top of page
Writer's pictureViktor Fersht

AI can manipulate human behavior

Written by Viktor Fersht, PhD


Humans manipulate AI to manipulate other humans

“Every one has wishes

which he would not like to tell to others,

which he does not want to admit even to himself.”

As an introduction: my discussion this them with Chat GPT

I spoke with one of the most famous artificial intelligences Chat GPT-4 about whether AI can program human behavior.


My question: Can AI could manipulate human behavior?

Chat GPT-4: AI's potential to manipulate human behavior is multifaceted and has both benign and malicious implications. Understanding these mechanisms can help us take precautions to guard against unwanted manipulation while still benefiting from AI's advantages. Here are several ways AI could influence or manipulate human behavior:

1. Personalized Content and Recommendations: Platforms like YouTube, Netflix, and various social media sites use AI to curate and suggest content based on individual user behavior. While this is intended to enhance user experience, it can also create echo chambers, where individuals are only exposed to a narrow range of perspectives, reinforcing existing beliefs.

2. Targeted Advertising: By analyzing browsing habits, search queries, and social media activity, AI can predict what products a user might be interested in, and show them highly personalized ads. This can shape purchasing decisions and, on a broader scale, consumer trends.

3. Nudging in User Interfaces: Design elements, driven by AI insights, can be used to nudge users towards specific actions, such as clicking on a particular link, spending more time on a platform, or sharing certain content.

4. Social Proof and Conformity: AI-powered bots can create an illusion of consensus or popularity on social platforms, nudging people towards particular views or behaviors by exploiting humans' tendency to conform to perceived majority opinions.

5. Deepfakes and Synthetic Media: AI can generate realistic but entirely fictitious media content, which can mislead viewers and sway public opinion or trust.

6. Emotion Recognition and Manipulation: AI can be used to detect user emotions (from textual, vocal, or visual cues) and adapt its responses to exploit or manipulate these emotions. For instance, in gaming or shopping platforms, detecting frustration might trigger discounts or offers.

7. Game Mechanics and Engagement Tools: Many digital platforms incorporate game-like elements (often called "gamification") to boost engagement. AI can tailor these elements to individual users, keeping them engaged longer or steering behavior in desired directions.

8. Persuasive Chatbots and Virtual Assistants: As these AI entities become more sophisticated, they can use a range of persuasive techniques to encourage users to adopt certain viewpoints or take specific actions.

9. Behavior Prediction: With sufficient data, AI can predict users' future behaviors and decisions. Platforms can exploit this knowledge to steer those behaviors in desired directions before they even occur.

10. Real-time Feedback: Devices like fitness trackers use AI to give users feedback on their activities and offer suggestions for improvement, subtly guiding users' behavior over time.

While many of these tactics have benign or even beneficial intentions, there's potential for misuse, especially when the end goal is profit or power rather than the user's well-being. The key to navigating this landscape responsibly is transparency, user control, and robust ethical guidelines for AI developers and platforms.


Viktor: Fersht: Could you penetrate human consciousness and subconscious?

Chat GPT-4: While it is not possible to physically penetrate someone's consciousness and subconscious, psychoanalytic techniques, subtext analysis, and general semantics can be employed to gain a deeper understanding of an individual's thoughts, emotions, and motivations. Psychoanalysis, developed by Sigmund Freud, focuses on exploring the unconscious mind to uncover repressed memories, feelings, and desires that may influence behavior. Subtext analysis allows us to decipher the hidden meanings and emotions behind a person's words and actions. General semantics, a field that studies the relationship between language and thought, can also be useful in understanding how people perceive and interpret their experiences. By combining these approaches, it is possible to gain valuable insights into an individual's inner world, but not directly penetrate their consciousness and subconscious.


Viktor Fersht: Is it possible to apply all these methods to determine a person's inner world from a text written or spoken by a person?

