Transformer for emotion

My solution is a large language model based on cutting-edge theories of educational psychology and developmental psychology. This model is structured with an architecture based on state-of-the-art algorithms that allow it, firstly, to conduct a precise assessment of an individual's level of socio-emotional competencies (conducting this evaluation process with criteria that exceed the limitations of current psychometrics) to then be able to carry out a personalized accompaniment and scaffolding process for these types of competencies.

The structuring of this technology will be part of my doctoral training process. Firstly, I will conduct some experiments to better understand how students react to the accompaniment of an artificial agent in guiding decision-making in situations of emotional and moral character, as opposed to decision-making in situations that require information processing of a different nature, such as cognitive situations.

Similarly, this project seeks to understand the causes of human trust in technology, especially centered on two types of trust, epistemic trust and social trust. The former is important for teaching knowledge content or objectives, and the latter is fundamental for thinking about accompaniment that aims at socio-emotional and moral development.

These investigations will take place in the comparison of populations that have daily contact with technology, such as students in the United States, as opposed to populations with minimal technological contact, such as rural populations in Colombia.

This project will seek to reduce gaps in education, as well as to make the most of technologies that have the potential to solve essential problems in current education.

My solution will significantly impact students, particularly in my home country of Colombia. According to various developmental psychology studies, these students face several developmental gaps and are not reaching their full potential. Currently, these opportunity gaps in my country primarily result from the limited access to quality education due to social, contextual, and economic factors.

Education psychology science has proven that there are numerous factors hindering a student's proper educational process, from primary education, through high school education, and subsequent university education. Among these factors, sociocognitive variables play a significant role. For instance, children's relationships with their families, where parents have not been educated for purposeful upbringing that seeks to generate outcomes in children beyond mere survival.

These gaps in parenting skills, in turn, generate chains of gaps in the sociocognitive and socioemotional development of children and adolescents, creating a vicious cycle.

My project aims to assist these children and parents, as well as teachers in various rural institutions, by providing them access to cutting-edge knowledge and practices for sociocognitive and socioemotional development. In this way, the inequalities regarding the lack of knowledge of adequate parenting practices, the lack of resources in institutions for the proper accompaniment in the development of socioemotional competencies, and a precise and personalized evaluation in younger students for their personalized accompaniment can become real objectives. Thanks to artificial intelligence technologies and the emerging large language models, this is now possible.

My team and I are ideally suited to implement this solution due to our deep academic knowledge in the sciences required for the proper structuring of this type of assistance.

Being primarily an academic team at the doctoral level ensures that the models we will use to design large language technologies will not be limited to technical specifications when constructing these models. Instead, they will be based on theories that allow us to have a well-founded understanding to overcome the challenges in the accompaniment of these emerging technologies.

Current large language model technologies have high quality in technical terms and concerning the algorithms they are based on. However, they still present shortcomings due to a lack of understanding of theories focused on providing appropriate accompaniment, such as sociocultural theories that speak to us about precise and well-studied concepts like scaffolding.

In this sense, the composition of my team ensures that the model we construct will not be one with specifications and capacities like current accompanying models. Instead, it will take into account current development evaluation technologies and a potential for accompaniment based on psychology. This approach combines accurate technical knowledge for the creation of artificial intelligence models with cutting-edge psychological theories for appropriate content.

Moreover, our interest in and sensitivity towards vulnerable populations in a country like Colombia mean that the technologies we structure will not be based on a profit motive. Instead, they will be driven by a spirit of solidarity for the reduction of gaps and justice in opportunities.

Given that this project is framed within a doctoral process, it aims to ensure that the final product is not merely an imitation of similar existing technologies, but instead contributes innovative aspects for the proper guidance and development of the competencies described. Currently, it is in a theoretical phase.

In this phase, we are first conducting an experimentation process on the trust in the guidance offered by an artificial agent in decision-making and how this relationship is characterized with a human, particularly with children and adolescents.

After understanding this relationship and based on the results found, we will proceed to construct the technological agent for the appropriate scaffolding of the skills described.

For these reasons, I select that the project is in a conceptual phase.

My solution is innovative as it aims to articulate the fields of Artificial Intelligence and Psychology, not merely seeking technical or theoretical potential, but striving for a solution that integrates cutting-edge theoretical proposals and appropriate technologies and algorithms to achieve social projection and impact.

One aspect in which new artificial intelligence technologies have been innovative is the technical potential they have brought with them. Models such as neural networks and transformer algorithms have achieved goals that seemed impossible a few years ago. However, when reviewing the proposals for support based on these technologies, particularly in educational, socio-emotional, or mental health support, it has been found that there is a bias in understanding that rigorous research is not necessary to deeply comprehend how these two worlds can be united.

On the contrary, current technologies that target these phenomena have been based on an undifferentiated mix between the technical world of advances in AI models and the theoretical world from disciplines such as psychology. This is evident insofar as the current large language models that seek to develop socio-emotional competencies or focus on mental health still have evaluative processes based on psychometrics, aspects that bring multiple limitations in understanding the phenomenon to intervene and subsequently, in the personalized interventions that are carried out.

Current models are based on training in theories of clinical psychology in particular, but not in models whose focus is on accompaniment, such as the fields of developmental psychology.

