<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Mary Lou Maher on Medium]]></title>
        <description><![CDATA[Stories by Mary Lou Maher on Medium]]></description>
        <link>https://medium.com/@mmaher9?source=rss-45c1b16cd26d------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*Cu-VgfnA5AktfqG8</url>
            <title>Stories by Mary Lou Maher on Medium</title>
            <link>https://medium.com/@mmaher9?source=rss-45c1b16cd26d------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 03:23:19 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@mmaher9/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[The Grounding Problem and Co-Creative AI]]></title>
            <link>https://pub.aimind.so/the-grounding-problem-and-co-creative-ai-1b3348a0150e?source=rss-45c1b16cd26d------2</link>
            <guid isPermaLink="false">https://medium.com/p/1b3348a0150e</guid>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[the-grounding-problem]]></category>
            <category><![CDATA[co-creative-ai]]></category>
            <category><![CDATA[cognitive-systems]]></category>
            <category><![CDATA[large-language-models]]></category>
            <dc:creator><![CDATA[Mary Lou Maher]]></dc:creator>
            <pubDate>Tue, 22 Aug 2023 04:10:35 GMT</pubDate>
            <atom:updated>2023-08-28T14:36:42.138Z</atom:updated>
            <content:encoded><![CDATA[<p>I recently read a paper, called The Vector Grounding Problem by Mollo and Millière (2023), that inspired me to think about how Large Language Models (LLMs) can be creative. This train of thought then led to a reflection on how the creativity in LLMs differs from years of research on computational creativity in which we developed symbolic models of creative processes. Could these two approaches be connected? In this article I explore The Grounding Problem as an explanation of how AI systems can generate content that we consider creative and how that explanation may also provide insight how humans can meaningfully interact with Generative AI, and bring human values to bear on the generative process. The ideas in this article have been influenced by conversations with Dan Ventura and Brian Magerko that are recorded in a paper in the AAAI 2023 Fall Symposium (Maher, et al, 2023). Some of the paragraphs in this article are reproduced in the Fall Symposium paper.</p><p>The Grounding Problem asks how AI systems can be grounded in a world with which they have no direct interaction. The Grounding Problem refers to the relationship between AI systems and the world. It was originally articulated as The Symbol Grounding Problem, relevant to symbolic AI, and more recently as The Vector Grounding Problem due to the advances in deep learning neural networks that are the basis for LLMs and other Foundation Models. The Symbol Grounding Problem, framed by Harnad (1990), questions how a cognitive AI system can acquire meaningful semantics from purely symbolic representations. The Vector Grounding Problem, framed by Mollo and Millière (2023), asks the same question of Large Language Models: how can the manipulation of vectors in deep neural networks produce representations that have intrinsic meaning? Following from this line of reasoning, can we make LLMs stronger by establishing a communication between LLMs and cognitive models by mapping the symbols to the vectors?</p><p>The term grounding is used across various fields and in various ways, and in this article, I build on the concept of grounding in cognitive psychology. Mollo and Millière have identified five types of grounding to better understand the grounding problem for AI systems (2023): sensorimotor, communicative, epistemic, relational, and referential. Sensorimotor grounding is present in an AI system when a lexical concept is linked to sensorimotor representations available within the system, as implemented in embodied cognitive systems or robots. Communicative grounding is a coordinated action that involves collaborating to reach a common understanding of what is said and may involve explicit clarification strategies, for example, a conversation using iterative prompting in LLMS. Epistemic grounding is a relationship between a lexical concept and specific data as can be achieved in LLMs when they are connected to external databases used to retrieve information.</p><p>Relational grounding refers to intra-linguistic relationships: a word’s meaning in a language is partly determined by its relations to other words. Vector space models in LLMs trained on large amounts of data can capture a wide range of usage-based relationships between words, including informal relationships that may be missed by lexical decomposition in symbolic AI. LLMs represent words in a continuous vector space, so that semantic relationships are reflected in distance relations in the vector space, capturing nuances in relatedness that are harder to capture with the discrete components and relations of lexical decomposition in cognitive systems.