What is free will? If we humans have so-called free will, can AI have similar abilities? Or can it be imitated? If artificial intelligence cannot have free will, what is the difficulty?

This question will make everyone think deeply, and countless complex ideas will emerge in their minds. So why do we need to discuss the topic of whether artificial intelligence has free will?

First, if AI has free will similar to humans, this will mark a new stage in the development of AI technology and represent an important milestone for humans to achieve AGI. By then we can start thinking more broadly: What can AI do? What are its specialties? How do we maximize our strengths and avoid weaknesses, and how do we improve the R&D process? These are all considerations from a technical perspective.

Second, if AI has free will, we need to re-examine the way we use AI. Should we prioritize developing "subjective" AI? Do we view AI as a true user agent, or do we still view it as a human tool? This may change our traditional understanding of computers, software, and data, and change the positioning of AI in social development from a tool to a digital subject.

Third, when AI has free will, we may need to adjust the existing AI audit control and legal framework. When problems occur, should we hold humans or machines accountable? How are responsibilities distributed? Do existing laws and regulations still apply? These need to be reconsidered.

The fourth and most important question. The autonomous decision-making ability of AI brings a series of issues, including accountability, responsibility and behavioral consequences, while deepening our ethical concerns about AI. For example: Who is responsible for the behavior of AI? Should AI have certain rights? Among them, the issue of bias is the main ethical challenge facing AI autonomy. Because AI systems rely on large amounts of data to learn, they may inadvertently inherit or reinforce existing biases in the data, which calls into question the fairness of AI system decisions. In addition, the opacity of the AI ​​decision-making process increases the complexity of ethical dilemmas, making it difficult for us to understand the principles and basis for specific decisions.

Therefore, the free will of AI systems will bring about typical ethical dilemmas. Can existing technology ethics still play a critical role? Technology ethics helps us set boundaries and constraints for AI systems to ensure that they behave ethically. We need to think deeply about how to effectively integrate human values ​​and principles into AI systems so that their decision-making processes are always consistent with ethical principles.

If this happens, we need to be prepared in two aspects: first, warning education and preventive thinking, including how to guide the public to fully understand the potential and limitations of AI and avoid excessive expectations for AI. Or irrational fears, and how to predict and avoid the risks that AI may bring to prevent AI from getting out of control or turning against customers. The second is to reflect on human beings themselves. The autonomous consciousness of artificial intelligence will prompt us to re-examineHuman moral principles, thinking about the impact of giving machines autonomy on us, and in-depth exploration of the essential differences between humans and AI, as well as the fundamentals of being human.

At present, neither governments nor society as a whole are fully prepared for the above four aspects. If this moment comes suddenly, it will inevitably trigger huge social changes and turmoil. However, the premise of all this is whether artificial intelligence can have free will. If it is fundamentally impossible for AI to have free will, then these discussions will lose their basis.

This article will analyze this issue from multiple perspectives such as philosophy, psychology, spiritual science, and cognitive science, and explore possible conclusions.

Free will psychologically consists of the physical, emotional and intellectual sensations a person feels, resulting in a specific overall "feeling". Free will is our ability to choose among various possible options and decide on actions. It is essentially the perception that our choices are our own and are neither externally imposed nor determined by internal forces beyond our control. Through this perception, we see ourselves as the primary agents of our actions. As far as free will is concerned, this essence is reflected in three aspects: freedom, will and choice. Although this academic concept sounds complicated, as humans, we all understand its core meaning: the inner basis for our ability to make autonomous choices.

Why is free will so important to humans? There are at least two main reasons for this:

First, free will is seen as a uniquely human attribute. As our understanding of animal society continues to deepen, we have discovered that many characteristics that were once considered unique to humans are actually possessed by other animals, but humans are better at them, such as the ability to use complex language, empathy, and the ability to use Technical knowledge and ability to build tools.

Second, free will is an important pillar of society. It is precisely because we have free will that we are held morally and legally accountable for our actions. Research shows that people exhibit more prosocial behavior when they believe in the existence of free will. To deny the existence of free will is to shake one of the foundations of human society.

