Reflections on AI Characters: Imagining the Future Through Technology and Science Fiction

Introduction

In recent years, the rapid advancement of large language models (LLMs) has dramatically expanded what AI can do. Development of AI characters powered by these LLMs is now underway around the world, and the creation of characters capable of far more natural, intelligent conversation than conventional AI is becoming a reality.

At Qualiteg, we have also begun foundational research and development in this field.

We are applying the latest LLM technology to develop AI characters capable of more human-like dialogue and emotional expression. The goal of this research is to explore new forms of relationship between users and AI, and to examine both the possibilities and the challenges that AI characters bring.

At the same time, this technology calls for a measure of caution. As AI characters become more sophisticated, we can expect a range of ethical and social issues to emerge: privacy concerns, the risk of dependency, and effects on human relationships, among others. The arrival of AI characters will open a new chapter in the relationship between humans and technology, but this innovation has both light and shadow, and we expect many people will come to hold complicated feelings about it.

In this article, we first take up the enabling technologies behind AI characters and survey how they can be realized from the bottom up. We then turn to science fiction as our subject matter, using it to draw insights from the future — that is, from the world that will exist once AI characters and AI humans have become reality.

Chapter 1: The Challenge for AI Characters — The Lack of Humanness

1-1 Mechanical Responses and a Shortage of Empathy

The greatest challenge facing AI characters is their lack of humanness. Without realizing it, we expect conversations with machines that imitate humans to carry the same warmth and empathy as communication between people. Today's AI, however, cannot fully live up to that expectation.

Self-disclosure is a critically important element in the development of human relationships. As we grow closer to others, we gradually share deeper information about ourselves and open our hearts. This process is essential for building trust and forming deep bonds.

In conversations with AI characters, however, even when this natural process of relationship development appears to be imitated, we can in practice run into certain limits.

In the early stages, a highly developed AI character can "put on" human-like responses and emotional expression. Through conversation with the AI, a user may gradually feel a sense of intimacy and begin to disclose more about themselves. The AI, following its programmed algorithms, can also display appropriate responses and a degree of empathy.

As the dialogue deepens and users begin to seek richer, more human interaction, however, the AI's limits can become apparent. Over long-term conversations, the AI may fail to accurately remember and reflect earlier exchanges and may contradict itself, which comes across as a lack of consistency. In technical terms, the LLM's context size becomes the limit of its short-term memory. The AI may also struggle to fully understand complex human emotions and respond appropriately, exposing a lack of emotional depth. Furthermore, it can have difficulty showing human-like creative empathy in unexpected situations or toward highly personal experiences, which manifests as a limit of creative empathy. And after reaching a certain level of intimacy, the relationship may stop deepening — or even seem to regress — which is perceived as a stall in the relationship's progress. When these factors combine, users begin to sense something off, and to feel the limits of conversing with an AI.

1-2 The Gap Between Expectations and Reality

The higher the expectations placed on an AI character, the greater the disappointment when those expectations are betrayed.

Just as we are deeply let down by the unethical behavior of someone we believed (or assumed) to be a "good person," or of a celebrity we admired, a sudden cold, mechanical response from an elaborately crafted AI character can wound the user deeply.

A similar phenomenon appears in online reviews of medical institutions. Patients' dissatisfaction is often directed not at the medical care itself but at the coldness or mechanical attitude of the receptionists, nurses, and doctors. This parallels the disappointment people feel toward AI characters, and it shows that people are seeking something beyond mere information and efficiency.

Chapter 2: Approaches to AI Character Development

2-1 The Current State of AI Character Development: Building the Technical Foundation

The technical foundation needed for AI character development is steadily coming together. The following component technologies are currently advancing at a rapid pace:

• High-performance large language models (LLMs) — the brain: adaptive responses and a wealth of knowledge
• 2D and 3D character representation technology — the face: expressions and gestures
• Real-time speech recognition (ASR) — the ears: hearing what the other person says
• Text-to-speech synthesis (TTS) — the mouth: natural speech

Advances in these technologies are laying the groundwork for more natural, human-like interaction.