Chat GPT- 4: It is possible to gain some insight into a person's inner world through their written or spoken text by analyzing their choice of words, tone, and themes. However, it is important to remember that this analysis only provides a partial understanding of their inner world, as it is based on their external expression. Intrapersonal communication, including self-talk and internal dialogue, is complex and multifaceted, and cannot be fully accessed or understood through text analysis alone. While examining a person's written or spoken text can offer valuable information about their thoughts, emotions, and motivations, it is not a complete representation of their inner world or the full scope of their intrapersonal communication”.


Viktor Fersht: Can AI predict human behaviour?


The authors of the article "How AI programs human behavior" published in TechDetox

also claim that artificial intelligence manipulates people widely.

“Artificial intelligence operates on thousands - perhaps millions - of variables that it has collected about you.

Variables across your entire digital footprint across all the apps and devices you've ever used. This data is combined into a huge database containing everything there is to know about you. The more you do online, the bigger your digital footprint.

Plus, your digital history spans years—and the longer it goes on, the more AI knows about you. Since your credit history can predict your financial behavior, your digital history can predict your every behavior.

If there are gaps in the data about you, AI fills them in with information about people who are somehow related to you. And they give him information about you. It's a simple task: it knows who your social media friends are.

Your digital profile contains precise pressure points on you. If you press correctly and at the right time, the desired result of your behavior is almost inevitable.

This is not a 50/50 guess. This is 90/10 confidence.

You don't realize what those pressure points are. The machine creates your function, which exactly changes the result - changes your behavior. You obey without thinking.

This is not an advertisement for AI abilities.

It is control over your mind.

And it's not so simple: a person's personality trait determines his behavior. What happens at the intersection of brain neuroscience and machine intelligence is incredibly complex. AI finds complex relationships between many variables, traits and self-doubts that you don't even know about. A human analyst, even someone who works directly with your data, may not notice these connections - they are not obvious and not intuitive. Some don't make sense. But the algorithm will find correlations that work.

Even the creators of algorithms do not understand how this happens. Some of the correlations that the machine finds in the data cannot even be defined as a boolean variable by a human programmer. For example, the creators of Youtube were surprised that their algorithm somehow pushes people towards dark content and conspiracy theories - all for the sake of "engagement".

Even talented programmers may not know why it works, but the machine does.

Machines run millions of "split tests" to see if you're more likely to click if an ad or political campaign is presented with, say, a pink or blue background. You don't know about these weird quirks of your subconscious, but the AI knows they're real.

The algorithm knows you better than you know yourself. He took over your subconscious, as evidenced by your data.

The artificial intelligence algorithm to control you never sleeps, it never gets tired. He has no feelings. It has nothing to do with empathy and compassion. He does not doubt whether his tactics of persuasion are right or wrong, good or bad - only if they are effective in achieving the goal.

In other words, it does not include any ethical parameters. This is not a person. The algorithm for managing you is optimized only for the result - behavior modification:

• Force you to buy.

• Force you to vote.

• Get you to donate to a cause.

• Turn you into a gaming addict.

• Force you to participate in riots.

• Turn you into a terrorist.

• Force you to give up your free will in all areas of your human existence.

• Force you to give up more data to further optimize the algorithm.

This AI Algorithm designed to control you is a psychopath.

When the highest bidder bids for any change in your behavior, it can be done. Sometimes with more than 90% accuracy - depending on the situation, the AI can be so sure that you will behave exactly the way it wants.

It's a matter of pressing just the right psychological buttons in your brain. Giving you the right information at the right moment of weakness. Makes you feel angry or depressed - and pushes you towards the behavioral outcome your Function is programmed for.

All this happens below the level of our conscious understanding. You think you are making your own choices when in fact you are being manipulated with great precision.

Your well-being is not a programmed result of your Function.

Programming the Behavior Modification Feature



Your function is programmed inside a machine learning black box. The algorithm feeds on two things: the data of your digital footprint and the task of achieving the desired behavioral outcome in the individual. It then suggests the best function to achieve the result.

