We present a novel multimodal preference dataset for creative tasks, consisting of over 250 million human votes on more than 2.2 million captions, collected through crowdsourcing rating data for The New Yorker's weekly cartoon caption contest over the past eight years. This unique dataset supports the development and evaluation of multimodal large language models and preference-based fine-tuning algorithms for humorous caption generation. We propose novel benchmarks for judging the quality of model-generated captions, utilizing both GPT4 and human judgments to establish ranking-based evaluation strategies. Our experimental results highlight the limitations of current fine-tuning methods, such as RLHF and DPO, when applied to creative tasks. Furthermore, we demonstrate that even state-of-the-art models like GPT4 and Claude currently underperform top human contestants in generating humorous captions.

Such creative software can be used for autonomous creative tasks, such as inventing mathematical theories, writing poems, painting pictures, and composing music. However, computational creativity studies also enable us to understand human creativity and to produce programs for creative people to use, where the software acts as a creative collaborator rather than a mere tool. Historically, it's been difficult for society to come to terms with machines that purport to be intelligent and even more difficult to admit that they might be creative. For instance, in 1934, some professors at the University of Manchester in the United Kingdom built meccano models that were able to solve some mathematical equations. Groundbreaking for its time, this project was written up in a piece in Meccano Magazine.

The titles of the eight symposia were as follows: AI for Gerontechnology (FS-12-01), Artificial Intelligence of Humor (FS-12-02), Discovery Informatics: The Role of AI Research in Innovating Scientific Processes (FS-12-03), Human Control of Bio-Inspired Swarms (FS-12-04), Information Retrieval and Knowledge Discovery in Biomedical Text (FS-12-05), Machine Aggregation of Human Judgment (FS-12-06), Robots Learning Interactively from Human Teachers (FS-12- 07), and Social Networks and Social Contagion (FS-12-08). The highlights of each symposium are presented in this report. The development of user-centered technologies that assist older adults to live independently and also reduce the burden on caregivers is gaining more attention due to increasing healthcare costs and the aging population. AI is central to these technologies as it deals with the process of transforming raw sensor data into human-interpretable abstractions, innovating new human computer interfaces, as well as planning and reasoning. The symposium provided an intimate setting for researchers from the disciplines of computer science, engineering, nursing, psychology, cognitive science, and health informatics to take stock of the state of the art, highlighting successes and failures, while discussing new problems and opportunities.

One obstacle to progress is the lack of a precise and detailed theory of how humor operates. Nevertheless, since the early 1990s, there have been a number of small programs that create simple verbal humor, and more recently there have been studies of the automatic classification of the humorous status of texts. In addition, there are a number of advocates of the practical uses of computational humor: in user interfaces, in education, and in advertising. Computer-generated humor is still quite basic, but it could be viewed as a form of exploratory creativity. For computational humor to improve, some hard problems in AI will have to be addressed.

How the Theory Explains Humor. Consider the following statement: Biological organisms are driven by the "Four Big F's": Feeding, Fighting, Fleeing, Mating. Some subjective observers who read this for the first time think it is funny. As the eyes are sequentially scanning the text the brain receives a complex visual input stream. The latter is subjectively partially compressible as it relates to the observer's previous knowledge about letters and words.

The notion of constructions, from Construction Grammar, is borrowed for use in joke recognition by a knowledge-based computational text analysis system. The joke recognizer is a proposed addition to an existing text analysis framework, Ontological Semantic Technology. Joke recognition is based upon calculation that the candidate text exhibits qualities similar to jokes already collected and represented in a taxonomy, with other processing input. Joke templates, based on constructions, provide semantic scripts against which texts are judged. With these scripts, meta-jokes, which conform almost but not completely to a known joke script, may also be recognized.

Some humorous texts can be detected by stereotyped patterns and terminology. But a humorous story or situation is often an exaggeration of patterns that also occur in serious texts: novelty, unusual plot twists, and situations that disrupt normal social conventions. The same methods for detecting novelty in serious texts can be adapted to detecting novelty in a humorous situation, but with additional tests for features that make it humorous. To interpret and reason about natural language texts, VivoMind Research has developed a cognitive architecture based on societies of heterogeneous intercommunicating agents that use conceptual graphs (CGs) as the knowledge representation. CGs are designed for representing semantics at the level of sentences and paragraphs, but they must be related to larger patterns that span an entire story, article, or book. For detecting and analyzing large-scale patterns, catastrophe theoretical semantics has proved to be surprisingly effective. This article discusses applications to both fictional and nonfictional documents of various kinds, both serious and humorous.

This talk will analyze responses to funniness of five versions of 10 different jokes. The responses of one of them will then be compared to theoretical analysis and representation of the same joke based on Script-based Semantics Theory of Humor, General Theory of Verbal Humor, and Ontological Semantic Theory of Humor.

This study extends humor theory to explain puns originating from orthographic conversion in Japanese computer-mediated communication (CMC). Standard word-processing software converts Romanized input to appropriate orthographic output consisting of phono-graphic kana and ideographic kanji . Such software may produce an output often semantically incongruent with the intended output, which can be humorous. The dataset analyzed here consists of 492 online submissions to the “Humorous Misconversion Contest” held by the Japan Kanji Aptitude Testing Foundation. Since not all misconversions are funny, the study accounts for how misconversions satisfy funniness conditions of the Semantic Script Theory of Humor. The study finds that script interpretability, as the basis of script compatibility, and script opposition are of most importance in humor perception. It also finds that script oppositeness resides not only within texts but also in outer contexts. As yet, very few academic studies have discussed humor in Japanese CMC. Since a majority of verbal humor is researched on alphabet-based languages, the observations here are expected to enrich and broaden our knowledge of humor.

The general goal of the symposium was to advance the state of the art in the direction of developing an AI system (the system) capable of understanding the mechanism of a joke at a level sufficient for providing a punch line to a human generated setup (even if unintentional) and conversely, for computer reacting competently to a human generated punch line that follows a setup, generated by either participant. The effort is multidisciplinary in nature, and the participants from several of the contributing disciplines, viz., computational semantics, knowledge representation, computational psychology, humanoid robotics, human-computer interface, human factors, to name just a few, took part in the work of the symposium.