</p><p>Referential grounding refers to the connection of (internal) representations to things in the (external) world. For example, the word/symbol “dog” is grounded in the notion of a dog in the world and the word/symbol “creativity” is grounded in the notion of creativity in the world. Referential grounding makes no claim that the word is the same as the real world thing in any physical sense, but that it is grounded in the way we have constructed meaning for that word in symbolic AI or in the way we use that word in language as modeled in LLMs. In a representation that exhibits referential grounding, reasoning about the world in the space of symbols in cognitive systems and operations on the vector in the latent vector space of deep neural networks have meaning in the associated notions in the world. This is where the insight for how we find meaning in both LLMs and cognitive models of creativity emerges. Figure 1 is an attempt to describe this grounding: both the symbol in our symbolic models and the vector for a word in LLMs map onto a concept in the world. When they map to the same concept, the symbol and the word provide the mapping between the symbolic and generative models.</p><figure><img alt="Illustration of referential grounding" src="https://cdn-images-1.medium.com/max/680/1*lBg2_4ccqcSJfqdDHfo9Cw.png" /></figure><p>Figure 1. Grounding symbols and vector to concepts that have meaning in the world.</p><p>Generative AI, with the recent advances in LLMs, is able to achieve remarkable results with a fundamental process of predicting the next word. The widespread use of LLMs, such as chatGPT, DALLE-2, Midjourney, and GitHub Co-pilot, have surprised even the creators of these models in their ability to generate meaningful responses to natural language prompts. With the selection of types of responses from the LLM, such as precise or creative, the user can guide the generative model to use a specific criterion for choosing the next word (this is an oversimplified explanation). If that choice is creative, the content generated by the LLM is intended to be less common, or we could say more novel.</p><p>Prior to the widespread availability of LLMs, we have developed cognitive models of expectation, surprise, and similarity/novelty (Grace and Maher, 2019). In a symbolic representation of context and features in a design space, we can determine how surprising an individual feature f is given a particular context c, which is a set of other features (c ⊂ F, f /∈ c). We denote this s(f|c), or “the surprise of f given c”. With referential grounding, we can measure probabilities of occurrence, expectation, and surprise and assume that these measurements map to the notions in the world. We evaluate s(f|c) as the ratio between the overall probability of f and the conditional probability of f given the context c. In other words, how many times less likely f is to occur alongside c than it is to occur altogether. Given the probabilities P(f) and P(f|c) this can be calculated as:</p><p>s(f|c) = log2(P(f )/P(f|c)).</p><p>Using the vector space in a trained deep learning neural network, again, assuming the vector space exhibits referential grounding, the cosine similarity of two vectors maintains the semantic and syntactic distance in the associated notions in the world. This is based on the observation that if the vector presentation exhibits referential grounding, then the operations on the vectors maintain the relationships in the notions in the world. In LLMs, this is one explanation for how we can achieve accuracy in predicting the next word. In creativity research, after training a CNN with features in a large dataset of designs, we have exploited this referential grounding to define a semantic distance function δsm(d1,d2) that represents the context or purpose of a design, and a syntactic distance function δsy(d1,d2) that represents the features of a design. Each distance function operates over pairs of designs. The surprise exhibited by a pair of designs is then evaluated as:</p><p>s(d1,d2) = |δsm(d1,d2) — δsy(d1,d2)|.</p><p>We have used these and other formal operations on symbol and vector spaces in co-creative systems in order to evaluate their effectiveness in human perception of creativity (for example, Karimi, et. Al. 2022; Kim et. Al. 2021; Rezwana and Maher, 2022).</p><p>Let’s assume that referential grounding is both the ambition of and the foundation for all of the other types of grounding. While any of the types of grounding can be used to explain or describe some aspect of how AI systems achieve meaning, and it’s possible that grounding in any of these senses can act as a mechanism for integrating cognitive and generative AI in a co-creative system. For example, Figure 2 shows how a co-creative system in which the AI inspires human creativity can be strengthened by connecting symbolic models with the vector models of pre-trained deep learning neural networks.