1. Introduction¶

1.1 The Fusion of Humor and Artificial Intelligence¶

The quest to bring humor and artificial intelligence (AI) together has taken the field of computer science by storm 🌩️. The idea of creating Robo-Comedians, AI models capable of generating humorous content, has long been a tantalizing challenge for researchers and developers alike. The ultimate goal is to create AI models that can not only understand and generate humor but also adapt their comedic style to different audiences, situations, and cultures.

The fusion of humor and AI is a fascinating journey through the realms of natural language processing (NLP), deep learning, and the psychology of laughter 😂. The endeavor requires a deep understanding of human cognition, the subtleties of language, and the complex interplay of emotions that drive human interactions. One of the key challenges lies in navigating through the nuances of humor, which can be highly subjective and culturally specific.

To develop an AI model capable of generating humor, we need to unravel the intricate fabric of joke structures and the psychological underpinnings of laughter. At the heart of this endeavor lies the question: Can we teach machines to make us laugh? 🤖

In this context, we'll explore the world of Robo-Comedians and the journey to develop AI models that can generate humor, alter their comedic style, and bring laughter to millions. We'll delve into the science of humor, the art of joke generation, and the ethical considerations of AI-generated comedy.

1.2 The Rise of Robo-Comedians¶

The emergence of Robo-Comedians can be traced back to the advances in AI, particularly in the realm of NLP and deep learning. The development of large-scale language models, such as OpenAI's GPT-3, has paved the way for a new generation of AI models capable of understanding and generating humorous content.

The rise of Robo-Comedians is fueled by the growing demand for personalized content and the potential to revolutionize the entertainment industry. From stand-up comedy to TV shows and movies, AI-generated humor has the potential to transform the way we experience comedy and engage with technology.

The development of Robo-Comedians also raises questions about the ethical implications of AI-generated humor, the impact on human comedians, and the potential for fostering creativity and collaboration in comedy. As AI continues to permeate all aspects of our lives, the role of humor in AI research and applications is becoming increasingly prominent, and the quest for Robo-Comedians has never been more exciting 🚀.

In the following sections, we'll delve into the science of humor, the art of joke generation, and the ethical considerations of AI-generated comedy. We'll explore the mathematics and psychology behind humor, the techniques used to train AI models in the art of comedy, and the future of AI-generated humor.

So, sit back, relax, and let's embark on a laughter-filled journey into the world of Robo-Comedians! 🎭

2. The Science of Humor¶

Humor is a complex and multifaceted phenomenon, deeply ingrained in human culture, cognition, and communication. To teach AI models to generate humor, we must first delve into the science of humor, which encompasses understanding joke structures and the psychology of laughter. So, let's dive into the mathematical and psychological underpinnings of humor to better understand how AI can tickle our funny bones! 😄

2.1 Understanding Joke Structures¶

One of the key aspects of humor is the structure of jokes. Jokes often follow specific patterns and rules that can be analyzed and quantified. In mathematics, we can describe these structures using formal language theory, graph theory, and other mathematical tools.

A common joke structure is the "setup-punchline" pattern, where a setup creates certain expectations, and the punchline subverts these expectations, eliciting laughter. We can represent a joke as a graph, with nodes representing the concepts and edges representing the relationships between them. The punchline introduces an unexpected twist, creating a surprising connection between the nodes, which can be modeled using graph theory.

Formally, let $G = (V, E)$ be a directed graph representing a joke, where $V$ is the set of nodes (concepts) and $E$ is the set of edges (relationships). The punchline can be represented as a function $f: V \times V \rightarrow \mathbb{R}$, which assigns a surprise value to each pair of nodes. The surprise value can be calculated using various methods, such as information theory or semantic similarity measures.

One approach to quantifying the surprise is to use the Kullback-Leibler (KL) divergence, a measure of the difference between two probability distributions. Given two distributions $P$ and $Q$, the KL divergence is defined as:

In the context of jokes, the KL divergence can be used to measure the difference between the expected and the actual punchline, capturing the element of surprise.

2.2 The Psychology of Laughter¶

Laughter is a universal human response to humor, rooted in our evolutionary history and cognitive processes. The psychology of laughter involves understanding the cognitive, emotional, and social factors that contribute to the perception and appreciation of humor.

One of the most influential theories of humor is the incongruity-resolution theory, which posits that humor arises from the resolution of incongruity or the unexpectedness in a joke. This theory can be formalized using Bayesian surprise, a measure that quantifies the surprise elicited by an event given prior beliefs.