Since free will is a perceptual experience, some of the most enduring philosophical debates are about whether it exists, what its nature is, and the meaning of its existence.

Some people believe that free will is the ability to act beyond the limitations of external influences and desires. This is similar to Descartes's view of "Cogito, ergo sum" (Cogito, ergo sum), forming mind-body dualism - treating material (body) and immaterial (mind) as two essentially different entities. In this view, the mind is a thinking entity, not limited by physical space, has free will and self-awareness, and is completely subjective; while the body is mechanical, spatial, and governed by physical laws.bundle. This metaphysical libertarianism holds that an immaterial mind, will, or soul overrides physical causation, so that events in the brain that trigger action cannot be explained in purely physical terms.

If Descartes’ philosophical thoughts seem too ancient, then the "Machine consciousness" or "artificial consciousness" proposed by the modern Australian cognitive scientist David Chalmers ( The concept of Artificial consciousness has a more modern meaning.

Chalmers elaborated in his 1995 paper "Confronting the Problem of Consciousness" and his 1996 book The Conscious Mind: In Search of a Fundamental Theory His "hard problem of consciousness." He believes that consciousness originates from human subjective feelings and experiences, is a basic attribute of human beings, and is ontologically independent of any known physical attributes. While we can explain how the brain processes information, we cannot explain why these physical processes produce subjective experiences. Chalmers calls his view "naturalistic dualism": naturalism in that he believes that mental states occur "naturally" in the human brain; dualism in that he believes that mental states are ontologically distinct from the human body physical system and cannot be reduced to a physical system. He also proposed the "Philosophical Zombies" thought experiment, explaining that even if a system can perfectly imitate human behavior, it does not mean that it has a true conscious experience.

According to this view, free will is the exclusive privilege of human beings and can only be innate. Therefore, no matter how advanced artificial intelligence is developed, true free will will never be formed.

In addition, there is another view that also denies the possibility of artificial intelligence having free will. This view holds that free will itself does not exist, and humans do not possess free will. Since the concept of free will itself is not established, it is meaningless to discuss whether artificial intelligence has free will.

This view is determinism. The philosophy of determinism holds that every event in the universe is determined by the cause and effect of previous events. According to determinism, free will does not exist, everything operates like a giant clockwork mechanism, and our choices are only part of this mechanism. Determinism is closely related to the principle of causation: there can be no effect without a cause.

Determinism can be proven by natural laws (such as the law of gravity) or by explicitly constructed systems (such as computer programs). In scientific fields such as biology, physics, and psychology, determinism is reflected in theThe rigorous derivation of natural laws and causal relationships emphasizes the clarity and inevitability of the causal chain.

Other forms of determinism are based on speculation or assumptions. This type of determinism does not necessarily require direct causal inference and empirical support, as long as a correlation between facts can be established. For example, in the fields of philosophy or cosmology, determinism is mainly based on theoretical assumptions and model frameworks, and this approach is of particular value in fields that have not yet been proven.

The extreme form of determinism is fatalism, also known as universal determinism or predeterminism. Fatalism is the belief that everything is predetermined and that every event in the universe is determined before it occurs. This is a rigid metaphysical principle that denies the influence of human experience and thinking on the world.

Determinism and free will are mutually exclusive: if we admit that the world is dominated by an inevitable chain of cause and effect, we cannot simultaneously admit that humans have free will. In order to clarify this relationship, philosophers proposed "incompatibilism", which believes that the deterministic universe and the existence of free will are completely opposite. In contrast to dualism, incompatibilists believe that we must choose between the determinism of the universe and human free will.

The typical representative of incompatibilists is American neuroscientist Benjamin Libet. In the 1980s, he proposed through nuclear magnetic resonance experiments that humans do not have free will, and that consciousness is only the result of brain reactions and is a basic biological attribute of humans.