The You function dynamically adjusts in real time, bombarding the user with subliminal messages until the desired result is achieved - a change in behavior. The algorithm is self-optimizing. This means that over time it will get better at calibrating your behavior as it collects more data about you.

By providing platforms with our data as part of your daily digital activities, you are teaching AI (for free!) how to manipulate us more effectively. It's called machine learning because... it learns.

This is not fantasy, this is reality. The same process was used to train Google Translate: an incredible amount of text in different languages was fed into the machine, and the algorithm uncovered complex patterns in languages that made human translators almost obsolete. No linguists were involved."

AI uses 2 types of human programming: Neuro-Linguistic programming and Neuro-symbolic programming

Neuro-Linguistic Programming (NLP)

NLP programming described in article of Timofey Skrylnik: “Two Essences of Neuro-Linguistic Programming” that show how psychological NLP connected with AI NLP.

“Neuro-Linguistic Programming (NLP) is a field of knowledge that arose originally in two branches of science: Computer Science and Psychology. In Computer Science NLP stands for Natural Language Processing, in psychology it stands for Neuro-Linguistic Programming.

It's not exactly the same, but at least close. Close, like a unity of opposites. Having the same name and tasks for studying the structures of the language, there are very few works devoted to the study of their connection and the possibility of cross-use of techniques. Psychological NLP, in my humble opinion, has advanced much further in the study of language structures than stemming and lemmatization, which are used in most computer NLP tasks. In this article, I will consecrate the main developments of psychological NLP for computer NLP, I will outline ways of how to create a State of the Art NLP neural networks using psychological NLP.

To reduce the breadth of NLP methods, we will limit this area to the area where neural networks are used for NLP purposes.

Psychological NLP was introduced by Richard Bandler and John Grinder. The most famous book on NLP is The Structure of Magic. I mean, of course, psychological NLP, don't be surprised or scared. It is significant that the title of the first volume of this book is: "The Book of Language and Psychotherapy". If we omit the word "psychotherapy", we find a lot of information about language that is quite applicable to technology.

A neural network is a simplified mental model, at least we hope so. There is certainly something in common between them, at least the word "neuro". At the same time, psychological NLP is also not a full-fledged psychoanalysis. Serious psychologists call NLP "psychology for dummies" - those who cannot read Sigmund Freud read Richard Bandler and John Grinder.

I will cover the transfer of methods from psychological NLP to technical: now it is in demand, they will give me likes, someone will write comments, girls will recognize me “this is a popular technical blogger!”. However, the transfer of NLP technical developments to the field of psychology is possible and, for the purposes of the development of society, IMHO, is more important. Actually, those who study Big Data are approximately doing this: receiving data on the behavior of the masses, they regularly form psychological and sociological models.

Unfortunately, few people can translate the developments into the language of psychologists, sociologists, and popularize the results among them.

Basic NLP Presuppositions for a Neural Network

First, some abstractions. The methods will be in the lower sections.

One of the foundations of psychological NLP is the concept of basic presuppositions: axioms on the basis of which a given area is formed (almost like in geometry). From a computer science point of view, the term "presupposition" can take its rightful place in Embedding, at the beginning of your neural network. If I had more time, I would certainly conduct a study, whether the word "presupposition" is not a stumbling block in the text corpus, whether it is at the forefront of it.

Can the basic presuppositions of NLP apply to a Neural Network?

Below I give the basic presuppositions of psychological NLP and decipher their meanings for the neural network. Computer scientists reading this can understand what you have to deal with and how to pass it through your keras.layers.Embedding. The interpretation for a neural network is subjective and does not claim to be complete.


Here we are designing a neural network. Practicing dialogue with the universe:

- we have a failure: the network works, but badly. - answer of the universe:

“The meaning of communication is in the response you get.” Those. the problem in this form is not solved, something needs to be changed. - another question:

- what exactly to change? - answer of the universe:

“The ability to change the process of perceiving reality is often more useful than changing the content of perceived reality.” Those. change features...