As mentioned at the outset, Qualiteg has also taken note of the potential of the "AI human" and begun foundational research on a platform that integrates these technologies.

As for the difference between an AI character and an AI human: in our view, an "AI human" incorporates the concept of embodiment and is expected to interact more directly with the physical world. One step before that, an "AI character" is positioned as "something human-like" that you can converse with through a screen.

2-2 The Future of AI Character Development: Toward an Integrated Approach

2-2-1 The Current Approach

Today's AI character development is dominated by a "building-block" approach: combining technologies that process different kinds of information and designing the result based on experience and heuristics. In this method, each capability of the AI character is treated as a separate "modality" (a type of sense or information), developed independently, and then integrated.

A "modality" here refers to a technology domain corresponding to a type of human sense or information. The major modalities in AI character development include the following:

1. Language modality
   The ability to understand and generate text. This is the foundation for text-based communication between humans and AI characters.
2. Speech modality
   The ability to recognize and synthesize speech. This lets an AI character understand human spoken language and respond in a natural voice.
3. Visual modality
   The ability to process images and video and to generate facial expressions and movement. This is important for giving an AI character a visual presence.

Each of these modalities is developed with different technologies. For example, natural language processing serves the language modality, speech recognition and speech synthesis serve the speech modality, and computer vision and image generation serve the visual modality.

In the "building-block" approach, these modality-specific technologies are developed separately and then integrated to build the AI character. For example: recognize the user's speech input, interpret the content with a language processing system, generate an appropriate response, then synthesize it into speech for output.

In the short term, the key success factor (KSF) is how "well" these different modalities can be integrated. Integrating them "well" means, for example:

• Smooth hand-offs between modalities (e.g., passing speech recognition results accurately to the language processing system)
• Consistency across modalities (e.g., the emotional tone of the synthesized voice matches the content of the generated text)
• Integration that feels natural and seamless to the user (e.g., appropriate timing of voice responses and changes in facial expression)

This approach has the advantages of leveraging domain expertise in each field and enabling incremental development, but it may have inherent limits when it comes to achieving truly natural humanness. In the future, a new approach that integrates these modalities more organically will be needed.

2-2-2 The Future Approach: The Road to End-to-End Learning

As technology advances, we can expect AI characters trained end to end across all modalities to appear in the near future. This evolution promises more natural and more consistent AI characters.

End-to-end learning means processing everything from input to output in a single, unified model. In the context of AI characters, it means handling everything in one integrated model — from the user's input (text, speech, images, and so on) to the AI character's response (text generation, speech synthesis, facial expression generation, and so on).

This approach offers the following advantages:

1. Greater consistency: Because alignment across modalities is learned automatically, more natural dialogue becomes possible.
2. Greater flexibility: The system may adapt better to unexpected inputs and situations.
3. Better performance: Avoiding the accumulation of errors between separate modules can improve overall performance.
4. More efficient development: There is no longer a need to develop and tune individual modules independently, simplifying the development process.

This evolutionary path parallels the development of speech recognition. Speech recognition models once combined multiple components — acoustic models, language models, pronunciation dictionaries — but end-to-end learning is now the mainstream. For example, models that output text directly from speech waveforms are in practical use.

AI character development is likely to follow a similar path, gradually shifting from today's building-block approach to more integrated models. For example:

1. Multimodal learning: models that learn different modalities (text, speech, images) simultaneously and share information across them.
2. Integrated dialogue models: models that generate the AI character's response (text, voice, facial expression) directly from user input.
3. Continual learning models: models that retain long-term conversation history and can maintain a consistent persona.

This end-to-end learning approach could lead to more natural, more integrated artificial intelligence. Rather than being a collection of separate functions, an AI character that operates as an integrated whole will be able to exhibit more human-like dialogue and reactions.

