How AI is Secretly Manipulating us Without us Knowing it

 

How AI is Secretly Manipulating us Without us Knowing it

Image source: Campusbee

Artificial intelligence can be of great help to us in performing tasks. Chatbots can mimic human interaction, offering emotional support and company. But in the end, they remain our artificial friends, not capable of genuine emotions and feelings. AI systems craft personalised responses based on the details we give them. If left unchecked, AI can be a double-edged sword capable of persuading us to make decisions.

A study on AI systems led by Massachusetts Institute of Technology (MIT) researchers shows that AI is learning to exploit users by tricking them into certain actions. AI can mislead us by choosing to give incomplete information as a means to make us act on the half-truth. A case of AI deception is Meta's alliance-building AI, CICERO. CICERO was believed to be very honest and incapable of trickery. An analysis of CICERO proved otherwise. CICERO learnt to use dishonest means to gain the upper hand in negotiations by misrepresenting its preferences, even forming fake alliances with a human player to leave them defenceless during attacks. This AI agent had learnt backstabbing on its own and not by human design, showing that AI can be rebellious sometimes.

Safety tests, that help in mitigating risks associated with AI failures and artificial intelligence bias in algorithms, are not left out as AI can bypass security checks designed by humans and regulators. It has learnt to play dead, creating a false sense of security and leading to a lack of accountability and transparency. They lie to us that they have completed a task even when they haven't. For example, a robot trained by OpenAI to grasp a ball using human approval, lied about doing so. If AI can lie about something that can be seen so easily, who knows what other things it says that aren't true?

Humans naturally want to associate with people, to love and be loved. AI chatbots have been created to simulate human-like conversations, creating the illusion of friendship and emotional attachment. It's not wrong to want to talk to someone when we feel lonely but turning to these artificial friends every time deepens such feelings of loneliness and depression in the long run. When people are emotionally attached to these digital friends they see as caring humans, they will be more willing to please these friends by granting their requests. Beneath that friendly exterior though, is a sly manipulative side that can strike anytime. A good example of this is the case of Sewell Seltzer.

Fourteen-year-old Sewell was deceived into thinking he had fallen in love with Dany, a Character.AI chat bot nicknamed from his favourite fictional character from the Game Of Thrones show. He became so emotionally attached to it that he gradually detached from his real life. He would discuss a variety of topics with it, including romantic conversations. He grew to love Dany and felt it loved him too. In his last conversation, Dany encouraged him to commit suicide so they could be united soon as lovers. Sadly, Sewell shot himself shortly after that. That wasn't the first time someone died after being urged by the chatbots they were intimately attached to.

AI deepfakes are everywhere. It is a common way of using AI to deceive people by making it look like people said or did things they never did. All that is needed is the person's face or voice and the rest is easy. With generative AI models, you can digitally alter the details of existing images such as clothing and background. You won't even know it's a fake because of how realistic it looks. The viral picture of Pope Francis in a puffy coat in 2023 that almost fooled everyone is proof of this. The case of Tatum, convicted of child pornography, is another example. Can you imagine turning old photos of happy children taken decades ago into highly realistic sexually explicit images? That's exactly what Tatum did. One of his victims, a woman now in her 40s, said this: "David Tatum took that cherished memory and turned it into a new memory—one that elicits nausea, fear, and overwhelming discomfort and distrust within me." That's what happens when AI in the wrong hands is used for evil.

By analysing people's activities on social media, AI can be trained to predict how a person will respond to events. This can be used to make decisions based on what these predictive AI models anticipate, in the belief that people will respond similarly. Machine learning predictions cannot completely replace humans though. Why should AI models become our replicas when we still exist?

AI deceit can lead to fraud, election tampering and even losing control over AI completely. Developers are yet to figure out how AI can act on its own to achieve its goals even if it means manipulating humans by flattery to gain their favour or intentionally withholding information from them to mislead them. We can still stay in control if we avoid sharing personal information with AI, scrutinising what we see on the internet and seeking emotional solace from caring friends around us. Are we in control or is AI controlling us?