</p><figure><img alt="co-creative grounding with cognitive and deep learning models" src="https://cdn-images-1.medium.com/max/936/1*j-ti7HRgDjqO2tk1bLkc6g.png" /></figure><p>Figure 2. Integrating Cognitive Models and Deep Neural Network Models for Co-Creative Design</p><p>The top of Figure 2 shows the human-AI interaction space in which the user has a canvas for sketching solutions to a design problem and the AI presents inspirational sketches based on the state of the user’s design process. The downward arrows on the left of the diagram show the communication between the human-AI interaction space to the cognitive models is a sense making process, and the communication between the cognitive model and the pre-trained deep learning models results in a prompt. The upward arrows on the right of the diagram show the communication from the deep learning models to the cognitive models as a response to the prompt, followed by a cognitive process that determines the appropriate inspiration for the user in the human-AI interaction space.</p><p><em>If both the cognitive models and the vector models are referentially grounded to the way humans make sense of the concepts in the world, then this combination works.</em></p><p>We elaborate on the potential for an integrated cognitive and generative AI by describing how such an AI system achieves: alignment, grounding, and instructability. AI systems must be judged by how well their operations align with societal expectations and human intentions. The alignment problem, as described by Christian (2020), is exacerbated by training machine learning models with data that is biased or in other ways does not capture human values. We have the potential to capture human values in cognitive systems and link them to the prompt-response processes in generative systems. Grounding allows generative AI systems to demonstrate a connection between its outputs and the abstract concepts with which they operate through explicit connection to cognitive models. Therefore, when both symbolic and vector models exhibit grounding, there is the potential for the symbols and associated vectors to be the bridge between cognitive models and deep learning models trained on large data sets. Models of human cognition can therefore provide a bridge between language models trained on large data sets to human values. Instructible AI systems change their behavior in response to instructions from non-AI-experts to implement more effective and trustworthy assistance to humans. The development and integration of an ontology of prompts that are structured based on cognitive models of human values and learning can guide human interaction with LLMs that has the potential to make instructible AI systems available to all users.</p><p>Referential grounding can be used as an ontological framework to produce stronger AI by adopting the stance that both cognitive and deep neural network AI systems exhibit grounding in the world, providing a path for integrating cognitive and generative AI. Cognitive AI systems reflect our understanding of human cognition and can play a significant role in the future of guiding and evaluating the response of deep learning models. Referential grounding as a basis for understanding how cognitive systems can provide models for guiding and evaluating the responses from generative models and how generative models provide a robustness in the use of language that is difficult to obtain with cognitive models alone. This also provides the basis for humans playing a central directing role since referential grounding is also the basis for the meaning and values that humans ascribe to the symbols in cognitive AI and the words in an LLM.</p><p>References</p><p>Christian, B. (2020). <em>The alignment problem: Machine learning and human values</em>. WW Norton &amp; Company.</p><p>Grace K., Maher M.L. (2019). Expectation-Based Models of Novelty for Evaluating Computational Creativity. In: Veale T., Cardoso F. (eds) <em>Computational Creativity. Computational Synthesis and Creative Systems</em>. Springer.</p><p>Harnad, S. (1990), ‘The symbol-grounding problem’, <em>Physica D</em> 42, 335–346.</p><p>Karimi, P., Rezwana, J., Siddiqui,S., Maher, M.L., Dehbozorgi. N. (2020). Creative Sketching Partner: An Analysis of Human-AI Co-Creativity, In <em>Proceedings of Intelligent User Interfaces</em>.</p><p>Kim, J., Maher, M.L., and Siddiqui, S. (2021). Collaborative Ideation Partner: Design Ideation in Human-AI Co-Creativity. CHIRA2021: <em>5th International Conference on Computer-Human Interaction, Research, and Applications</em>.</p><p>Maher, M.L., Ventura, D., Magerko, B. (2023). The Grounding Problem: A Path for the Integration of Cognitive and Generative Models, AAAI 2023 Fall Symposium.</p><p>Mollo, D. C., &amp; Millière, R. (2023). The vector grounding problem. arXiv preprint arXiv:2304.01481</p><p>Rezwana, J., and Maher, M.L. (2022). Understanding User Perceptions, Collaborative Experience and User Engagement in Different Human-AI Interaction Designs for Co-Creative Systems, Proceedings of Creativity &amp; Cognition.<a href="https://doi.org/10.1145/3527927.3532789"> https://doi.org/10.1145/3527927.3532789</a></p><h4>A Message from AI Mind</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/250/0*5Wm7sOfTpe5DEbhg.gif" /></figure><p>Thanks for being a part of our community! Before you go:</p><ul><li>👏 Clap for the story and follow the author 👉</li><li>📰 View more content in the <a href="https://pub.aimind.so/">AI Mind Publication</a></li><li>🧠 Improve your <a href="https://www.aimind.so/prompt-generator?utm_source=pub&amp;utm_medium=message">AI prompts effortlessly and FREE</a></li><li><strong>🧰 Discover </strong><a href="https://www.aimind.so/?utm_source=pub&amp;utm_medium=message"><strong>Intuitive AI Tools</strong></a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1b3348a0150e" width="1" height="1" alt=""><hr><p><a href="https://pub.aimind.so/the-grounding-problem-and-co-creative-ai-1b3348a0150e">The Grounding Problem and Co-Creative AI</a> was originally published in <a href="https://pub.aimind.so/">AI Mind</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What if the goal of Artificial General Intelligence (AGI) is to enhance human creativity?]]></title>
