parameters can be a challenging job. Data scientists used to work on a lot of the details of the math formula, produce low-level matrix calculations, write a lot of code, and spend a huge amount of time debugging the code.
Frameworks like TensorFlow® or PyTorch® have simplified the work dramatically. These frameworks offer ready-made building blocks that significantly improve the speed at which even newcomers can implement machine learning architectures and train a decent model quickly.
Though training models is now easier and faster than ever, it’s critical to approach this step of AI with care.
The Importance of Training
Let’s take a look at one popular machine learning model: the decision tree. A person is trying to make a decision: should I go to my colleague’s party? A few pieces of information will go into that decision: Am I free? Is it going to be fun? Is it near public transit?
The person will try to answer each of these questions. Some will be yes, some no. Some are more important than others — if it’s going to be fun, maybe it matters a lot less if it’s near public transit. There’s an intrinsic weight to each of these inputs, and the person will answer the question by doing some mental math to combine them all; if the result passes some threshold, they’ll go to the party.
Humans weigh these factors and arrive at a decision without even thinking about it, but computers have to be told explicitly how to make a decision. It’s very important that the data scientists and business people responsible for the creation of a model clarify and articulate the thresholds of decision-making ahead of time because the model’s decisions are only as good as the inputs it knows about.
Perhaps this would-be partygoer learns their former significant other will be there. That might sway the decision to an absolute “no,” regardless of the other answers. But a decision tree created to decide whether or not to go to parties wouldn’t take that input into account unless it knows to handle former partners and had been trained on some examples. In other words, without the proper training data, the AI would make the “wrong” decision.
AI is More Human Than You’d Expect
Without the right training data, AI models cannot be effective, and the ramifications can be far greater than mistakenly attending a party where you might run into an ex.
If your training data is not sufficient in quantity and quality, your model could return inaccurate predictions or inadvertently discriminate on the basis of gender or race.
At the end of the day, AI is created by humans. If we want effective, responsible AI, it is up to us to create it, through proper training.
For more advice on building effective, responsible AI, you can find Real World AI on Amazon.
Alyssa Rochwerger is a customer-driven product leader dedicated to building products that solve hard problems for real people. She delights in bringing products to market that make a positive impact for customers. Her experience in scaling products from concept to large-scale ROI has been proven at both startups and large enterprises alike. She has held numerous product leadership roles for machine learning organizations. She served as VP of product for Figure Eight (acquired by Appen), VP of AI and data at Appen, and director of product at IBM Watson. She recently left the space to pursue her dream of using technology to improve healthcare. Currently, she serves as director of product at Blue Shield of California, where she is happily surrounded by lots of data, many hard problems, and nothing but opportunities to make a positive impact. She is thrilled to pursue the mission of providing access to high-quality, affordable healthcare that is worthy of our families and friends. Alyssa was born and raised in San Francisco, California, and holds a BA in American studies from Trinity College. When she is not geeking out on data and technology, she can be found hiking, cooking, and dining at “off the beaten path” restaurants with her family.
Wilson Pang joined Appen in November 2018 as CTO and is responsible for the company’s products and technology. Wilson has over nineteen years’ experience in software engineering and data science. Prior to joining Appen, Wilson was chief data officer of Ctrip in China, the second-largest online travel agency company in the world, where he led data engineers, analysts, data product managers, and scientists to improve user experience and increase operational efficiency that grew the business. Before that, he was senior director of engineering at eBay in California and provided leadership in various domains, including data service and solutions, search science, marketing technology, and billing systems. He worked as an architect at IBM prior to eBay, building technology solutions for various clients. Wilson obtained his master’s and bachelor’s degrees in electrical engineering from Zhejiang University in China.