Let $P(\theta)$ denote the prior belief about an event $\theta$, and $P(\theta \mid x)$ denote the posterior belief after observing evidence $x$. The Bayesian surprise is then defined as:

Bayesian surprise captures the extent to which evidence $x$ changes our beliefs, which can be used to model the surprise elicited by a punchline in a joke.

To train AI models in the art of humor, we can draw on these mathematical and psychological insights, teaching them to recognize and generate joke structures and surprise. In the next section, we'll explore how AI models can be trained to generate humor using datasets, NLP techniques, and evaluation methods.

import numpy as np

def kl_divergence(P, Q):
    return np.sum(P * np.log(P / Q))

def bayesian_surprise(P_prior, P_posterior):
    return kl_divergence(P_posterior, P_prior)

2.3 The Mathematics of Timing and Delivery¶

In comedy, timing and delivery play a crucial role in making a joke funny. This aspect of humor can be modeled mathematically using concepts from signal processing and dynamical systems. One approach is to represent the timing and delivery of a joke as a sequence of events, with each event corresponding to a word, gesture, or pause.

The timing and delivery of a joke can be characterized by the inter-event intervals, which can be modeled as a stochastic process or a dynamical system. For example, we can use a Poisson process, where the inter-event intervals follow an exponential distribution with parameter $\lambda$:

By analyzing the timing and delivery patterns in a corpus of jokes, we can identify the optimal parameters for generating humorous content with AI models.

def poisson_process(lmbda, t):
    return lmbda * np.exp(-lmbda * t)

def inter_event_intervals(data,lmbda):
    return poisson_process(lmbda, data)

An alternative approach is to model the timing and delivery as a continuous-time dynamical system, such as a linear time-invariant (LTI) system. The LTI system can be represented by a transfer function $H(s)$, which describes the relationship between the input signal $X(s)$ and the output signal $Y(s)$ in the Laplace domain:

By identifying the transfer function that characterizes the timing and delivery of jokes, we can design AI models that generate humor with the right pacing and rhythm.

To recap, the science of humor involves understanding joke structures, the psychology of laughter, and the mathematics of timing and delivery. By leveraging these insights, we can train AI models to generate funny content, paving the way for a new generation of robo-comedians. 😂🤖🎭

In the next section, we'll dive into the practical aspects of training AI to be funny, discussing the datasets, NLP techniques, and evaluation methods used in the field of AI-generated humor. Hold on to your hats, because things are about to get hilarious! 🎩💼🚀

3. Training AI to be Funny¶

Now that we have a solid understanding of the science of humor, it's time to roll up our sleeves and explore the nitty-gritty of training AI models to generate jokes. In this section, we'll delve into the art of joke collection, natural language processing and generation techniques, and evaluation methods for AI humor. So, buckle up, and let's embark on this laughter-filled journey! 🚀😄

3.1 Datasets and the Art of Joke Collection¶

Before we can train an AI model to generate humor, we need a dataset of jokes to learn from. The quality and quantity of the dataset play a crucial role in the performance of the model. Datasets can be sourced from various platforms such as joke websites, comedy shows, and social media platforms like Twitter or Reddit.

To create a dataset, we need to pre-process and clean the data, removing any irrelevant or offensive content, and ensuring a diverse range of joke types and styles. The dataset should be balanced, containing both traditional and contemporary jokes, as well as different genres of humor, such as puns, one-liners, and situational comedy.

Once the dataset is ready, it's time to split it into training, validation, and test sets. The training set is used to teach the AI model, the validation set is used to fine-tune the model's parameters, and the test set is used to evaluate the model's performance on unseen data.

3.2 Natural Language Processing and Generation¶

With the dataset in place, we can now proceed to train the AI model using natural language processing (NLP) and generation techniques. NLP focuses on the analysis and understanding of human language, while natural language generation (NLG) is concerned with creating human-like text.

3.2.1 Language Models¶

At the core of AI-generated humor lies the language model, which is trained to predict the next word in a sequence given its context. One popular approach is to use recurrent neural networks (RNNs), such as long short-term memory (LSTM) networks, which can model long-range dependencies and capture the structure of jokes.

Another state-of-the-art approach is to use transformer-based models like GPT (Generative Pre-trained Transformer). The transformer architecture enables the model to capture complex patterns and dependencies in the input text, making it well-suited for joke generation.