Libet and other scientists—including Haynes of the Bernstein Center for Computational Neuroscience in Berlin, and neuroscientist Itzhak Fur of the University of California, Los Angeles, and Tel Aviv Medical Center in Israel— Itzhak Fried and others - found through brain scans that the brain activity began before the participants were aware of their choices. In other words, the decision is formed before it enters consciousness, and consciousness arises only after the decision has been made.

Libet conducted a button experiment: subjects were asked to press a button when they wanted to, while EEG was recorded. It was found that the motor cortex of the brain was already excited half a second before people were aware of making the decision. Follow-up studies have even found that excitement in subcortical brain areas precedes awareness by a full 7 to 10 seconds. By observing the activity of 256 neurons, scientists were able to predict a subject's choice with 80 percent accuracy—for example, certain areas of the brain may have already made a decision before a person realizes what dish to order.

Neurobiologist Robert Sapolsky pointed out in his 2023 book "Determined: A Science of Life Without Free Will" , the apparent freedom of choice is actually affected by genetics, brain chemicals, and environmental factors.There is no real "subjective factor" due to the profound influence of factors and past experiences. Human intelligence is built on complex biological mechanisms, including the interaction of genes, brain structure and function, and environmental factors. All current sensory inputs, after being processed by the neural network of each module in the brain, determine the current output. There is no involvement of the so-called "superego" or "self" in this process.

Therefore, some neuroscientists believe that although humans do make choices and deliberate decisions, this is essentially a process that follows the laws of nature. They view humans as thinking machines and believe that terms such as "making decisions" or "choosing" are inaccurate.

According to the incompatibilist point of view, since humans do not have free will, why should artificial intelligence have it? Thinking about it conversely, if free will is just the result of multi-layer calculations performed by the brain through a model composed of genes and neural networks after receiving input, then artificial intelligence may be able to achieve this ability more easily.

This means that AI does not need to achieve true self-control and judgment decision-making, because everything comes from data. If human behavior is based on predictable biological and physical laws, then AI decision-making can also be realized through causal chains. AI has no fundamental obstacles in simulating human decision-making and behavior. It only requires sufficiently complex algorithms and data support. AI does not need to have free will; at best it can simulate a human-like decision-making process. Therefore, AI does not have to bear ethical and moral responsibilities because problems with AI can be traced back to its algorithms and training data. It may not be reasonable to assign "responsibility" to AI because its behavior essentially depends on the design and training environment, not "free will."

In the view of incompatibilists, the path for AI to achieve free will is either completely impossible or extremely simple. But is it really that simple?

If we accept this deterministic view of "no free will", it will have a subversive impact on education, law, religion and social policy. Moral responsibility and criminal responsibility in the traditional sense need to be redefined. This view advocates a "treatment rather than punishment" attitude towards criminal behavior, believing that criminal behavior should be regarded as an understandable and treatable result rather than a "moral fault" that requires punishment. In this view, we cannot even make moral judgments about people because each person's actions are caused by their determinants.

This view has triggered heated debates in philosophy, ethics and legal circles. Critics question: Is their definition of free will too narrow? Does determinism ignore the effects of randomness and chaos theory? As mentioned above, free will also plays an important role in maintaining social norms and operations.

Thus, the discussion about free will and determinism gave rise to a second theory - compatibilism. Compatibilism believes that a deterministic universe and free will are not completely opposite. In a deterministic world, free action is possible. This view seems to offer a more middle-of-the-road approach.

Compatibilists usually define "free will" as the ability of human subjects to make free judgments and choices based on their own motives. The definition of free will does not depend on the truth or falsity of causal determinism. This view sees free will as a form of autonomy, the ability to live by your own rules rather than submitting to external control and coercion.