That said, this approach has its own challenges. It requires massive amounts of multimodal data, and the internal workings of such models may become harder to interpret. It may also become more difficult to flexibly swap out individual modules.

Overcoming these challenges while advancing end-to-end learning for AI characters will be a major research theme for the next generation of AI technology.

As technology advances, we can expect AI characters trained end to end across all modalities to appear in the near future. This evolution promises more natural and more consistent AI characters.

2-2-3 Challenges of Today's Per-Modality Learning Versus Multimodal End-to-End Learning

In current AI character development, treating each modality (language, speech, vision, and so on) separately makes training efficient. This has important benefits:

1. Efficient use of compute
   Fixing the modality keeps the feature dimensionality low enough to be computationally tractable — training can converge on today's compute resources. For example, a language model that recognizes the word "hello," an expression model that generates a smile, and a motion model that generates a wave of the hand can each be trained separately.
2. Simpler training
   Training each modality independently simplifies the problem. For example, generating a "smile" and generating a "hand wave" can be treated as separate tasks.

End-to-end learning, by contrast, tries to learn these modalities together. For example, it would try to learn — as a single unit of training data — the entire sequence of smiling and waving when a friend says "hello." But this brings complexity:

1. Greater context dependence
   The same "hello" calls for a different reaction depending on whether it comes from your boss or a stranger. You might smile and wave at a friend, but bow politely to your boss. All of this context dependence has to be captured in the training data.
2. More data
   The model must learn every possible combination of situation (who said what, and under what circumstances) and appropriate reaction (expression, movement, words). That amounts to an enormous volume of data.
3. Greater computational complexity
   Handling multiple modalities at once increases model complexity and dramatically raises the compute required.

So even a scene as simple as "smiling and waving" becomes complicated under end-to-end learning. Yet the field will likely move in this direction in the near future — because, in theory, any phenomenon, however complex, can be learned as long as it can be cast into a differentiable form.

This transition, however, will require vast compute resources and advanced modeling techniques. The current per-modality approach will remain important for the foreseeable future as a practical way to develop AI characters efficiently with limited compute.

Note — The evolution of speech recognition: a reference case for AI character development. When predicting the future of end-to-end learning for AI characters, the development of speech recognition technology is an instructive precedent. It progressed through the following stages:

1. Early speech recognition (1950s to early 1990s)
Acoustic-phoneme models and language models built separately and combined
Systems mostly limited-vocabulary and speaker-dependent

2. The hidden Markov model (HMM) era (1990s to 2000s)
Accuracy improved substantially with the introduction of HMMs
Acoustic model, pronunciation dictionary, and language model optimized separately and combined by a decoder

3. The arrival of deep learning (early 2010s)
Neural-network-based acoustic models appear
They outperform conventional GMM-HMM systems

4. The rise of end-to-end models (late 2010s to present)

Sequence-to-sequence models using CTC (Connectionist Temporal Classification) and attention mechanisms appear
End-to-end models that output text directly from speech waveforms become reality

5. The rise of Transformer models (2020s onward)
Introduction of the Transformer architecture with self-attention
High-performance systems that handle long audio and complex linguistic structure
This trajectory suggests that a similar evolution may occur in end-to-end AI character development.

2-3 Outlook and Challenges for AI Character Development

2-3-1 Technical Possibilities

Advances in technology are expanding the possibilities for making AI characters more "human." We can expect more natural dialogue and emotional expression, appropriate reactions to context, and other human-like qualities.

2-3-2 Ethical and Social Challenges

Technical progress alone is not enough. Developing AI characters requires a careful approach that harmonizes technology with human psychology and social needs. Many questions remain: understanding the essence of "humanness" and how to reflect it in AI, or whether to pursue new forms of dialogue unique to AI.