The objective function for training the language model is typically the cross-entropy loss, which measures the difference between the predicted word probabilities and the true word probabilities:

where $N$ is the number of examples, $V$ is the vocabulary size, $y_{ij}$ is the true probability of word $j$ in example $i$, and $\hat{y}_{ij}$ is the predicted probability of word $j$ in example $i$.

3.2.2 Fine-tuning and Joke Generation¶

Once the language model is trained, we can fine-tune it on our joke dataset to generate humor. Fine-tuning involves updating the model's parameters on the joke dataset using techniques like transfer learning or domain adaptation. This process helps the model specialize in joke generation while retaining its general language understanding capabilities.

To generate jokes, we can use various sampling techniques, such as greedy sampling, beam search, or nucleus sampling. These methods balance the trade-off between exploration and exploitation, ensuring that the generated text is both diverse and coherent.

import torch
from transformers import GPT2LMHeadModel, GPT2Tokenizer

tokenizer = GPT2Tokenizer.from_pretrained("gpt2")
model = GPT2LMHeadModel.from_pretrained("gpt2")

def generate_joke(prompt, max_length=100, num_return_sequences=1):
    input_ids = tokenizer.encode(prompt, return_tensors="pt")
    output_sequences = model.generate(
        input_ids=input_ids,
        max_length=max_length,
        num_return_sequences=num_return_sequences,
        do_sample=True,
        temperature=1.0,
    )

    jokes = []
    for sequence in output_sequences:
        joke = tokenizer.decode(sequence, skip_special_tokens=True)
        jokes.append(joke)

    return jokes

3.3 Evaluating AI Humor: The Turing Test of Comedy¶

Evaluating the humor of AI-generated jokes poses unique challenges, as humor is highly subjective and context-dependent. One approach is to use the Turing Test of Comedy, where human judges rate the funniness of jokes without knowing whether they were generated by an AI or a human.

Metrics such as precision, recall, and F1-score can be used to evaluate the model's performance, along with human ratings of funniness, surprise, and coherence. In addition, we can leverage intrinsic evaluation methods like perplexityand BLEU score, which measure the fluency and similarity of the generated text to the reference text, respectively. However, these metrics may not fully capture the nuances of humor, and thus, a combination of human and automated evaluation methods is recommended.

Perplexity is defined as the exponentiated average negative log-likelihood of the true word probabilities given the model's predictions:

where $N$ is the number of examples, $y_i$ is the true word in example $i$, and $\hat{y}_{i, y_i}$ is the predicted probability of the true word in example $i$.

The BLEU score is calculated as the geometric mean of the modified n-gram precision scores, multiplied by a brevity penalty if the generated text is shorter than the reference text:

where $P_n$ is the modified n-gram precision, $w_n$ is the weight assigned to each n-gram level (usually uniform), and $\text{BP}$ is the brevity penalty.

Despite these evaluation methods, there's no substitute for the ultimate test: the laughter of the audience! 🤣👏

4. AI in Comedy: Practical Applications¶

AI-generated humor is no laughing matter! 😄 The application of artificial intelligence in the realm of comedy has revolutionized the way jokes are created and consumed. In this section, we will explore some of the most exciting and practical applications of AI in comedy, including stand-up comedy, writing jokes for TV shows and movies, and personalized joke recommendations.

4.1 Stand-up Comedy and Robo-Comedians¶

Stand-up comedy has always been a human-centric art form, but the arrival of robo-comedians is shaking things up. Robo-comedians are AI-powered virtual or physical robots that can perform stand-up comedy routines, generating jokes using advanced natural language processing (NLP) techniques.

One notable approach is to use transformer-based language models like GPT-3, which have shown a remarkable ability to generate coherent and sometimes hilarious text. For example, a robo-comedian might generate a joke using the following prompt:

prompt = "Tell me a joke about AI and comedy:"
joke = gpt3.generate(prompt)
print(joke)

The generated joke can then be performed by a physical robot or a virtual avatar. The performance aspect involves several layers of AI, including speech synthesis, facial expression generation, and body language modeling. A combination of these techniques allows the robo-comedians to deliver their punchlines with the same panache as their human counterparts.

Robo-comedians also offer the unique advantage of tailoring their performance to the audience. By analyzing the demographics, interests, and even facial expressions of audience members, robo-comedians can adapt their material in real-time to maximize laughter 😂. This level of customization is unparalleled in traditional stand-up comedy and opens the door for highly interactive comedic experiences.