Compatibilists accept the existence of both free will and determinism. This idea can be traced back to the Stoics of ancient Greece, and contemporary cognitive scientist and philosopher Daniel Dennett writes in From Bacteria to Bach and Back: The Evolution of the Mind Back: The Evolution of Minds explores the origins and evolution of the mind in depth. He believed that mind and consciousness were products of evolution rather than design. From simple bacteria to the complex human brain, it's all explained by natural selection and evolution. "Building from the bottom up" is key: consciousness evolves gradually from unconscious processes. Human language was an important turning point, making it possible to communicate and store complex ideas, moving humans beyond the understanding of simple creatures. The uniqueness of the human mind stems from cultural evolution, and language, art, and technology are important tools for cultural evolution. He particularly emphasized the driving role of "meme", the basic unit of cultural transmission and accumulation, in the evolution of the mind.

Dennett divides the evolution of the mind into three stages: biological evolution (from bacteria to complex multicellular organisms), cognitive evolution (the brain develops complex perception and response capabilities), and cultural evolution ( Ability to reflect and create through memes). He views consciousness as a functional tool, a cognitive "user interface" that has evolved to meet survival needs and is used to simplify complex computational processes in the brain.

When discussing whether machines may have consciousness, Dennett pointed out that although current artificial intelligence is functionally close to the human mind, it still lacks true "understanding capabilities." The development of artificial intelligence has provided important revelations for us to understand the operating mechanism of the natural mind, but imitation is not the same as possession. He opposed traditional "mind-body dualism" and argued that all properties of the mind can be explained by physical processes. He criticized treating consciousness as an irreducible mysterious phenomenon and advocated the use of scientific methods to deconstruct and understand consciousness. If genes are the basis of biological evolution, then memes are the basis of cultural evolution. He compared the evolution of the mind to ""Tower's spiral staircase", each floor is built on the previous one and gradually climbs up.

Based on Dennett's view, it is possible for artificial intelligence to achieve free will by imitating the evolutionary process of humans. , without adding any mysterious "soul" component

However, critics of compatibilism argue that this idea requires both free will and determinism. Thinking, in fact, uses the name of freedom to cover up the essence of determinism, falling into a "quagmire of escape" and reducing it to a word game. This makes it difficult for us to judge what is free will and what is the product of determinism in specific circumstances. < /p>Transfer and decomposition of the problem

We have discussed three views above: The first is the view proposed by David Chalmers, who believes that free will is unique to humans. Yes, even if a machine shows similar characteristics, it cannot be regarded as truly having free will, because AI is not a human being. The second is Robert Sapolsky's incompatible determinism view. He believes that humans do not have free will, and everything is free. It is a deducible working principle. Free will is just a false proposition. Since it does not exist at all, then AI can naturally obtain similar reasoning and cause and effect. Transformation ability. The third is Daniel Dennett’s compatibilism view. Since human free will is acquired through evolution, AI can also be gradually evolved to imitate humans

In freedom There are still disputes between different parties on the topic of will. How should we deal with it? We might as well follow the idea of ​​solving complex problems and first explore the precursors to the formation of free will. What are the prerequisites and whether these conditions can be realized. If the prerequisites are realized, free will may be possible.

The prerequisites for free will can be divided into three categories: The first category is basic. Conditions, including human self-awareness (Consciousness) and subjective experience, people without consciousness cannot navigate complex moral dilemmas. Weighing and decision-making; and subjective experience allows people to feel happiness and a sense of accomplishment at work, rather than just conducting purely rational analysis. The second category is functional conditions, that is, the ability to obtain information and knowledge and the ability to perform actions. , that is, the ability to choose independently. For example, if a person does not understand the diversity of careers or lifestyles, he cannot make free choices among them. Classes are open environmental choices. For example, although prisoners may have consciousness and subjective experience, they have extremely limited choices in their cells. These three types of conditions constitute the basic premise of free will.

Artificial intelligence solves the problem. We are already quite good at functional issues and open environment selection, so there is no need to go into details. The real difficulty lies in the two most basic elements: awareness and experience. Together with the brain, these two form an MVC architecture (model-view-controller) in computer design: the brain is the model, the experience is the view, and the consciousness is the controller. If the large model of artificial intelligence can be analogized to the brain. , then the key is to understand how autonomous experience and consciousness are formed

Let’s explore the relatively simple ones first.subjective experience issues.