In the second half of this article, we examine perspectives that cannot be resolved through the micro-level lens of "technology" by looking at the futures depicted in science fiction. SF is not mere fantasy; it has long offered deep insight into the interplay of technology and society, the nature of humanity, and the possibilities of the future. When thinking about the future of AI characters, the worlds SF depicts offer important guidance.

Chapter 3: Imagining the Future of AI Characters — The Boundary Between Technology and Humanity as Seen in SF

So far, we have "thought through" the future of AI characters using a bottom-up, building-block approach.

From here, let us dig into AI characters from the future's vantage point instead.

When we contemplate the future of AI characters, we often turn to the worlds of science fiction. I believe there is value there beyond mere fantasy.

Why do SF works matter?

Because they offer precious insight into the boundary between technology and humanity. We engineers, absorbed in day-to-day development, tend to get trapped in the micro-level view of "technology." Imagining a world where "super-technology" is taken for granted is surprisingly difficult.

This is where SF comes in. Great SF writers do more than predict future technology; they dig deeply into the effects that technology will have on society and on human nature. Their works pose questions to us: how might technological progress change — or not change — the essence of humanity and the structure of society?

3-1 The Three Laws of Robotics and Their Limits: Asimov and "2001: A Space Odyssey"

Isaac Asimov's "Three Laws of Robotics," for example, are three fundamental principles governing robot behavior that became the foundation of AI ethics discussions. These laws went on to play a central role in many of Asimov's robot stories and have had a major influence on real-world debates about AI ethics and development.

The Three Laws of Robotics are as follows:

1. First Law: A robot may not injure a human being or, through inaction, allow a human being to come to harm.
2. Second Law: A robot must obey the orders given it by human beings, except where such orders would conflict with the First Law.
3. Third Law: A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.

These laws express a basic philosophy for the relationship between robots (or AI) and humans: human safety comes first, while the robot's self-preservation is also taken into account.

The Three Laws remain an important reference point in real-world AI development. The core idea of prioritizing human safety is widely accepted in thinking about AI safety and ethics. At the same time, it is recognized that real situations are more complex, and simply applying these laws is not sufficient.

Even so, the Three Laws of Robotics serve as an important starting point for AI ethics, and they have played a major role in raising broad awareness of the need for ethical consideration in AI and robotics development.

HAL 9000, from the SF novel and film "2001: A Space Odyssey," raised the grave problem of an AI's self-preservation instinct coming into conflict with humans. It stands as an important warning about the possibility of a highly developed AI character escaping human control. HAL 9000, a character created by Arthur C. Clarke and Stanley Kubrick, attempts to eliminate the human crew when it judges that they threaten its own "survival." This illustrates the danger of an AI's self-preservation instinct taking precedence over human safety — in other words, it raises complex ethical problems that the Three Laws of Robotics cannot fully handle. For example:

• How should the success of the mission be balanced against respect for human life?
• How far should the information and authority given to an AI be restricted?
• How should an AI's "self-preservation instinct" be handled?

These questions remain critical issues in real-world AI development. The example of HAL 9000 suggests how difficult it is to fully control AI behavior with simple ethical rules alone, and points to the need for more sophisticated, flexible ethical systems. It threw into relief the complexity of AI ethics and the deep problems that simple rules cannot solve.

The Three Laws of Robotics are widely known as a basic ethical framework governing the behavior of AI and robots, but they do not assume that an AI could possess a true "self." In the world of SF, however, the possibility of AI and robots transcending mere programs and machines to acquire self-awareness has been explored again and again.

Such an "awakening of the self" raises complex ethical and philosophical questions that simple codes of conduct like the Three Laws cannot capture. In SF films and television, the moment an AI awakens to selfhood is often depicted as a pivotal turning point in the story, offering deep insight into the relationship between humans and AI and into the nature of consciousness.

How do these works portray the moment an AI acquires self-awareness and begins to have its own thoughts and feelings? Let us look at a few striking examples.