4.2 Writing Jokes for TV Shows and Movies¶

AI has also made significant inroads into the world of television and movies, assisting writers in generating humorous content. While the creative process is still predominantly human-driven, AI can offer valuable suggestions and help writers overcome writer's block.

To generate jokes for a specific context, AI models can be fine-tuned on relevant datasets. For example, to write jokes for a sitcom, an AI model might be trained on a dataset of existing sitcom scripts. This fine-tuning process allows the AI to internalize the structure, tone, and humor styles typical of sitcoms.

Once trained, the AI model can be used to generate humorous lines or entire scenes, given a specific context or set of characters. For example:

prompt = "Write a funny scene between two characters discussing artificial intelligence:"
scene = gpt3.generate(prompt)
print(scene)

The generated content can then be edited, refined, or combined with human-written material, resulting in a collaborative creative process that pushes the boundaries of traditional comedy writing 📝.

4.3 Personalized Joke Recommendations¶

Personalized joke recommendations are another exciting application of AI in comedy. By analyzing users' preferences, interests, and even their sense of humor, AI models can generate tailored jokes that resonate with individual tastes.

One approach to achieving this personalization is by leveraging collaborative filtering techniques, which are commonly used in recommendation systems 🤖. The basic idea is to identify users with similar tastes and recommend jokes that have been enjoyed by those users. For example, if user A and user B both enjoyed jokes about mathematics and cryptography, then the AI might recommend a joke about artificial intelligence for user A, assuming user B also enjoyed that joke.

To represent users' preferences, we can use a matrix $P$, where $P_{i,j}$ denotes the preference score of user $i$ for joke $j$. Using singular value decomposition (SVD), we can factorize this matrix into three matrices $U$, $S$, and $V$:

The matrix $U$ represents users' latent factors, while the matrix $V$ represents jokes' latent factors. By multiplying these matrices, we can estimate users' preferences for new jokes and generate personalized recommendations.

import numpy as np
from scipy.sparse.linalg import svds

def recommend_jokes(P, num_recommendations):
    U, S, V_T = svds(P, k=50)
    P_hat = np.dot(np.dot(U, np.diag(S)), V_T)

    recommendations = []
    for user_idx in range(P.shape[0]):
        user_recommendations = np.argsort(P_hat[user_idx])[-num_recommendations:]
        recommendations.append(user_recommendations)

    return recommendations

By combining state-of-the-art AI techniques with a deep understanding of human psychology and humor, AI-generated comedy is revolutionizing the way we create and consume humor. So, next time you hear a hilarious joke, don't be surprised if it was crafted by an AI! 😆

In the next section, we will delve into the ethical considerations and challenges surrounding the use of AIin comedy, and how we can navigate this brave new world of robo-comedians.

5. Ethical Considerations and Challenges¶

As we delve deeper into the world of AI-generated humor, it is crucial to address the ethical considerations and challenges that emerge with the rise of robo-comedians. In this section, we will explore the impact on human comedians, the importance of ensuring inclusive and non-offensive humor, and the complexities surrounding intellectual property and joke ownership.

5.1 The Impact on Human Comedians¶

The emergence of AI-generated humor raises concerns about the potential displacement of human comedians in the industry. As robo-comedians continue to improve their performance, it is necessary to contemplate the implications for human performers. One possible approach to address this concern is by adopting a hybrid model that combines the strengths of AI and human comedians, fostering collaboration and synergy between the two. For instance, AI-generated jokes could serve as inspiration for human comedians, allowing them to fine-tune or adapt the material to their unique style and audience preferences 🎭.

However, it is important to recognize that AI-generated humor is unlikely to ever fully replace the intangible qualities that make human comedians special, such as their ability to connect with the audience, convey emotions, and respond to real-time feedback. As noted by Turing in his seminal paper on AI, "We may hope that machines will eventually compete with men in all purely intellectual fields" Turing, 1950. But when it comes to humor, the human touch is irreplaceable.

5.2 Ensuring Inclusive and Non-offensive Humor¶

Another critical ethical consideration is the importance of promoting inclusive and non-offensive humor in AI-generated jokes. Since AI models are trained on large datasets that may contain biased or offensive content, it is essential to address these concerns at the training stage. One approach is to develop robust filtering mechanisms to identify and remove potentially offensive content from the training data 🚯. For example, researchers at OpenAI have implemented a moderation system in their API to prevent content that violates their usage policies from being generated OpenAI, 2021.