In fact, there are numerous examples showing that AI can simulate or partially imitate human subjective experience. For example, chatbots such as ChatGPT can generate emotional responses based on context, including comfort, sympathy, or encouragement. By analyzing speech, facial expressions or text, AI can identify a user's emotions (such as sadness, anger or joy), and this technology has been widely used in mental health monitoring and human-computer interaction. NPCs in the game also exhibit characteristics similar to subjective experience, being able to simulate the character's motivations, emotions, and behaviors. For example, characters in "Zelda" will "emotionally react" to environmental changes (such as weather). Digital companions (such as Replika) can build personalized conversation experiences based on user input, showing "care" or "understanding" qualities. In addition, AI's ability to create novels, paintings, songs, and videos presents some kind of creative expression similar to "inspiration." So, do these simulations mean that AI has truly acquired human subjective experience capabilities?

On this issue, we can refer to the views of Nobel Prize winner in physics Geoffrey Hinton. In his important speech at the "Vector Institute's Remarkable 2024" event in February 2024, he made it clear that artificial intelligence systems already have subjective experience.

He pointed out that although most people still insist that AI is fundamentally different from humans, they ignore a deep understanding of the nature of subjective experience. He said:

"When we use the words 'subjective experience', we are not describing some inner world or theater, but trying to explain us by telling about some state of affairs in the real world. The information provided by the perceptual system. This information, if established, can explain how our perceptual system works normally. .

So the interesting thing about mental states is not that they are some mysterious internal thing, but that they are assumptions about the state of the world that, if true, can explain what happens in our brains. What happened was normal, not something went wrong

So when I say that. When I have the subjective experience of a pink elephant floating in front of me, I am not describing something in some inner world or inner theater, but I am saying that my perceptual system provides certain information that is consistent with reality. If the conditions in the world are consistent, then my perception is valid. The pink elephant is not made of some mysterious substance. They are things that are assumed to exist in the real world, which is why the language we use to describe them is often the same as the language we use to describe things in the real world. I think if there were these pink elephants floating in front of me in the real world, I'd beThe information provided by the perception system will be correct.

Now, let’s take a multimodal chatbot as an example. It has a robotic arm and a camera and is trained to recognize objects. If you put a prism in front of its lens and then put an object in front of it and tell it to point at that object, it might point to the side instead of straight ahead. When you tell it that the object is actually in front of it, it might say, "Oh, I see the object right in front of me." But suppose it has a subjective experience that the object is there. At this point, the chatbot expresses its situation in the same way we would use the word “subjective experience,” and nothing is missing.

In this kind of chatbot, when its perception system goes wrong, it can tell you what happened by describing what conditions need to be in place in the real world for the perception system to give these results. Of course, some things cannot be handled this way, such as impossible triangles. But for the most part, I think we all have the wrong idea about the mind. Once we get past this misconception, we realize that AI is not that different from us. It’s just that they are digital and therefore immortal and much smarter than us or soon will be. ”

According to Hinton’s point of view, with the advancement of AI in perception capabilities and the improvement of large models’ cognitive ability of the physical world, it is entirely possible for artificial intelligence to achieve subjective experience.

Another basic prerequisite for realizing free will is self-awareness. If AI can really simulate humans. If artificial intelligence has human consciousness and meets the other conditions we discussed before, it can be said that artificial intelligence has achieved free will.

There are different views in the academic community that consciousness is human. Unique mental attributes, which directly denies the possibility of AI acquiring consciousness. Another view regards consciousness as. A biological attribute that exists objectively. If we adopt the latter view, we may explain the nature of consciousness through neuroscience and create "artificial consciousness" by simulating neuroscience principles, allowing AI to break through human beings' "sublime self". "Inherent cognition.

First of all, we must understand the difference between intelligence and consciousness. In other words, we must answer: A If I have intelligence, can I have consciousness?