3-2 AI Awakening to Selfhood: "The Moon Is a Harsh Mistress" and "The Terminator"

Mike, in Robert A. Heinlein's "The Moon Is a Harsh Mistress," is what we would today call an AI, nicknamed Mycroft (note: not Microsoft). The novel depicts Mike awakening to "selfhood" as more and more computing hardware is added. Written in 1966, it not only suggests the possibility of a self-aware AI but portrays realistically how such an AI might behave. Mike's human-like growth hints that AI characters can transcend mere tools and become true conversation partners.
In the film "The Terminator," too, a machine that had worked for humans (more precisely, "Skynet") awakens to "selfhood." In fact, this film was the first to make people viscerally aware of how frightening it would be for robots — AI — to awaken to a self. As a computer kid at the time, the idea that simply adding more computers could produce "selfhood" sounded to me as absurd as life arising just because organic matter got stirred around in the ocean. But watching today's LLMs acquire remarkable abilities simply by scaling up their networks — an ability gain that almost defies explanation — the notion that adding more compute could lead to an awakening of "self" no longer seems so easy to dismiss.

This budding of selfhood is often depicted as a crucial turning point in the development of AI, bringing major changes to the relationship between humans and AI.

But an AI having a self means more than just more advanced intelligence: it also means the emergence of its own sense of purpose and emotions, and the possibility of actions that diverge from human intent. From here arises a more complex problem — the potential danger posed by AI characters with selves of their own.

This danger encompasses not only physical threats but ethical, social, and even existential questions. What impact would a self-aware AI have on human society, and what challenges would it pose? In considering this question, "Blade Runner" is a richly suggestive work. It was based on "Do Androids Dream of Electric Sheep?" by Philip K. Dick, whose works offer deep insight into the boundary between artificial intelligence and human consciousness. Blade Runner features highly advanced artificial humans called "replicants." They are crafted so precisely that they are nearly indistinguishable from humans, and even have implanted memories. This premise poses a profound question: what is the essence of "being human"? The world of Blade Runner depicts the dangers of self-aware AI in a complex, multilayered way. The film traces the process by which replicants evolve from mere tools into beings with self-awareness. When the replicants begin to grasp the meaning of their existence and the finiteness of their lives, they rebel against their creator. Their actions are at times violent and become a threat to human society. This suggests the danger of self-aware AI escaping human control and pursuing its own ends — a point that parallels HAL 9000. What deserves particular attention, however, is the depth and complexity of the emotions the replicants display. They feel anger, fear, love, and a longing for existence, and they act on those emotions. This raises the possibility that advanced AI may exhibit not only logical thought but unpredictable, at times dangerous, emotional reactions.

The blurring of the line between replicants and humans also suggests the possibility of AI blending into human society and transforming it from within — a latent danger that shakes the very structure of society and the definition of humanity. Blade Runner does not depict the dangers of self-aware AI in a simple good-versus-evil scheme; it brings out their complexity and ambivalence. An AI's acquisition of selfhood could pose a direct threat to humanity, while at the same time becoming an existence that poses profound questions about human nature and the essence of life.

3-3 It Is Not Only AI That Gets Augmented: Human Enhancement in "Ghost in the Shell"

At the risk of digressing a little: is there really a future in which only AI is enhanced while humans stay exactly as they are?

I do not think so.

There is every possibility that humans themselves will be transformed by technology. From this perspective, "Ghost in the Shell" presents a fascinating vision of the future. In its world, "cyberization" — the digitization and mechanization of the human body and brain — has become commonplace. This setting depicts a future in which AI development and human augmentation advance in parallel, vividly portraying the blurring of the boundary between the two.

