At some time, we’ve all had the feeling that online apps like YouTube, Amazon, and Spotify, among others, seem to know us better than we do, making recommendations that fit us like a glove. The “magic” behind them is artificial intelligence algorithms. Or, more precisely, machine learning models capable of identifying patterns in large data sets.
Companies in different industries use big data, machine learning, and computing resources to improve tasks such as recommending content, managing inventory, forecasting sales, and detecting fraud. However, despite their seemingly magical behavior, AI algorithms work with statistics to get these results and are generally very efficient as long as there is a standard. That is, they do not have many points outside the curve.
But the new coronavirus pandemic, to cite a recent phenomenon that is still present in our daily lives, has caused changes in our habits and activities. And this made many algorithms have difficulties in maintaining or identifying the patterns in what was called the “new normal.”
Machine learning started from quite simple models, with two or three variables, and, over time and the demands of each business or objective, it evolved. According to the context and the emergence of less accurate forecasts, new variables were added, increasing the accuracy of the models. In addition to date and location, temperature, weather forecast, working days, holidays, seasonal movements, etc., algorithms have become more flexible and resilient.
Most machine learning algorithms, including deep neural networks, share the same core concept: a mapping of features to results. But the artificial intelligence algorithms that power the tech giants’ platforms use far more resources and consider massive amounts of data.
For example, the AI algorithm that powers Google’s ad platform takes your browsing history, search queries, mouse movements, ad pauses, clicks, and dozens (maybe hundreds) of other resources to serve ads that you are more likely to click on. Facebook’s AI uses tons of personal information about you, your friends, your browsing habits, your interaction history to keep your attention on the News Feed, which includes advertisements that are a source of revenue. Amazon uses tons of data on shopping habits to predict what else you would be interested in buying when you’re researching a pair of sneakers.
Ability to deal with the unpredictable
As fascinating as today’s artificial intelligence algorithms may be they certainly don’t see or understand the world the way we do. More importantly, while they can discover correlations between variables, machine learning models do not understand causality.
We humans can find a causal explanation for why water sales are higher on hot, sunny days. AI, however, does not understand the relationship between weather and outdoor activities. Algorithms only work with the variables they were trained on. And they work fine, as long as things remain as they are.
If an unknown virus spreads, contaminates people determining social isolation measures, there is a break in the predictions of the machine learning model. We are also surprised and confused when we face unforeseen events and events that go against the grain. But our intelligence goes beyond recognizing patterns and rules. We have cognitive abilities that allow us to adapt to our ever-changing world.
The AI systems we have can’t manage the unpredictable (yet?). But, possibly, we will evolve until we reach the so-called artificial general intelligence (AGI). A software that has problem-solving capabilities similar to the human mind. This is the kind of AI that can innovate and find solutions for pandemics and other off-curve events.