Our research aims to address these current problems and enhance innovative evaluation strategies that are not simply based on self-perception processes, as well as enhance intervention processes based on appropriate theories.

Primarily, we will be working with natural language processing, as we see structuring large language models as a sustainable and accessible alternative for the main population we will be working with, as opposed to other technologies like social robots.

Furthermore, aspects related to deep learning are fundamental for the appropriate structuring of our technology. This is because one of the main critiques found in literature regarding the evaluation of socio-emotional competencies is precisely that their evaluation is based on psychometric technologies, which are founded on self-perception processes. This generates results that can be biased or imprecise, complicating subsequent support.

In this sense, deep learning technologies are appropriate for finding evaluative strategies based on criteria that transcend simple self-perception, to find specific patterns that give us a much clearer and specific idea of the socio-emotional profile of the participant for their subsequent personalized support. This will be done using textual information, as well as visual information that the constructed model can process to find different relationships between variables that the current psychological scope is not yet capable of.

Finally, through natural language processing technologies, individual forms of appropriate support and intervention can be offered, seeking to be a precise complement for both parenting and formal education processes.

I am confident that this solution will work, based on the progress already made in understanding how artificial intelligence models, particularly large language models, can potentially support mental health.

Furthermore, experiments have shown that children can even exhibit positive biases towards artificial agents, such as robots.

Similarly, research on social and epistemic trust that children place in artificial agents provides evidence of the feasibility of emotional scaffolding based on artificial intelligence technologies.

I would like to refer to four fundamental articles that substantiate my research work.

https://www.sciencedirect.com/...

https://link.springer.com/arti...

https://pubmed.ncbi.nlm.nih.go...

https://www.frontiersin.org/ar...

The primary strategy we have been implementing to avoid these types of biases involves characterizing and understanding the human-machine relationship, not based on currently available data from studies in first-world countries, but from a proper characterization with the Colombian population who will be the main beneficiaries of this technology.

By conducting rigorous research based on the specific needs and characteristics of the Colombian Latin population, we ensure the avoidance of some algorithmic biases that are based on training models with data from users in other countries.

Likewise, the training of the model will be carried out with local data, through a process of participation of urban and rural Colombian populations, so that the Artificial Intelligence model can take into account the unique characteristics of the vulnerable population, from an understanding of their particularities and not external generalizations.

Finally, this model aims to promote equity and fair participation by reducing gaps in the development of vulnerable populations in Colombia, aiming at accessible solutions that seek social transformation and tangible impact on society through concrete technological solutions.

Full-Time Staff:

Johan Sebastián Galindez Acosta, Ph.D Student.

Facultad de Psicología y Ciencias del Comportamiento

Universidad de la Sabana

Chía, Colombia

Part-Time Staff:

Juan José Giraldo Huertas, Ph.D.

Facultad de Psicología y Ciencias del Comportamiento

Universidad de la Sabana

Chía, Colombia

Judith Danovitch, Ph.D.

Department of Psychological and Brain Sciences

University of Louisville

Kentucky, USA

My plans for the pilot are based on structuring various trust evaluation tools in decision-making according to the accompaniment agent, whether they are artificial or not. I can provide a link to the data collection tool that we have prepared for the first pilot in this regard.

Likewise, we already have the theoretical section prepared regarding the systematic review of literature that has been undertaken for the foundation of these technologies. I offer a link for the review of this work.

Additionally, we have permissions to carry out the evaluation process in various schools in the country for the use of the instrument. These schools consist of both rural and urban populations in Colombia. The emails and documents can be shared for verification.

With this process, we guarantee the fulfillment of the theoretical step of the research for the subsequent implementation and testing of the tool based on artificial intelligence.

My solution is accessible insofar as we have evaluated the costs of various intervention strategies using artificial intelligence technologies. Among these evaluated strategies, we have discarded all technologies related to social robots. Although literature demonstrates that these types of robots can generate social trust processes, and therefore, scaffolding processes, they are costly and not easily scalable.

Technologies based on large language models have a better opportunity for scalability, particularly due to their more affordable hardware costs.

Finally, I ensure that access to these technologies is guaranteed through some policies in my country aimed at technological access for the population. These projects have ensured that both the rural and urban educational population can have access to the necessary hardware for the implementation of my technological solution.

I am currently facing economic barriers, as the various costs associated with reaching rural populations in my country and conducting rigorous evaluative and investigative processes to ensure the fulfillment of my research objectives are particularly high.

These costs include transportation, data collection from junior researchers, and the expenses related to the analysis of results and the structuring and training of the proposed model.

In addition to economic barriers, I also face challenges in terms of access to rural populations, given the geographical complexity of my country. To overcome these challenges, I am seeking alliances with local organizations that can facilitate access to these communities. These alliances will also be vital to ensure that the implementation of my solution is conducted in a culturally sensitive and relevant manner.

Moreover, I am in the process of forming a broader research team that can assist in data collection and the analysis of results. This team will consist of globally recognized researchers in the field and technical experts associated with Artificial Intelligence.

The Bill & Melinda Gates Foundation could assist me by providing the necessary resources to address the issues described above.