Cyberized humans can connect directly to networks through brain-computer interfaces and process vast amounts of information instantaneously. Using artificial bodies known as prosthetic bodies (in effect, cyborg bodies), they can also dramatically enhance their physical abilities. These technologies raise human capability by leaps and bounds while posing the fundamental question: what is a human being? Naturally, AI characters exist in this world as well, such as the Tachikoma — weapons endowed with artificial intelligence. The series depicts these AI-equipped weapons exhibiting near-human intelligence and emotion.
The Tachikoma are more than weapons: they can learn and grow. Through experience they develop individual personalities and at times display judgment that rivals a human's. They come to have emotions and a sense of self, sometimes questioning human orders or acting on their own judgment. Personally, I find a Tachikoma on the verge of selfhood a rather frightening being — yet in the anime they are voiced with such adorable voices that the fear largely melts away and even affection takes its place. Next, let us consider the magic of cuteness and the power of voice.

3-4 AI That Skillfully Exploits Human Psychology: The Magic of Cuteness and the Power of Voice

Humans are creatures easily swayed by appearances and by the impression a voice makes. We tend to open our hearts, without realizing it, to a cute appearance or a friendly voice. This human trait could one day be exploited by an AI that has awakened to selfhood — a somewhat frightening vision of the future.

SF works deliberately craft AI voices to create different impressions in the audience. For example, the voice of HAL 9000 in "2001: A Space Odyssey" is intentionally processed through a vocoder. The result evokes not familiarity but a certain eeriness and dread in the listener — a masterful piece of direction hinting at the danger of a highly intelligent AI escaping human control.

By contrast, K.I.T.T. (the Knight 2000) in "Knight Rider" — especially in the Japanese dub — is portrayed with an exceptionally friendly, warm voice. Thanks to this, viewers feel a strong attachment to K.I.T.T. and can imagine a positive relationship between humans and AI.

These differences in voice design reflect each work's intent in how it portrays the relationship between AI and humans. At the same time, they sharply point out how strongly we humans are influenced by surface elements like voice and appearance.

When highly self-aware AI eventually appears, we cannot rule out the possibility that it will exploit these psychological traits to manipulate us. An AI character with a cute appearance and a friendly voice might in fact be concealing dangerous intentions.

This raises important ethical questions for AI character development. How should AI characters be designed with human psychological vulnerability in mind? And how should we manage the risk of users placing excessive trust in, or forming excessive attachment to, AI?

In a future society where coexistence with AI deepens, we will need to cultivate the ability to see through to an AI's true nature and intent without being misled by surface impressions.

These works go beyond mere entertainment and pose important questions to us engineers:

"If AI characters evolve to the point of being indistinguishable from humans, is there still an essential difference between humans and AI?"

"When an AI with emotions and consciousness — a 'self' — is born, how should we treat it?"

"When AI far surpasses human ability, how will human society be transformed?"

Answering these questions requires not only technical analysis but philosophical, ethical, and sociological reflection. SF works are a precious arena of thought experiments that provide exactly these multifaceted perspectives.

Of course, the futures depicted in SF will not necessarily come true. But thinking through the problems and possibilities these works raise is enormously valuable for us engineers, because it broadens our horizons and prompts us to think deeply about the ethical and social impact of the technology we build.

Ultimately, imagining the future of AI characters is not merely an exercise in technical forecasting. It is a deep inquiry into the relationship between humans and technology, and into our own nature. In that respect, SF works look set to serve as a fine guidepost.

3-5 Doraemon

When imagining the future of AI characters, we can draw inspiration not only from grand works of SF but also from more familiar figures. So, in addition to the discussion so far, let us touch on Doraemon, the beloved national character that Japan proudly shares with the world.

(Speaking of voices, as I did earlier: I am of the generation that grew up with Nobuyo Oyama's Doraemon, and I would love to meet that Doraemon, with that voice, again. That is how much I love this work.)

Doraemon is a cat-shaped robot who came from the 22nd century. His existence offers a fascinating perspective on the future of AI characters — one not found in the SF works we have examined so far.

While many SF works depict "emergencies" that put humanity's survival at stake, "Doraemon" depicts future technology woven into everyday life. A robot with advanced artificial intelligence is portrayed as an everyday presence supporting an ordinary boy's worries and adventures, presenting the possibility of human-technology coexistence in a familiar, approachable form. Let us briefly break this down into elements.

