AI solutions are becoming more and more popular, but do we know how tools like ChatGPT work? Creating AI is a complex process, and the creators use a mysterious “black box”.
Black box is a term commonly used in aviation. It is the plate from this device that often allows you to diagnose the cause of a malfunction or a serious accident. However, the term is also extremely important in artificial intelligence research and development. This is the basis for running AI tools like ChatGPT or DALL-E 2.
AI Black Box - how artificial intelligence “thinks”.
Visually, the process of creating AI can be simply represented as follows. We create an algorithm to perform a specific task. The algorithm needs to be provided with a series of training data - after training, an AI model is generated on them. Only it is used in services that people use.
A black box is a term for an artificial intelligence system whose operation is incomprehensible to the creator himself. Simply put, we put the data we know into a box, a certain series of actions take place inside, and then a result comes out of it, which we can check as correct.
Let’s show this with an example. We just created an AI model that recognizes photos of dogs and distinguishes them from cats. That’s why we throw thousands of pet photos into the black box. In a few moments. AI sorts images into two groups with high precision.
We are sure that the artificial intelligence system worked faster, and sometimes even more efficiently than a human. But how did he recognize the animals? By color, ears, paws, muzzle, height at the withers, teeth, tails, fur, body position? We just don’t know.
The logic or precise reasoning of algorithms in action (especially deep learning algorithms) differs significantly from human reasoning. AI does not work according to “common sense”, and the machine’s reasoning is devoid of all layers associated with … humanity. Thanks to this, AI can solve problems in ways that we will never imagine.
AI - black box vs glass box
Most often, the AI model is placed in a black box. The creators of artificial intelligence are interested in the result, and not how the machine works inside. In addition, in this way he can protect the model as his intellectual property - competitors will not copy what the creator himself does not fully know. However, hidden from the world can be an algorithm or a set of training data.
The opposite of popular trends is the “glass box” - Glass Box. This approach to AI is based on the maximum possible transparency of the process of creating artificial intelligence. Companies that create AI in a glass model often make their algorithm, the data used to train it, and the finished model itself publicly available.
An entirely new branch of AI research is also slowly being built. It is an artificial intelligence that can be explained (by a human). The study of so-called explainable AI attempts to explain and understand exactly how advanced algorithms work and on the basis of which they make the right decisions.
Glass boxes will be extremely important as AI advances. In many cases, artificial intelligence will help in sensitive topics where there can be no mistakes. A good example of transparent AI should be e.g. a system that diagnoses patients or offers specific treatment for their diseases.
Source: Wprost
I am George Brown, author at Daily News Hack. I mostly cover economy news and I have been doing this for quite some time now. I have a lot of experience in this field and I’m always looking for new opportunities to learn more.
