Like it or not, AI learns how to influence you

When I was young I had four Ai agents In my life. Their names are worthless, blinky, pinky and clyde and they try their best to hunt me. This is the 1980s and the agents are the four colorful ghosts in the iconic arcade game Pac-Man.

By today's standards they are not particularly smart, but they seem to be chasing me with cunning and desire. It's been decades before Neural networks has been used in video games, so their behavior is controlled by simple algorithms called heuristic that dictates how they will chase me around the maze.

Most people do not realize this, but the four ghosts are designed with different ”Personalities. “Good players can observe their actions and learn to predict their behavior. For example, Red Ghost (Blinky) is already programmed with a” chasing “personality that directly charges you. The pink Ghost (Pinky), on the other hand, was given an “ambush” personality predicting where you were going and trying to get to her first. That made him really turn away from you.

I remember because in 1980 an experienced person could observe AI agents, decompose their unique personalities and use those views to make them. Now, 45 years later, Tides is about to turn. Like this or not, AI agents are about to be deployed duty in decoding your personality So they can use those views Optimally influence You

The future of AI manipulation

In other words, we will all be accidental players in “The Game of Human” and this will be the AI ​​agent who is trying to earn a high score. I mean this literally – most AI systems are designed to be able to maximize an “Reward“Which earns points for achieving goals. It allows AI systems to quickly find the best solutions. Unfortunately, without regulatory protections, we are likely to be the goal that the AI agent Assigned to the optimization.

I most remember the Communication agents That will contact us in the friendly dialogue throughout our day -to -day life. They will talk to us through photorealistic avatars on our PCs and phones and soon, by Ai -powered glasses It will guide us in our days. Unless there are clear restrictions, these agents are designed to talk to us for information so they can identify our behavior, tendencies, personalities and preferences, and use those qualities in I -Maximize their persuasive effect When working to sell the products to us, we will have services to us or convince us to believe in misinformation.

It's called “AI manipulation problem”And I warn regulators about danger Since 2016. So far, policy manufacturers have not taken decisive action, looking at the threat that is very far in the future. But now, with the release of the Deepseek-R1, the final barrier to the widespread expansion of Ai- the cost of real-time processing-has been greatly reduced. Before this person exits, AI agents will be a new form of target media like that interactive and adaptiveIt can optimize the ability to influence our thoughts, guide our emotions and drive our behavior.

SUPERHUMAN AI 'SALESPEOPLE'

Of course, human salespeople are interactive and also fitting. They interact with us in the friendly dialogue to size us, quickly finding the buttons they can press to replace us. Ai agents They will look like amateurs, they will be able to draw information to us like a fine, it will scare a seasoned therapist. And they will use these views to adjust their tactics in real-time conversation, working at Encourage us More effective than any used car salesman.

It will be asymmetric encounters where the artificial agent has an upper hand (almost speaking). After all, when you contact someone who is trying to influence you, you can feel their motives and honesty. This will not be a fair fight against AI agents. They can make you superhuman skills, but you won't be able to size them. That's because they look like, sound and act so man, we'll unintentionally to trust them When they smile at empathy and understanding, forgetting that their face that affects is just a simulated façade.

In addition, their voice, vocabulary, speech style, age, gender, race and face features are likely to be customized for each of us personally I -Maximize our reception. And, unlike human salespeople who need to size every customer from the beginning, these virtual creatures may have access to stored data about our backgrounds and interests. Then they can use this personal data quickly Earn your trustAsking you about your kids, your work or perhaps your beloved New York Yankees, who lightest you to accidentally neglect your guard.

When AI achieves the high -high -high -power -high power

To teach manufacturers of AI-Powered Risk policy, I helped make an award-winning short film entitled Privacy is gone It was made of the responsible metaveverse alliance, Mindero and the XR Guild. The fast 3 minute narrative describes a young family eating at a restaurant while wearing a glasses of autmented reality (AR). Instead of human servers, the avatars take the orders of each diner, using the power of AI to overthrow them in personalized ways. The film is considered sci-fi when released in 2023-but just two years later, Big Tech engages in an all-out Race of arms to make an AI-powered eyewear That is easy to use in these ways.

In addition, we need to consider the psychological effect that will occur when people begin to believe that AI agents who give us advice are smarter than us In almost every front. When AI achieves an noticed state of “cognitive supremacy” with respect to the average person, it is likely to cause us blind to accept its guidance rather than use our own critical thinking. This respect for a detected superior intelligence (whether truly superior or not) will make the agent manipulated easier to deploy.

I am not a fan of excessive aggressive regulation, but we need clever, narrow AI restrictions to avoid Superhuman manipulation by the agents of the conversation. Without protections, these agents can convince us to buy things we don't need, to believe in things that are not true and accept things that are not in our best interests. It is easy to tell yourself that you are not susceptible, but with the AI's optimization of every word they say to us, we are probably all in -outmatched.

A solution is to inhibit AI agents from establishing Feedback loops where they can optimize their encouragement by reviewing our reactions and repeating their tactics. In addition, AI agents should be required to inform you of their goals. If their goal is to convince you to buy a car, vote for a politician or force your family doctor for a new drug, goals should be stated in front. And finally, AI agents should not have access to personal data about your background, interest or personality if such data may be Before you.

In today's world, the target influence is an extreme problem, and it is often deployed as fired by buckshot in your overall direction. Interactive AI agents will be the target influence on the heat -seeking missiles that find the best path to each of us. If we do not protect this risk, I am afraid we will all lose the game of people.

Louis Rosenberg is a computer scientist and has -set known who is pioneering Halo -The truth and established Unanimous AI.