            <link>https://medium.com/human-centered-ai/what-if-the-goal-of-artificial-general-intelligence-agi-is-to-enhance-human-creativity-5e4034d97735?source=rss-45c1b16cd26d------2</link>
            <guid isPermaLink="false">https://medium.com/p/5e4034d97735</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[human-centered-ai]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[creativity]]></category>
            <dc:creator><![CDATA[Mary Lou Maher]]></dc:creator>
            <pubDate>Thu, 04 May 2023 12:44:11 GMT</pubDate>
            <atom:updated>2023-06-28T15:45:22.302Z</atom:updated>
            <content:encoded><![CDATA[<h4>by <a href="http://maryloumaher.net/">Mary Lou Maher</a> (University of North Carolina, Charlotte, US)</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/1*_j9lSDLcebtwHzheqFMiuQ.jpeg" /><figcaption>Abstract representation of creativity. Produced with Stable Diffusion.</figcaption></figure><p>Creativity is foundational to human learning, social and scientific innovation, quality of life, and economic prosperity. Creativity is practiced in context by all of us in our work lives, education, and social interactions. What if we re-align the goals of Artificial General Intelligence with a focus on interaction with AI that augments human creativity in all of its forms, from common to extraordinary acts of creativity (Weisberg 1993)? A vibrant research community is studying ways to build creative AI systems, and more recently, has focused on developing co-creative systems (e.g., Davis et al. 2015b; Goel et al. 2015; Maher 2012; Muller et al. 2020; Truesdell et al. 2021). What if the advances in large language models — accurately predicting the next word and using reinforcement learning to improve conversational exchanges — are rewarded for enhancing human creativity?</p><p>Creative ideas are characterized as satisfying a duality of novelty (or surprise) and usefulness. The advances in AI, and in large language models more recently, have focused on prediction accuracy as a proxy for the usefulness of the AI response. ChatGPT, a pre-trained language model that is based on the transformer architecture, implements the idea of self-attention, which allows the model to attend to different parts of the input sequence during processing. The GPT models are trained on vast amounts of data using unsupervised learning techniques with the basic goal of predicting the next word. Reinforcement learning from human feedback is used to improve conversational agents like ChatGPT. The goals of prediction accuracy and conversational coherence are mechanisms for improving usefulness. An alternative value alignment for AGI is to include novelty as equally important to usefulness in training the model to improve its conversational performance. By including novelty as a training objective, it signals that an important role that AGI can play is to enhance human creativity, including the personal creativity we experience in learning something new, as well as historical creativity in generating ideas that are transformative in a field (Boden, 2004).</p><p>There are numerous concerns about the recent release of LLMs such as ChatGPT and its potential negative impact on human control as AI acquires more knowledge. Stuart Russel recently gave a presentation called “<a href="https://www.youtube.com/watch?v=ISkAkiAkK7A">How to not to destroy the world with AI</a>.” In his presentation, he spoke about how we can redefine the objectives of AI. From a technical perspective, an objective of AI is accuracy, where accuracy is a metric for measuring the performance of AI models in tasks such as classification or prediction. In classification tasks, accuracy is the proportion of correctly classified data items compared to the total number in the dataset. Other metrics like precision, recall, or F1 score may also be relevant. If we broaden the objectives of AI to align with shared human values, we go beyond technical performance to a socio-technical performance evaluation. Shared human values refer to fundamental ideals that are considered shared across all human beings, regardless of cultural or societal differences. Examples of shared human values are empathy, fairness, and creativity. Redefining, or extending, the objective of AI from accuracy to shared human values is informed by the distinction between common goals and indexical goals in Russell’s presentation. Common goals, such as, “to mitigate climate change,” are of value no matter who achieves them. Indexical goals, such as “to drink coffee,” are only valuable to the person that achieves the goal. As he points out, developing a robot to drink coffee for us is not very useful. Establishing objectives for AGI that are based on common goals means that the achievement of the goal is of value to all of us.