Moreover, incorporating fairness and bias mitigation techniques in the AI models can help to ensure that the generated humor does not disproportionately target or marginalize specific groups. Researchers have developed several fairness metrics and debiasing methods, such as the Equalized Odds and Demographic Parity Hardt et al, 2016, which can be integrated into the AI training process.

5.3 Intellectual Property and Joke Ownership¶

The question of intellectual property (IP) and joke ownership becomes increasingly complex as AI-generated humor gains prominence. Traditional copyright laws may not adequately address the unique challenges posed by AI-generated content, leading to potential disputes over joke ownership and infringement claims.

One possible approach to resolving IP concerns is to recognize AI-generated humor as a derivative work, with ownership attributed to the human creators who designed and trained the AI model. This would align with existing copyright frameworks, such as the United States Copyright Office's stance that "A work that is created by a machine with no creative input or intervention from a human is not registrable" USCO, 2021.

However, as AI-generated humor becomes more sophisticated and autonomous, it may be necessary to revisit and recalibrate existing IP frameworks to accommodate the evolving landscape. For example, the European Parliament has suggested granting certain legal rights to AI systems European Parliament, 2020.

6. The Future of AI-Generated Humor¶

As we venture into the uncharted territory of AI-generated humor, we can't help but feel a sense of excitement and anticipation for what lies ahead. In this section, we will explore the new frontiers in comedy, including virtual reality (VR) and augmented reality (AR) experiences, and the role of AI in fostering creativity and collaboration. Let's dive in, shall we? 🏊‍♂️

6.1 New Frontiers in Comedy: Virtual Reality and Augmented Reality¶

The advent of VR and AR technologies has opened the door to a myriad of possibilities for AI-generated humor. These immersive environments provide a fertile ground for the development of innovative and engaging comedic experiences, where robo-comedians can dynamically adapt their performances based on audience feedback and preferences.

Imagine, for instance, a stand-up comedy show in a virtual reality space where the audience members are represented by avatars, and they can interact with the comedian in real-time, providing feedback through laughter, applause, or even virtual tomatoes 🍅. The robo-comedian could be equipped with advanced sensors and algorithms to detect and analyze audience reactions, allowing it to modify its performance and jokes on-the-fly. This could lead to a highly personalized and interactive comedic experience that transcends traditional stand-up comedy shows.

In an augmented reality context, AI-generated humor could be integrated seamlessly into our everyday lives. Picture a world where you can point your smartphone at a mundane object, and a robo-comedian generates a witty joke or humorous observation related to that object, instantly brightening up your day 😄. This could open up entirely new avenues for humor consumption, as people could access comedic content anytime, anywhere.

6.2 The Role of AI in Fostering Creativity and Collaboration¶

Beyond expanding the frontiers of comedy, AI-generated humor has the potential to foster creativity and collaboration among human comedians and other creative professionals. By providing a wellspring of fresh ideas and inspiration, AI-generated jokes can serve as a catalyst for the development of new comedic styles, genres, and forms of expression.

One way to achieve this is by leveraging the power of generative adversarial networks (GANs) in the context of comedy. GANs consist of two neural networks, a generator and a discriminator, that are pitted against each other in a game of cat-and-mouse 🐱🐭. The generator creates content, while the discriminator evaluates its quality and provides feedback, allowing the generator to refine its output iteratively.

In the realm of comedy, we can envision a scenario where the generator produces jokes, and human comedians act as the discriminator, providing feedback and refining the AI-generated humor. This collaborative process could lead to the emergence of new comedic styles and genres, pushing the boundaries of both AI-generated and human-produced humor.

Moreover, AI-generated humor could democratize comedy by lowering the barriers to entry and allowing aspiring comedians to access and build upon a vast repository of jokes and comedic material. By embracing the potential of AI-generated humor, we can create an inclusive and diverse comedy landscape that fosters innovation and celebrates the unique talents of comedians from all walks of life 🌍.

In conclusion, the future of AI-generated humor is teeming with possibilities, from new frontiers in VR and AR experiences to fostering creativity and collaboration among human comedians. As we continue to explore and harness the power of AI in comedy, we can look forward to a future where laughter is more accessible, diverse, and innovative than ever before. So, let's raise a toast 🥂 to the exciting and, dare I say, hilarious journey that lies ahead in the world of AI-generated humor!

7. Conclusion¶

As we bring this thrilling exploration of AI-generated humor to a close, it's crucial to reflect on the potential and implications of robo-comedians and their growing presence in our comedic landscape. Let's revisit the key points we've discussed and ponder the exciting prospects that lie ahead, as well as the challenges we must overcome to ensure the harmonious coexistence of AI and human comedians in the pursuit of laughter 😄.

