Move Over Machine Learning; Hello Data-Centric AI Approaches
One step closer to AI thinking like humans.
Photo Credit: CIO
Were you finally starting to understand the big jump in how artificial intelligence processes are becoming more humanlike through machine-learning models?
Well, now we are hearing the next big jump won’t come from machine-learning models, but from data-centric artificial intelligence.
Before we jump into what data-centric AI is and why it might be such a big deal, let’s do a quick review of the progress AI has made over the past decade.
Ten years ago, AI programs only worked with huge amounts of data that ran on a static program. More and more data was needed to increase the probability of getting the right answer. A good example of this would be Facebook’s previous popularity-based algorithm. The more popular a link was, the more data it collected and the chances of the link appearing on other feeds increased.
The need for more data changed with the advent of the machine-learning process, whereby looking at where the data goes within some of the approaches allows the AI process to reprogram itself. In other words, the AI would learn from the process that was happening without human intervention.
At that point, the AI process would decide what might be the right answer. This decision wasn’t just based on larger data sets and probability, but on what it saw was happening.
TikTok used this data-centric approach to show users new videos, and if the users interacted with the videos, the algorithm would expose them to more people. Hence, without human interaction to change the process, the machine-learning model found a solution on its own.
Sometimes the solutions extend to the point where human programmers can’t explain why it works. However, we do know that with more and more data, the machine-learning model adapted to the data set and became more and more humanlike in making decisions.
Photo Credit: MARK RALSTON/AFP via Getty Images / Forbes
If Andrew Ng, computer scientist and one of the premier AI thinkers, is right, this is all about to change and become more data-centric.
Ng’s past experiences include directing the Deep Learning project at Google, directing AI labs at Stanford University and founding Coursera. He believes the next big jump in AI approaches will come from data-centric fine tuning.
We are going to stop treating each piece of data we collect as the same, but focus more on the collection of data and smaller subsets. This will ensure decisions aren’t as easily influenced by low-quality data, which have ruined many efforts in the past.
Gone will be the days of “just give me more data” to improve the AI process. The future will enforce capturing high-quality data, and we will be able to make better decisions from smaller subsets of data.
While we won’t dive into more detail on Ng’s approaches, he and his new company, Landing AI, are helping manufacturers reduce defects in their production systems. Ng doesn’t just add data to eliminate the effects of bad data. Instead, he looks at a subset of the data that isn’t working with the machine-learning model correctly and tries to adapt the AI process for only that subset.
For example, if a speech recognition machine-learning model has trouble recognizing a spoken word correctly due to music in the background, Landing AI will only work with that one subset of data to fine tune the process rather than alter the rest of the data set that doesn’t have music in the background.
So, the next time you are talking to your phone or looking for directions with loud music playing in the background, but your phone still processes your speech successfully, you should credit the new data-centric AI approach in helping make that happen.
There will be many more examples like this to come as the data-centric approach gains acceptance. One step closer to AI thinking like humans.
For Further Information:
This article shares how data-centric communities will be defining the difference between what’s high-quality data and what’s not.
Forbes explained why data-centric AI is the next big breakthrough approach while discussing Andrew Ng’s campaign.
Here is a detailed article on Ng’s thoughts of how data-centric AI can be used.
Best,
Craig