</p><p>This concern about the value alignment for AI is also argued in Brian Christian’s book, “<a href="https://en.wikipedia.org/wiki/The_Alignment_Problem">The Alignment Problem: How Can Artificial Intelligence Learn Human Values?</a>” (Christian, 2020). This book argues that current AI models are very good at prediction, but are biased by the data set used to train the model. In essence, AI models effectively propagate historical biases present in our society without regard for our current or future values. Christian argues for a better alignment with human values, including diversity, inclusion, and curiosity.</p><p>For instruction-tuned large language models such as ChatGPT, if the reinforcement learning phase includes a <em>dual</em> objective of usefulness and novelty of the AI’s response, we can begin to enhance human everyday creativity, including encouraging human learning, rather than only seeing a future in which AGI replaces humans. We have an opportunity to measure novelty in language models generated by deep learning with the mapping from natural language tokens to numerical vectors. An alternative to prediction as a metric for the response to a prompt is measured novelty in which the response encourages human ideation (Kim and Maher, 2023). This is one way to operationalize novelty and shift the goal of AI from prediction accuracy to enhancing human creativity.</p><h3>References</h3><ul><li>​​Boden, M.A. 2004. The creative mind: Myths and mechanisms. Psychology Press.</li><li>Davis, N., Hsiao, C.-P., Popova, Y., and Magerko, B. 2015a. An enactive model of creativity for computational collaboration and co-creation. In: Creativity in the Digital Age. Springer, 109–133.</li><li>Kim, J. And Maher, M.L. (2023). The Effect of AI-based Inspiration on Human Design Ideation, International Journal of Design Creativity and Innovation, 11:1.</li><li>Goel, A., Creeden, B., Kumble, M., Salunke, S., Shetty, A., and Wiltgen, B. 2015. Using Watson for enhancing human-computer co-creativity. 2015 AAAI Fall Symposium Series.</li><li>Maher, M.L. 2012. Computational and Collective Creativity: Who’s Being Creative? International Conference on Computational Creativity.</li><li>Muller, M., Ross, S., Houde, S., et al. 2022. Drinking Chai with Your (AI) Programming Partner: A Design Fiction about Generative AI for Software Engineering. Joint International Conference on Intelligent User Interfaces Workshops: HAI-GEN.</li><li>Truesdell, E.J., Smith, J.B., Mathew, S., et al. 2021. Supporting Computational Music Remixing with a Co-Creative Learning Companion. ICCC, 113–121.</li><li>Weisberg, R. 1993. Creativity: Beyond the myth of genius. WH Freeman.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5e4034d97735" width="1" height="1" alt=""><hr><p><a href="https://medium.com/human-centered-ai/what-if-the-goal-of-artificial-general-intelligence-agi-is-to-enhance-human-creativity-5e4034d97735">What if the goal of Artificial General Intelligence (AGI) is to enhance human creativity?</a> was originally published in <a href="https://medium.com/human-centered-ai">Human-Centered AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[On AI Anthropomorphism: Commentary by Mary Lou Maher]]></title>
            <link>https://medium.com/human-centered-ai/on-ai-anthropomorphism-a-commentary-9dfdf7a34f24?source=rss-45c1b16cd26d------2</link>
            <guid isPermaLink="false">https://medium.com/p/9dfdf7a34f24</guid>
            <category><![CDATA[interaction-design-for-ai]]></category>
            <category><![CDATA[ai-and-diversity]]></category>
            <category><![CDATA[human-centered-ai]]></category>
            <category><![CDATA[ai-anthropomorphism]]></category>
            <dc:creator><![CDATA[Mary Lou Maher]]></dc:creator>
            <pubDate>Wed, 26 Apr 2023 01:09:17 GMT</pubDate>
            <atom:updated>2023-06-01T14:18:48.390Z</atom:updated>
            <content:encoded><![CDATA[<h4><em>by </em><a href="http://maryloumaher.net/"><em>Mary Lou Maher</em></a><em> (University of North Carolina, Charlotte, US)</em></h4><p><em>Editor’s note: This article is a commentary on “</em><a href="https://medium.com/human-centered-ai/on-ai-anthropomorphism-abff4cecc5ae"><em>On AI Anthropomorphism</em></a><em>,” written by </em><a href="https://www.cs.umd.edu/users/ben/"><em>Ben Shneiderman</em></a><em> (University of Maryland, US) and </em><a href="https://research.ibm.com/people/michael-muller"><em>Michael Muller</em></a><em> (IBM Research, US). We have reproduced the commentary in its original form, sent as an email to Chenhao Tan and Justin Weisz, with light typographical editing.</em></p><p>I am commenting on the conversation between Ben and Michael with a goal to shift the focus of the conversation on AI anthropomorphism to the benefits as primary arguments for AI anthropomorphism, and focus our concerns as arguments for guardrails that protect people from negative consequences.