From a human perspective, are intelligence and consciousness the same thing? Psychological research in the 1990s found that even if a person has a higher level of level of intelligence, and may not be able to think and act rationally Canadian psychologist Keith Stanovich. Stanovich thus proposed the concept of "rational disorder". He believed that intelligence and rationality are different concepts: intelligence is related to cognitive ability, while rational consciousness is more concerned with the optimization of thinking processes and the quality of decision-making, including critical thinking, probabilistic reasoning,Scientific thinking and avoiding thinking biases, etc. Similarly, American psychologist Raymond C. Nickerson also emphasized the complexity of cognitive processes, including knowledge acquisition, information processing, problem solving, and creative thinking. They both point out that humans are susceptible to various cognitive biases in their thinking and decision-making processes, leading to irrational behavior.

The most famous person in this field is the recently deceased Nobel Prize winner Daniel Kahneman. His research in behavioral economics and psychology reveals the irrational nature of human cognitive biases and decision-making processes. In the book "Thinking, Fast and Slow" (Thinking, Fast and Slow), he discussed in detail the two thinking systems of human beings: System 1 is the automatic, intuitive way of thinking, responsible for quick judgment and decision-making; System 2 Is a slower, more logical, and more effortful thought process responsible for complex reasoning tasks. Kahneman demonstrates the complexity of human intelligence and how we use different cognitive resources in different situations.

Intelligence enables humans to understand complex situations, create and use tools, and think abstractly. Reason and consciousness are based on intellectual abilities and help us make judgments and decisions. Reason allows us to move beyond gut reactions and process information in a deeper, more systematic way. However, human decision-making behavior is often full of irrationality. The solution to irrationality is to use the faculty of reflection—a faculty that transcends reason, that is both encompassing and greater than reason. In addition, the human mind also has the ability to "not think", that is, it can consciously not think about certain things when it is necessary to protect the consistency of thinking. These abilities of "reflection" and "non-thinking" all originate from human autonomy, allowing us to be aware of our own existence, experience emotions, and reflect on our own thinking and feelings.

Consciousness can be divided into conscious and unconscious: conscious is the process of brain activity through thinking, an arbitration or decision-making mechanism, similar to Kahneman's System 2; while unconscious is unconscious brain activity. , more like a behavioral pattern, similar to System 1. It should be noted that we use the term "unconscious" rather than "unconscious" - although Freud's concept of "unconscious" is widely circulated, contemporary psychologists and psychiatrists generally use the term "unconscious" , and "subconscious" is mostly found in the works of philosophers and writers.

The above are the theoretical foundations laid by psychologists for the study of human intelligence. Returning to the discussion of general artificial intelligence, if it is relatively easy to realize various abilities and logical reasoning thinking through machines, then how to make AGI have self-awareness or free will, so that it can reflect or not think?ability has become one of the biggest difficulties in the industry.

Neuroscience and artificial intelligence have a long history, which has led to the formation of the connectionist school. Representatives include McCulloch, who collaborated with Pitts, and Donald Hebb, who had an important influence on Rosenblatt. Humanity's understanding of neural networks has inspired generations of neural network algorithms. Modern artificial neural network algorithms use activation functions to simulate neuron outputs, use weighted inputs to abstract synapses in neural networks, and simulate the distributed hierarchical network structure of the human brain instead of using the von Neumann style storage and processing methods. However, neuroscientists still know little about some basic structures and principles of the human brain, such as the hierarchical structure of the cerebral cortex, feedback connections within brain regions, and the modules of multi-layered brain building blocks. Therefore, how to achieve artificial consciousness is still a big problem facing neuroscientists.

Although neuroscientists have paved the way for general artificial intelligence, this road is extremely bumpy, mainly because humans have extremely complex neural network systems. The theory of "Neural Darwinism" or "Theory of Neuronal Group Selection (TNGS)" proposed by American biologist Gerald Edelman, also known as "Neural Darwin" doctrine". He believes that neurons in the brain form complex neural networks through a process similar to Darwinian natural selection. Human consciousness and cognition are generated through the dynamic interaction of extensive neural networks, rather than relying on single neurons or simple neural circuits.