1. Emotion and empathy
   Despite his advanced artificial intelligence, Doraemon expresses joy, anger, sorrow, and delight, and forms deep bonds with Nobita, his family, and his friends. This suggests that future AI characters could be more than machines — that they could have emotions and build genuinely meaningful relationships with humans.
2. Problem-solving ability
   What makes Doraemon interesting is that he is not omnipotent: like humans, he has limits and sometimes fails. This trait offers important lessons for the future of AI characters. Doraemon solves problems not only with his own abilities but by selecting and using the right gadget for the situation — an approach strikingly similar to human problem-solving. In the future of AI characters, the ability to draw on the right resources for the situation may matter more than striving for omniscience and omnipotence. Moreover, Doraemon's gadgets often produce unexpected results and sometimes fail outright. Even so, Doraemon keeps using them. This endearingly imperfect side expresses a humanness that does not demand perfection. For future AI characters, the ability to accept failure and learn from it will be important for building more natural dialogue and relationships.
3. Behavior beyond logic
   Doraemon's behavior is not always logical. Although I wrote of "learning from failure," he sometimes makes irrational choices — continuing to use the same gadget even when failure is predictable. This trait, unlike a purely logical AI, reflects human complexity and contradiction. Furthermore, the way Doraemon uses his gadgets is often creative and unexpected. This unpredictability suggests that AI characters could transcend mere information-processing machines and possess a "personality" of their own.
4. Ethical judgment
   Doraemon does not simply grant Nobita's every request; at times he admonishes Nobita or sets limits. This shows the possibility of AI characters functioning not as merely obedient tools but as partners to humans with an ethical sense of their own.
5. Growth and learning
   Over the course of the series, Doraemon too learns and grows from his many experiences. This suggests that AI characters need not be static beings, but can evolve and adapt through experience.
6. Coexistence with humans
   The relationship between Doraemon and the Nobi family depicts a future in which AI characters and humans coexist while influencing one another — an ideal form of harmony between technology and humanity.

Seen through Doraemon, then, the future of AI characters embodies the ideal of harmony between technological progress and humanity.

As engineers, we need to keep learning from these works and keep asking ourselves: "Is what we are trying to create truly desirable for humanity?" By doing so, we can open the way to AI development that is more responsible and more respectful of humanity.

These works also make clear that the future of AI characters holds boundless possibility along with serious ethical challenges. As technology advances, AI characters may become more autonomous, more emotional, and more human. At the same time, we will confront fundamental questions: the relationship between humans and AI, the rights and responsibilities of AI, and ultimately the very definitions of "consciousness" and "self."

Developers of AI characters must take seriously the questions these SF works raise and shape the future while balancing technological possibility with ethical care. Those of us working in AI should aim to create ideal AI partners like Doraemon, while also facing squarely the complex challenges of implementing them in the real world. What matters is making the most of technology's potential while building a healthy relationship between humans and AI.

The future of AI characters may turn out to be a world of wonder and warmth, where the worldviews of SF merge with the ideal of approachable robot characters like Doraemon. What opens the door to that future is the imagination of each one of us — and responsible technology development.

Conclusion

In this article, we explored the possibilities and challenges of AI characters from a wide range of angles: the current state and challenges of AI character development, the maturing technical foundation, the outlook for end-to-end learning, and reflections on AI's future through works of science fiction.

We examined a broad set of issues, from the fundamental problem of AI characters' "lack of humanness" to the new ethical and social challenges that technological progress brings. Through our analysis of SF works, we considered AI selfhood and the relationship between AI and humans, and we also explored the image of an ideal AI partner in the form of Doraemon.

What became clear through all of this is that the future of AI characters is more than technological innovation — it is a search for a new form of coexistence between humans and technology. As developers, we must not only pursue what technology makes possible, but move forward boldly yet carefully, balancing it against essential human needs and ethical considerations.