Our lives are all about making decisions. Will you take an umbrella on your walk or not? How about if there’s a 25% chance of rain? How about if you’re wearing something you really don’t want to get wet?
According to Ajay Agrawal, economist, professor, and co-author of Prediction Machines: The Simple Economics of Artificial Intelligence, we use a combination of prediction and judgment to make every decision, big or small. The power of AI comes from its ability to take care of the prediction part of that equation — to tell you the chance of rain. Ultimately, you’ll use human judgment to decide whether you should bring an umbrella based on how much you dislike getting wet versus how much you dislike carrying an umbrella when it doesn’t rain.
Artificial intelligence models can process massive amounts of data to identify patterns and generate accurate predictions. There’s no doubt AI is enhancing how individuals work, helping them be more productive, improving how they collaborate, and upleveling their skills. And when companies use AI for predictive analytics, it can transform how leaders make decisions, helping them better serve customers, allocate resources, and create new and improved processes.
Ajay recently joined our Work Evolved webinar series to talk about the power of AI and predictive analytics. We sat down with Ajay to continue the conversation on predictive AI — here’s what he had to say about its potential for changing entire industries and which employees have the most to gain from using predictive AI tools.
Some people view AI as smart machines, robots capable of talking or thinking like humans. As an economist, how do you think about AI?
Artificial intelligence helps us with prediction. Prediction is using information you do have to generate information you don’t have.
That’s what generative AI models like ChatGPT are doing — using prediction to generate human-sounding language. Generative AI models predict the next token, or word, in a sequence to create a human-sounding message. Another example would be a bank using AI for fraud detection, processing data from past transactions and user habits to accurately predict whether a purchase is fraudulent.
A basic principle of economics is that when something becomes cheaper, we use more of that thing. The rise of AI represents a drop in the cost of prediction. And as prediction gets cheaper, we’ll use more of it.
Let’s talk more about prediction. What is predictive AI and what are some real-world examples?
Predictive AI uses input data to generate output. As a traditional example, we can use 20 years of historical sales data to predict third-quarter sales for next year. The historical data is the input, and the sales prediction is the output.
Less traditionally, we can use the pixels in a medical image to predict the label on a tumor as malignant or benign. The pixel data is the input, and the label on the tumor is the output. That’s also a prediction.
One interesting feature of AI is that, unlike prior statistical techniques, it can utilize multimodal data (e.g., pictures, video, language), not just numbers, as input data, and it can produce predictions in the form of pictures, video, and language.