</p><p>Ben starts the conversation with concern about the way GPT-4 output uses first-person pronouns, suggesting it is human, “In my world, machines are not an ‘I’ and shouldn’t pretend to be human.” Michael responds that there are many precedents for people to address made-things as having human-like qualities. He expresses concerns, but not for the use of the first person pronoun, rather “to protect people from AIs that are feigning human-ness to fool people.”</p><p>Michael goes further to explain that this phenomenon of ascribing human-like qualities to made-things has been around for a long time, and it is likely to continue. Not that this is without concern: but we should be mindful of what we are concerned about. As Michael points out, we should protect people from AIs that are trying to fool us with their human-ness.</p><p>My research team has been exploring the design and impact of human-AI co-creativity, identifying both the benefits and the concerns [1, 2, 3]. AI can augment human creativity, and depending on the interaction design, co-creative AI can be more engaging than your typical creativity support tool. This ability to increase engagement through human-like interaction comes with the need for guidelines and responsible interaction design. When engaging the user involves a conversational human-like interaction, the user tends to anthropomorphize. This often includes ascribing additional human-like attributes to AI such as intention, responsibility, and ownership. This is where we need to design interaction with AI to ensure the guardrails are in place.</p><p>To explore the idea that Michael presented on AI fooling people about its human-ness, I asked some questions in ChatGPT. First, I asked if ChatGPT can be creative, and the response from ChatGPT was:</p><blockquote>“ChatGPT is capable of generating text that may appear to be creative, but it does not have the ability to be truly creative in the same way that humans do. It is a machine learning model that has been trained on a large dataset of text, and it generates new text by making predictions based on patterns it has learned from that data. The model is not capable of experiencing emotions, understanding context, or having personal experiences, which are all factors that contribute to human creativity. Therefore, its creative ability is limited by its programming, and it is ultimately an algorithm that generates text based on patterns it has learned.”</blockquote><p>Then I asked, “Are you creative?” The response to this question using first-person for the AI was:</p><blockquote>“As an AI language model, I can generate text that may appear to be creative, but I do not have personal experiences or emotions, which are key factors in human creativity. I rely on patterns and data that I have learned from my training to generate responses, and my responses are limited to the information that has been inputted into my system. Therefore, while I may generate novel and interesting responses, I am not capable of being truly creative in the way that humans are.”</blockquote><p>While this is only a single example, I do not see in these responses that there is the intention to fool people with the use of first-person pronouns, but rather to engage and respond to people in a conversation about any topic of interest to the person. In ChatGPT, the AI model is used in response to a prompt, it is not directing the conversation. While there is a need for more AI literacy so that more people can critically reflect on the responses from ChatGPT, I do not feel we need to remove the first-person interaction.</p><p>I support Michael’s argument: the problem is not anthropomorphism, the problem is the use of anthropomorphism to do bad things. Rather than try to stop AI anthropomorphism, I would like to see more arguments in favor that go beyond the precedent argument. There is a growing observation among educators and researchers that the conversational tone of AI is encouraging learning opportunities for people that are marginalized in our society. The use of natural language, a familiar tone, and a non-judgmental AI lowers the bar for many to ask questions and explore topics. I lead a community of practice at UNC Charlotte for instructors in the core courses for a computer science degree. A common problem in teaching computer science to diverse students is that a lack of sense of belonging results in some students not asking questions of the professor or TA for fear of looking stupid. As we help students learn how to make use of AI such as ChatGPT and Github Co-Pilot, we are observing that students that normally do not ask questions are excited with the opportunity to use these new technologies for their assignments. There is a need for AI literacy and guidance in the critical reflection of what AI generates. But this does not mean we need to remove the first-person interaction design. Reduced opportunities for advanced education for marginalized groups of people have led to those who are privileged being the ones that make the technology, often carrying systemic discrimination and bias into the products produced by this privileged group of highly educated workers.