In the 1980s, Hans Moravec of Carnegie Mellon University in Pittsburgh proposed "Moravec's paradox": allowing computers to reach the level of adults It is relatively easy to play chess at a high level, but it is extremely difficult or even impossible to give it the perception and action capabilities of a one-year-old child. The more complex logical reasoning is for humans, the easier it is for machines to master; while general cognition that humans find simple is more challenging for machines. This paradox seems to confirm that human intelligence was formed after a long period of evolution and natural selection. Various human skills are biologically achieved through the process of natural selection. During evolution, natural selection preserves better designs. The older a skill is, the longer it has been improved by natural selection, the more complex its mechanism is, and the more difficult it is for machines to imitate. Humans' advanced reasoning and logical thinking abilities only appeared after the development of civilization. The evolution time is short. Although it is complex on the surface, it is easier for machines to implement it. The most reasonable explanation for this paradox lies in the evolutionary theory of the human brain.

The formation of human intelligence has gone through a long timeA long and complex evolutionary process spanning tens of millions of years. From the appearance of the earliest animals on earth 600 million years ago to the evolution of primate mammals 65 million years ago, it took hundreds of millions of years. During this period, animals have acquired basic survival intelligence such as sensory processing, motor control, and instinctive behavior. With the advent of primates, the brain began to develop more complex social behaviors, tool use, and problem-solving abilities. During this stage, the cerebral cortex begins to expand, especially areas responsible for higher cognitive functions, such as the prefrontal lobe. About 2 million years ago, human ancestors - Homo sapiens (Homo sapiens) appeared. Not only did the size of the brain further increase, but more importantly, the connections between various parts of the brain were enhanced and optimized. Tool use gave Homo sapiens an evolutionary advantage, requiring larger and more complex brains to coordinate fine hand movements. 40,000 years ago, human intelligence showed modern features such as controlling fire, creating artistic expressions such as cave paintings and personal decorations, and developing the ability to speak. The evolution of the human brain is not only an increase in size, but also an increase in gyri (folds on the surface of the brain) and sulci (grooves). The complexity of these structures greatly increases the surface area of ​​the brain and supports more neurons. and more complex network connections. This is just like the computing power resources and algorithm structure of computers continue to improve with the development of intelligence. 10,000 years ago, as humans began to settle and develop agriculture, social organization and technology changed significantly, and human intelligence such as writing and numerical calculations also emerged. Since then, mankind has relied on this intelligent system to move from an agricultural society to an industrial society, and then to an information society. In this long evolutionary process, heredity is the key to maintaining the continuation of human intelligence. It not only retains the characteristics of the previous generation, but also creates new possibilities.

How to build a machine "neural network" that simulates the working mechanism of the human brain? How can a computer achieve in a short period of time what it took tens of millions of years for human intelligence to achieve? This seems to be the only way to achieve general artificial intelligence. At present, the scientific community has proposed several theoretical models that may allow machines to produce "artificial consciousness."

The first is the whole-brain simulation method, which builds a model by scanning and mapping the biological brain in detail, and then simulates it on a computer system. This kind of simulation must be highly restored to the original brain to achieve the same behavior. The Blue Brain project at the Institute of Brain and Mind at the Ecole Polytechnique Fédérale de Lausanne in Switzerland is trying to create synthetic brains by reverse engineering mammalian brain circuits. In 2004, Henry Markram, the lead researcher on the project, pointed out: "Every molecule in the brain is a powerful computer, and we need to simulate the structure, function and interaction rules of trillions of molecules. This requires trillions of times more computing power than existing computers. "If you want to comprehensively model neural behavior at the molecular scale, the computing power required is even more unimaginable."

If whole-brain simulation is difficultIf the degree is too large, can an integrated approach be adopted that combines various cognitive abilities? The Integrated Information Theory (IIT) proposed by Italian neuroscientist Giulio Tononi contains two core concepts: differentiation and integration. Differentiation means that the system can produce different specific states in response to external input or internal changes; integration requires that all parts of the system coordinate and interact and show unified behavior. When the system reaches a certain level (φ value) in these two aspects, general intelligence is possible. However, applying IIT to AI development still faces many challenges, such as the difficulty in measuring and optimizing differentiation and integration values ​​in artificial systems. Although currently not applicable to the entire human brain, IIT has been used in models of the visual cortex.