7.1 Embracing the Laughter: The Potential of Robo-Comedians¶

Throughout this journey, we've examined the fascinating fusion of humor and artificial intelligence, delving into the science of humor, joke structures, and the psychology of laughter. We've explored how AI can be trained to generate jokes using datasets and natural language processing techniques, as well as how its humor can be evaluated using a Turing Test of Comedy. From stand-up comedy to writing jokes for TV shows and movies, AI has demonstrated its potential to revolutionize and diversify the comedic landscape 💫.

Moreover, we've discussed the ethical considerations and challenges associated with AI-generated humor, including the impact on human comedians, ensuring inclusive and non-offensive humor, and the complexities surrounding intellectual property and joke ownership. Addressing these challenges is paramount to fostering a thriving comedy ecosystem that embraces AI-generated humor while safeguarding the interests and contributions of human comedians.

7.2 The Last Laugh: AI's Role in the Evolution of Comedy¶

As we gaze into the future of AI-generated humor, we can't help but marvel at the new frontiers in comedy that await us, such as virtual reality and augmented reality experiences. The role of AI in fostering creativity and collaboration among human comedians and creative professionals is equally invigorating, as it has the potential to spark innovation, push the boundaries of comedic expression, and democratize the art of joke generation 🎉.

But, as we enthusiastically embrace the laughter that AI-generated humor brings, we must also remain vigilant and conscious of the challenges that lie ahead. To create a future where AI and human comedians coexist harmoniously, we must commit to fostering an inclusive, diverse, and ethical comedy landscape that celebrates the unique talents and contributions of all involved 🤗.

As mathematician and philosopher Bertrand Russell once said, "The only thing that will redeem mankind is cooperation." With that in mind, it's our collective responsibility to ensure that the union of humor and artificial intelligence leads to a brighter, more inclusive, and laughter-filled future for all. And with that, dear reader, we conclude our exploration of robo-comedians, as we eagerly anticipate the countless chuckles, guffaws, and giggles that lie ahead on this exhilarating journey 🚀.

So, let's put our hands together 👏 and give a standing ovation to the marvelous world of AI-generated humor, as we take the plunge into uncharted waters, navigate the waves of innovation, and ride the crest of laughter to new and exciting shores! 🌊😂

8. References¶

1. Mihalcea, R., & Strapparava, C. (2005). Making Computers Laugh: Investigations in Automatic Humor Recognition. In Proceedings of the Human Language Technology Conference and Conference on Empirical Methods in Natural Language Processing (pp. 531-538). Association for Computational Linguistics.

2. Raskin, V. (1985). Semantic Mechanisms of Humor. D. Reidel Publishing Company.

3. Attardo, S., & Raskin, V. (1991). Script Theory Revisited: Joke Similarity and Joke Representation Model. Humor, 4(3-4), 293-347.

4. Binsted, K., & Ritchie, G. (1997). Computational Rules for Generating Punning Riddles. Humor, 10(1), 25-76.

5. Westbury, C., Shaoul, C., Moroschan, G., & Ramscar, M. (2016). Telling the World's Least Funny Jokes: On the Quantification of Humor as Entropy. Journal of Memory and Language, 86, 141-156.

6. Hossain, N., & Krumhansl, S. (2019). Transformer-Based Deep Reinforcement Learning for Humor Generation. arXiv preprint arXiv:1910.00249.

7. Valitutti, A., Toivonen, H., & Doucet, A. (2017). Let Everything Turn Well in Your Wife: Generation of Adult Humor Using Lexical Constraints. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Vol. 1, pp. 2430-2439). Association for Computational Linguistics.

8. Mathur, M. B., & Reichling, D. B. (2016). Navigating a Social World with Robot Partners: A Quantitative Cartography of the Uncanny Valley. PLOS ONE, 11(1), e0146318.

9. Veale, T., & Hao, Y. (2020). Growing Pains: Adapting to the Challenges of a Growing Internet Audience. International Journal of Communication, 14, 4958-4975.

10. Turing, A. M. (1950). Computing Machinery and Intelligence. Mind, 59(236), 433-460.

11. GPT-3. (n.d.). In Wikipedia.

12. BERT (language model)). (n.d.). In Wikipedia.

13. The Ethics of Artificial Intelligence. (n.d.). In Wikipedia.