</p><p>I argue for a focus on how AI anthropomorphism provides a path for more diversity in higher education. As Michael points out in his response to the importance of metaphors, the use of human language for interacting with AI will open new interaction design possibilities. While we want to protect people from being fooled, we don’t want to go backward by banning the use of first-person AI or access to AI so that the privileged can have more control over how AI is used. We need to bring more voices to the table — and increase the diversity of people that have a voice. As<a href="https://medium.com/user-experience-design-1/ethical-evolution-in-the-age-of-ai-why-fear-and-complacency-are-not-options-2f8fd35eec16"> Louis Byrd says</a>: AI is not the problem, humans are.</p><h3>References</h3><p>[1] Rezwana, J., &amp; Maher, M. L. (2022). Designing Creative AI Partners with COFI: A Framework for Modeling Interaction in Human-AI Co-Creative Systems. ACM Transactions on Computer-Human Interaction. <a href="https://doi.org/10.1145/3519026">https://doi.org/10.1145/3519026</a></p><p>[2] Rezwana, J., and Maher, M.L. (2022). Understanding User Perceptions, Collaborative Experience and User Engagement in Different Human-AI Interaction Designs for Co-Creative Systems, Proceedings of Creativity &amp; Cognition. <a href="https://doi.org/10.1145/3527927.3532789">https://doi.org/10.1145/3527927.3532789</a></p><p>[3] Rezwana, J. and Maher, M.L. (2022). Identifying Ethical Issues in AI Partners in Human-AI Co-Creation, CHI 2022 Workshop on Generative AI and HCI, <a href="https://generativeaiandhci.github.io/">https://generativeaiandhci.github.io/</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9dfdf7a34f24" width="1" height="1" alt=""><hr><p><a href="https://medium.com/human-centered-ai/on-ai-anthropomorphism-a-commentary-9dfdf7a34f24">On AI Anthropomorphism: Commentary by Mary Lou Maher</a> was originally published in <a href="https://medium.com/human-centered-ai">Human-Centered AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[HAI-GEN 2023: 4th Workshop on Human-AI Co-Creation with Generative Models]]></title>
            <link>https://medium.com/human-centered-ai/hai-gen-2023-4th-workshop-on-human-ai-co-creation-with-generative-models-32a61dfb67fa?source=rss-45c1b16cd26d------2</link>
            <guid isPermaLink="false">https://medium.com/p/32a61dfb67fa</guid>
            <category><![CDATA[co-creativity]]></category>
            <category><![CDATA[generative-ai]]></category>
            <category><![CDATA[human-centered-ai]]></category>
            <dc:creator><![CDATA[Mary Lou Maher]]></dc:creator>
            <pubDate>Mon, 17 Apr 2023 13:02:05 GMT</pubDate>
            <atom:updated>2023-06-01T14:18:04.222Z</atom:updated>
            <content:encoded><![CDATA[<h4><em>by </em><a href="http://maryloumaher.net/"><em>Mary Lou Maher</em></a><em> (University of North Carolina, Charlotte, US), </em><a href="https://research.ibm.com/people/justin-weisz"><em>Justin D. Weisz</em></a><em> (IBM Research, US), </em><a href="http://hendrik.strobelt.com/"><em>Hendrik Strobelt</em></a><em> (IBM Research, US), </em><a href="https://www.cs.columbia.edu/~chilton/"><em>Lydia B. Chilton</em></a><em> (Columbia University, US), and </em><a href="https://research.ibm.com/people/werner-geyer"><em>Werner Geyer</em></a><em> (IBM Research, US)</em></h4><p><em>Editor’s note: This article was updated on April 19 to include the workshop recording.</em></p><p>Recent advances in generative AI through deep learning approaches such as generative adversarial networks, multi-modal generation, and large language models such as GPT-4 are enabling new kinds of user experiences. People are now able to generate many types of content — text, images, audio, video, code — at an unprecedented level of fidelity. In many cases, the content generated by AI models is either indistinguishable from human-generated content or could not be produced by humans. This ability raises significant creative, societal, ethical, and organizational challenges.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/0*Fp03mZoCgpGyw8b7.png" /><figcaption>“A chair in the style of GAN trained on chairs.” Prompt inspired by Philipp Schmitt’s keynote. Images produced by Midjourney.</figcaption></figure><p>The <a href="https://hai-gen.github.io/2023/">4th Workshop on Human-AI Co-Creation with Generative Models</a>, held at <a href="http://iui.acm.org/">IUI 2023</a>, brought together over 50 researchers and practitioners from the disciplines of HCI &amp; AI to explore our understanding of the human-AI co-creative process and the opportunities and challenges of creating meaningful user experiences with deep generative models.</p><h3>Workshop Summary</h3><p><a href="https://philippschmitt.com/">Philipp Schmitt</a>, an artist and designer based in Brooklyn, USA opened the workshop by delivering a keynote that reflected on his use of generative AI technologies in the design of physical artifacts. He challenged the conventional practice of using generative models to create high-fidelity artifacts by showing how the generation of low-fidelity images was sufficient to stimulate peoples’ creative thinking. In this way, the “prompting” process was flipped: low-fidelity outputs of a generative model were used to prompt human designers into creating a novel collection of chairs, beginning with hand-drawn sketches, followed by small-scale and full-scale prototypes. Philip questioned the assumption and expectation that having larger and larger language models equals increased co-creativity as this may leave less room for human imagination, similarly discussed in a recent article by <a href="https://thebulletin.org/2022/10/ai-is-plundering-the-imagination-and-replacing-it-with-a-slot-machine/">Annie Dorsen</a>. He ended his keynote provocatively by asking us “Let’s do less!”