"Marvel's Secret Invasion": AI and Human Cognition

A piece of news that circulated widely on social media related to pop culture in the past month was about the premiere of the TV series "Marvel’s Secret Invasion," which used artificial intelligence (AI) in the generation of images. The information was confirmed by one of the producers of the series, Ali Selim, who stated that generative AI was used to create the artistic landscapes that appear in the opening credits. In general terms, generative AI can be defined as AI that has the ability to create new information based on existing data sets. And it is at this point that the news reignited the controversy surrounding the use of AI for artistic content production.

Generative AI and Artists

For some time now, the use of generative AI has sparked protests from the artistic community, especially with the popularization of generative image AIs, such as Midjourney or OpenAI's DALL-E. Basically, you provide a prompt, and the platform generates the requested image based on a repertoire of pre-existing information with which the AI is trained. One of the main concerns raised by several artists is the ethical question behind how this pre-existing information is utilized by the AI.

Besides the fact that there is no reference to the previously used training data, which serves as the basis for generating the new images, some people see this problem as an advanced form of plagiarism. The new image generated by the AI has no mention or reference to the content that was used as the basis. Similarly, ChatGPT (a generative text AI) is seen as a potential plagiarizer for not citing sources or references for its responses.

The aim of this text is not to make a judgment on the ethical issue behind the generation of images or texts by AI but to highlight how this characteristic related to generative AI shows how close an AI can come to human cognition. After all, human beings also create novelty based on a repertoire of previously internalized knowledge. In this case, let us specifically analyze memory and imagination.

Defining Memory and Imagination

Psychologist and researcher from Aalborg University (Denmark) Jaan Valsiner defines the existence of two processes: internalization and externalization. Internalization refers to how human beings assimilate the world around them, bringing the concrete objects of the real world into the intrapsychological world. Externalization, on the other hand, is defined as how human beings communicate to their social and cultural environment what has been previously internalized. The main point is that when we externalize something, we do not literally repeat what has been internalized but, rather, communicate a new synthesis that represents a tension between personal and collective culture. This tension is based on the idea that we externalize what we subjectively understand from the material that has been previously internalized, while this externalization also retains attributes of what is universal or culturally shared.

The concepts of internalization and externalization help us understand an important characteristic of memory: its malleability. Old studies, such as those of Sir Frederic Bartlett, or more recent ones like Elizabeth Loftus', show how memory is flexible, malleable, and subject to alterations. Among other things, these studies demonstrate how people recall past events and how the new is constructed through mnemonic modifications influenced by social suggestions, cultural elements, and imagination.

Imagination can be defined as the human capacity to create events or objects that may or may not exist in the real world and to project possible future scenarios. It is through imagination that we create soundscapes, visual landscapes, and sensory experiences based on a vast repertoire of information that we internalize throughout our lives. In other words, we can only imagine something based on previous experiences. In this sense, Tania Zittoun's research from the University of Neuchâtel proposes a model that explains how imaginative movements work. According to her, imagination occurs through a loop (Fig. 1), where future projections are made based on the retrieval of information from past experiences. The loop's handle movement involves two dimensions: abstraction/generalization and implausibility.

Loop of Imagination

Source: Joao R.R.T. da Silva

Abstraction/generalization refers to how removed from the real/concrete the imagined object is. For example, imagining a dragon is much further removed from our real world than imagining a dog. In this case, there is a high level of abstraction or generality. On the other hand, implausibility represents how improbable what is being imagined can be. For instance, imagining myself winning the lottery is much more implausible than envisioning my workday tomorrow. These two dimensions come together in the prospecting of something new that is being imaginatively created.

Creating Novelty From Memory and Imagination

Let's take the creation of a painting representing a dragon being defeated by a medieval knight as an example. To imagine a dragon, the painter must have previously encountered representations of this "creature." Whether through stories, poems, descriptions, cartoons, movies, or depictions in books, all this prior exposure becomes a framework of elements that can be recalled when creating their own representation of a dragon. Even if the painted dragon is a completely new representation, laden with the painter's intrinsic subjectivity, it will be based on this repertoire of previously internalized information. The same goes for the medieval knight and the potential battle scene to be painted (mountains, valleys, rivers, cloudy sky, etc.). Every new creation, even if unique, will carry influences from the past (collective culture) and the artist's subjectivity (personal culture).

Artificial Intelligence Essential Reads

Can We Feel Empathy for AI?

AI Can Make Healthcare More Empathic

Generative AI and Human Cognition

Returning to the issue of generative AI, I raise the following question: How close can an AI come to human cognition? Because the fact that generative AI uses a repertoire of information that was utilized during its training to create something new is not very different from our ability to create something new based on previously internalized knowledge. Of course, there is still a long way to go in the field of AI before we can truly say that it has come close to human cognition. Affective elements, for example, are difficult to emulate in an AI. A painting created by a generative AI today does not carry affective elements or the subjectivity of a creator. How abstract or implausible can an image or text created by AI be? These are open questions that, with time, we may find answers to. In this sense, let us wait and see how far technology will go. In the meantime, I leave you with a painting of a dragon fighting a knight generated by a DALL-E Mini.

A dragon fighting a knight generated by DALL-E Mini

Source: DALL-E Mini/OpenAI