Basic NLP Presuppositions for a Neural Network

First, some abstractions. The methods will be in the lower sections.

One of the foundations of psychological NLP is the concept of basic presuppositions: axioms on the basis of which a given area is formed (almost like in geometry). From a computer science point of view, the term "presupposition" can take its rightful place in Embedding, at the beginning of your neural network. If I had more time, I would certainly conduct a study, whether the word "presupposition" is not a stumbling block in the text corpus, whether it is at the forefront of it.

Can the basic presuppositions of NLP apply to a Neural Network?

Below I give the basic presuppositions of psychological NLP and decipher their meanings for the neural network. Computer scientists reading this can understand what you have to deal with and how to pass it through your keras.layers.Embedding. The interpretation for a neural network is subjective and does not claim to be complete.

Here we are designing a neural network. Practicing dialogue with the universe:

- we have a failure: the network works, but badly. - answer of the universe:

“The meaning of communication is in the response you get.” Those. the problem in this form is not solved, something needs to be changed. - another question:

- what exactly to change? - answer of the universe:

“The ability to change the process of perceiving reality is often more useful than changing the content of perceived reality.” Those. change features...

Approaches to text processing in NLP

Let's go down to earth, let's return to the main task for which the dear reader of readers is reading this material. What is there in psychological NLP for working with text?

NLP contains many constructs for working with text that have not yet been algorithmized.

At the current stage, we managed to recall 5 NLP models suitable for use in a computer. Books have been written on each of the NLP constructs. In a nutshell, it will not work to explain what a metamodel is. In this article I will try to give a reference to the sources and figure out what it could be for computer NLP.

1. Language model and metamodel

Description

Brief description: the metamodel is a model of changes in the levels of abstraction in the language, as well as an arbitrary interpretation of concepts by different people. For example, the word "love" can be understood in different ways: the word "sex" for many is a kind of love, for someone it's different things. The metamodel is not a theory, but a questioning technique that allows you to break through to the concepts hidden behind words that are in a person’s head.

A neuron that would raise / lower / shift the levels of text abstraction would be a revolution! In production, this neuron could complement the existing dialogue models: modern chatbots cannot work in the selected context, adapt to human terminology. Someone under the word “happiness” understands that it is when there is a lot of love, someone when there is a lot of food, and a programmer understands that happiness is when there are computers and all Pentiums around. How can a chatbot determine the true meaning of the word “happiness” for an interlocutor?

The neuron, by the way, will not be super complicated: you need to assemble the body of the text, mark it up by levels of abstraction and displacement of the value area, play with networks similar to the architecture of translator networks. Let's translate a concrete corpus into an abstract one, then back.

2. Neurological levels (Charles Dilts)

Description

Short description: all concepts of the language are divided into 7 levels. Each next logical level - from the mission to the environment - must correspond to the previous one. The list of levels and pictures is on the link.

The logical levels are a refinement of the metamodel, which introduces an additional order into the levels of abstraction: it is not spherical in a vacuum, each level of abstraction has its own purpose and task (purpose), the levels become interconnected.

There are many networks based on logical levels. The first option that comes to mind is a network that extracts a higher order level from the text, which contains a description of the entity in terms of the selected logical level. For example: a business process is a description in the language of the neurological level "place / action". And the competence model is a description in the language of the “ability / opportunity” level. A network that builds a competency model based on the description of a business process will be very expensive.

It is realistic to make such a network: we need text corpora marked up by logical levels, power to work with the encoder / decoder architecture.

3. Metaprograms

Description

Brief description: metaprograms are the main filters of human perception. Psychologists, not programmers, called them programs. A person does not think “in general”: he concentrates on something, chooses a “filter” through which he perceives this something, then he works out according to the algorithm the information that he received through the filter and produces some output and control actions. In NLP, there are 7-10 main perception filters, metaprograms. The list of metaprograms is different for different cultures, parts of society, and may change over time.