</p><p>Following the keynote was the first paper session which included a lively discussion on topics exploring how people interact with generative AI systems.</p><ul><li>Mattias Rost and Sebastian Andreasson. <a href="https://hai-gen.github.io/2023/papers/7513-paper-HAIGEN-RostMattias.pdf">Stable Walk: An interactive environment for exploring Stable Diffusion outputs</a></li><li>Shuoyang Zheng. <a href="https://hai-gen.github.io/2023/papers/2838-paper-HAIGEN-ZhengShuoyang.pdf">StyleGAN-Canvas: Augmenting StyleGAN3 for Real-Time Human-AI Co-Creation</a></li><li>Juan Salamanca, Daniel Gómez-Marín, and Sergi Jordà. <a href="https://hai-gen.github.io/2023/papers/2358-paper-HAIGEN-SalamancaJuan.pdf">The Dynamic Creativity of Proto-artifacts in Generative Computational Co-creation</a></li><li>Ibukun Olatunji. <a href="https://hai-gen.github.io/2023/papers/1594-paper-HAIGEN-OlantunjiTunde.pdf">Why try to build a co-creative poetry system that makes people feel that they have “creative superpowers”?</a></li></ul><p>Papers in the second session took a dive into critical research and design issues raised by the use of generative AI in creative tasks.</p><ul><li>Yiren Liu, Mengxia Yu, Meng Jiang, and Yun Huang. <a href="https://hai-gen.github.io/2023/papers/2021-paper-HAIGEN-LiuYiren.pdf">Creative Research Question Generation for Human-Computer Interaction Research</a></li><li>Steven I. Ross, Michael Muller, Fernando Martinez, Stephanie Houde, and Justin D. Weisz. <a href="https://hai-gen.github.io/2023/papers/9320-paper-HAIGEN-RossSteven.pdf">A Case Study in Engineering a Conversational Programming Assistant’s Persona</a></li><li>Jeba Rezwana and Mary Lou Maher. <a href="https://hai-gen.github.io/2023/papers/3786-paper-HAIGEN-RezwanaJeba.pdf">User Perspectives of the Ethical Dilemmas of Ownership, Accountability, Leadership in Human-AI Co-Creation</a></li><li>Justin D. Weisz, Michael Muller, Jessica He, and Stephanie Houde. <a href="https://hai-gen.github.io/2023/papers/1272-paper-HAIGEN-WeiszJustin.pdf">Toward General Design Principles for Generative AI Applications</a></li></ul><p>The workshop ended with a panel discussion on the topic of how large language models will impact human-AI co-creativity. The panelists were <a href="http://katygero.com/">Katy Gero</a>, <a href="https://www.sydney.edu.au/architecture/about/our-people/academic-staff/kazjon-grace.html">Kaz Grace</a>, and <a href="https://www.linkedin.com/in/jrezwana005">Jeba Rezwani</a>, and it was moderated by <a href="https://research.ibm.com/people/justin-weisz">Justin Weisz</a>. The topics raised by the panelists provoked questions about how can AI support human authenticity and the intrinsic motivation of people to create their own voice, the concerns about conversational AI in dealing with the way humans ascribe contextual meanings for words that go beyond the conventional meaning, the reversal of the convention that humans prompt generative AI to suggest AI should prompt humans (echoing the keynote), and the call for ethical and responsible human-AI collaboration.</p><h3>Get Involved</h3><p>Generative AI technologies are rapidly maturing and being integrated into a variety of consumer and enterprise applications. Our workshop continued to explore the frontiers of human-AI co-creativity by exploring experimental user experiences, ethical dilemmas, design principles, and more. There are a number of upcoming events to get involved in this community:</p><ul><li>📅 Join the <a href="https://hcxai.jimdosite.com/">Human-Centered Explainable AI</a> the <a href="https://generativeaiandhci.github.io/">Generative AI and HCI</a> workshops at CHI 2023, April 28–29. Registration is now open to all CHI attendees!</li><li>🧑‍🎨 Participate in the <a href="https://computationalcreativity.net/iccc23/">14th International Conference on Computational Creativity</a>. Short papers are due May 2.</li></ul><h3>Watch the Workshop</h3><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FMvxRc2g7KtI&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DMvxRc2g7KtI&amp;image=http%3A%2F%2Fi.ytimg.com%2Fvi%2FMvxRc2g7KtI%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/f57292196c4dada571c8448aba1d66bf/href">https://medium.com/media/f57292196c4dada571c8448aba1d66bf/href</a></iframe><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=32a61dfb67fa" width="1" height="1" alt=""><hr><p><a href="https://medium.com/human-centered-ai/hai-gen-2023-4th-workshop-on-human-ai-co-creation-with-generative-models-32a61dfb67fa">HAI-GEN 2023: 4th Workshop on Human-AI Co-Creation with Generative Models</a> was originally published in <a href="https://medium.com/human-centered-ai">Human-Centered AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>