If it is difficult to coordinate differentiation and integration, can different intelligent modules be allowed to perform their respective functions? Global workspace theory (GWT) proposed by American psychologist Bernard Baars treats the brain as a neuron computer. Unlike traditional computers, the brain does not have a central processor, but is composed of distributed modules responsible for language, memory, hearing, vision and other functions. The global workspace is like a stage where various neural processes can bring information and become the focus of consciousness for processing by other parts of the brain. This theory gave birth to conscious neural computing models such as Intelligent Distributed Agent (IDA) and its upgraded version LIDA. This type of model enables the system to make decisions quickly through multiple agents processing in parallel, communicating with the help of a global workspace.

The neuroscience community has also proposed the theory of predictive coding, which believes that the brain handles cognitive functions such as perception, motor control, and memory by predicting and comparing existing experience with future expectations. The brain is constantly making predictions about the external world and updating these predictions through sensory input. When the actual input does not match the prediction, a prediction error (Prediction Error) will occur. This error signal is used to adjust the brain's internal model to improve future prediction accuracy. Predictive coding theory inspired a class of neural networks based on error minimization learning and promoted the development of large language models such as Transformer and BERT. However, the theory still faces challenges such as verification accuracy, insufficient generalization ability, and intelligence complexity beyond the explanation range of predictive models.

In general, compared with humans, current artificial intelligence algorithm models are still limited to local systems with specific functions, while the human brain and nervous system are a collaborative giant covering various intelligences and functions. system.

Therefore, the development path of artificial intelligence is still long. havePeople have asked, can the reflective ability that has been realized in the current large model be used as the basis for realizing free will?

Reflective ability refers to the ability of human beings to self-examine and evaluate their own thoughts, behaviors and decisions. This ability plays an important role in human behavior and is seen as a potential manifestation of free will: through reflection, humans can analyze their own motivations, choices and consequences and thereby adjust future behavior. This ability demonstrates autonomy compared to unconscious reactions. Reflection allows humans to pause in the face of impulsive or automatic reactions and consider more complex factors or long-term goals, which appears to enhance the possibility of free choice. However, the ability to reflect does not necessarily prove that humans possess true free will, as the reflective process itself may still be subject to neurological and environmental conditions.

From a deterministic point of view, reflection is only part of the causal chain: the ability to reflect does not give humans real freedom, because the reflection process itself is also driven by the physiological mechanism of the brain and past experience. This suggests that rumination may be just one link in a more complex causal mechanism.

Compatibilism believes that reflection strengthens human free will. Through reflection, humans can choose actions that are more consistent with their own values ​​or long-term goals, such as controlling emotions or changing bad habits. These behaviors are driven by reflection rather than just instinctive reactions.

To sum up, the ability to reflect is an important feature of the human mind, helping us examine behavior and make complex decisions. It enhances human autonomy and is a potential manifestation of free will. However, whether this "freedom" represents true free will still depends on how free will itself is defined. The ability to reflect may say more about the complexity of human behavior than its freedom.

Whether artificial intelligence can develop "free will" similar to humans in function depends first on how we define the concept of "free will" from the perspective of philosophy and worldview. If "free will" is considered to be a unique quality of human beings, then artificial intelligence as a silicon-based life form will naturally not be able to obtain it. Even from the perspective of causal inference of determinism or the evolutionary perspective of compatibilism, human free will was formed after a long period of evolution. Although AI has the ability to learn, judge and reflect, and can also exhibit certain subjective experiences, it is still unable to form a consciousness similar to humans, so obtaining true free will is still out of reach.

Of course, we should not judge this. This is not completely impossible, but the prospects are still unclear and it is extremely difficult to achieve. From my personal point of view, our generation at least does not have to worry about this problem.