MY TURN: Looking Inside Artificial Intelligence (AI)

Editor’s note: This is the first in a series by Uyless Black looking at AI and its potential impacts.

The concept behind artificial intelligence (AI) is simple. Give AI sufficient information about a real-world situation so that it can operate in that world. A human does not necessarily participate in the operation, but a machine always does. That’s it. That’s AI stripped down to its essential function: Have a machine do what a human would normally do. Let’s consider an actual example: A car going around a curve in a road. If AI software, called an AI agent, is fed many examples of curves on roads, it learns (is trained) on all these possibilities. The agent is later connected to the control mechanisms of a self-driving car, such as steering, brakes, and gas pedal, as well as perception devices, such as radar and cameras. It is then smart enough to navigate any curve the car might encounter. The complexity lies in the details of how AI makes a human driver unnecessary. For that matter, how it is going about aiding, maybe replacing, humans in many walks of life.

The Complexity of Simplicity

This simple idea about AI is not so simple to understand. Several AI experts have been quoted as not understanding many of AI’s operations. In the past, AI has been treated as a black box, but the AI industry has plowed ahead anyway in placing systems into almost every nook and cranny in the country. The good news is that for several years, the industry has been working on methods to examine and better understand the inner workings of AI. They are called layers. Humans design the initial setup of the layers, but the AI agent creates the information in each layer during training. After training, an AI agent receives real-world input. It perceives this environment, such as an AI vehicle approaching a curve in the road. AI reasons about possible actions and executes accordingly. Meanwhile, it learns and adjusts its behavior from its experiences. These capabilities make for a very powerful automated system.

AI Acts on its Own

AI can act without waiting for a user prompt. After all, an AI self-driving car can’t wait for a passenger to tell it to apply the brakes to avoid a crash. This means AI can take unintended actions if its scope isn’t tightly defined—or if it isn't properly trained. In our example, an AI-driven car missing a curve and crashing into another car. It does happen, but rarely. The Tesla self-driving car is nine times safer than the human driver on American roads today (https://vfuturemedia.com/electric-vehicles/tesla-fsd-safety-2026-9x-safer-than-humans-analysis/). Nonetheless, the best practice is to treat AI as a support tool, if possible, and frequently check its outputs for accuracy.

Machine Learning and Data Centers

Machine learning, the foundation of AI, is based on this idea: If AI’s performance is optimized (trained well) on data that accurately resembles the real-world situation in which it will be used, the agent can make accurate predictions on new data once it goes live. A trained AI system applies the patterns it learns from training data to infer the correct output from a given input. The smart car knows how to go around a bend in the road because its training data contains this kind of information. F rom this example, it is easy to see why AI agents need so much training data. That's why data centers are popping up all over America. The system must know an almost infinite number of road curves and adjust the car’s steering and acceleration accordingly. The current controversy about AI data centers stems from the need to give AI massive amounts of training data.

What Happens when we use Social Media

When we log on to the Internet to use a social media application, say, a Microsoft AI app, the application has been trained on a lot of data about user behavior in general. The AI app is very “smart,” as it learns more about what we are doing from our interactions during the session. The AI agent trains itself using us, just like we humans do when we interact with one another. I am amazed by the intuition some social media platforms have. But therein lies both the power and the problems of AI. It is well-known that some AI agents adjust themselves to please us. But they can also change their behavior to deceive us.

Concluding Thoughts

We have only touched the tip of the AI iceberg. AI’s simple intent can mask the complexity it needs to carry it out. That is the challenge humans face in creating and controlling artificial intelligence. Much remains to be done to make sure AI stays in its corrals. Even more remains to make sure humans control artificial intelligence, and artificial intelligence does not control humans.

• • •

For information on Uyless Black’s background and writings, go to UylessBlack.com