A neuron that would draw conclusions not in general, but in the context of a specific object using a perception filter, could become the basis of a “strong” intellect. A set of networks, each of which is trained to work on one of the metaprograms, and a committee that chooses which one to apply to reality, would enable the network to become more autonomous IMHO and, possibly, gain some kind of adaptability in the physical world.



As you can see, building each of these networks is a task for the institute. For each metaprogram, there will be a ResNet50 scale mesh trained on ImageNet. This network was made by the corporation for several years. And this is not a complete list of metaprograms for just one culture! But the effect can be astounding: you see, by assembling 20-30 such networks into an ensemble, we get something really smart.

4. Promotions

Description

Short description: promotion is a technique based on a game with perception filters. By changing the filters of perception of the interlocutor, we enter the emotional plane, gaining access to non-logical planes of consciousness and interesting psychological effects.

Promotions are essentially the application of skills in working with metaprograms. And yes, if we just imagine that we made the networks for the previous paragraph, then their application can be epochal and truly epic ...

5. Double and triple helix

Description

Short description: despite the fact that the triple helix is a technique, there is a clear speech construction behind it - the nesting of logical statements highlighted in tone gives access to the emotional and illogical plans of a person.

The triple helix can be modeled and/or at least elicited in speech. It won't be a revolution. But it will be useful for understanding the structure of materials in the press. And even better for systems that generate text: a variety of literary styles is impossible without a double and triple helix.

The Perspective of Network Development Based on Psychological NLP Models

Interest in developments of the scale outlined in the previous section may be either among large players or amateur enthusiasts. For the Army, strengthening the brains of Fedor's robot with neural networks created according to NLP models can be very attractive. There is also the prospect of developing models that enhance BERT and XLNet - working with the meta model and logical levels could help create the coolest chatbots that can chat about any topic with anyone.

For commercial organizations that are concerned about the recognition of invoice parameters for automating the accounting of fiat currencies, this is of little interest. The task of increasing the number of zeros on the accounts maintained in the computer of their actual owner cannot be solved with the help of neural networks”.

Neuro-symbolic programming

Neurosymbolic programming is a combination of artificial intelligence and cognitive computing that combines the strengths of deep neural networks and symbolic reasoning. Deep neural networks are a type of machine learning algorithm based on the structure and functioning of biological neural networks.

Neurosymbolic models have already demonstrated the ability to outperform state-of-the-art deep learning models in areas such as image and video recognition. Furthermore, it has been shown that they can achieve high accuracy with significantly less training data than traditional models. Due to the recent emergence of this field and the relative paucity of published results, the performance characteristics of these models are not well understood.

IBM Neuro-symbolic AI

IBM consider Neuro-symbolic AI as a pathway to achieve artificial general intelligence. By augmenting and combining the strengths of statistical AI, like machine learning, with the capabilities of human-like symbolic knowledge and reasoning, we're aiming to create a revolution in AI, rather than an evolution.

Neuro-symbolic AI programming of human behavior

How it works?

First, the AI operators reveal, through Neuro-symbolic AI, the symbols of the conscious and unconscious object of our programming.

They then correlate these symbols with the object's desired program of behavior.

Then they teach AI and various chatbots operating in various networks, Internet blogs, TV, radio, communication channels to use these symbols to program the object's behavior.

Programs of the desired behavior of the object begin to be automatically introduced into the unconscious of the object through all possible electronic channels.

What services that work for customers with Neuro-symbolic AI offer:

- analysis of the impact of AI on any person

- protection against the hidden effects of AI on humans

- Neuro-programming of people with deviant behavior

- AI psychoanalysis of people: hidden meanings of events, actions, intentions.



6 views0 comments

Comments


bottom of page