2026 SEPA POSTER AI and Learning .pdf

Learning with Artificial Intelligence:

Humans Versus AI and Internet of Things

Chuck Robertson, Michele Hill, Gabi Faulkner, Ava Andrick, Carson Sutherland, & Harley Ansbro

Introduction

Artificial Intelligence (AI) has evolved and drastically changed how people work, study and

learn, and live their day-to-day lives. Specifically, students and faculty are increasingly using

AI for understanding complex topics, finding evidence in research endeavors, and editing and

revising their writing (Black & Tomlinson, 2025). This study investigated several forms of AI

(Conversational and Generative), to see how students might learn similarly to studying with a

friend from class. Generative AI (ChatGPT and Gemini) has primarily been defined by its

ability to perform functions that are traditionally associated with humans, such as decision-

making, learning, and completing simple/complex tasks (Collins, et al., 2021). Assistive or

conversational AI appliances known as Internet of Things (IoT) are consumer AI devices that

gather and analyze data together to monitor/control environments, while also achieving

common goals (Choudhary, 2024). The purpose of this research was to compare how well

students learn with different types of AI-enabled systems compared to a human study partner

(study buddy). A traditional pre-test post-test design with general trivia questions was the

comparison task.

Hypotheses

• A main effect of studying, participants' post-test scores will be significantly higher than

pre-test scores.

• A main effect of study partner, with not all partners showing the same gains.

• A significant interaction with learning gain differences based on the type of study partner

(Study Buddy, ChatGPT, Gemini, Alexa, Google Home).

Methods

Participants

• Participants were 100 undergraduate students (age range = 18–40 years, M = 20.12, SD =

3.02). The sample included 60 females, 39 males, and 1 participant identifying as other.

• Most participants identified as White (n = 84), with smaller proportions identifying as

Hispanic/Latino (n = 8), African American (n = 3), Asian (n = 2), or multiracial (n = 2).

• Participants represented a range of majors, with psychology being the largest group (n =

43), followed by biology (n = 6), cybersecurity (n = 5), business (n = 5), kinesiology (n =

5), criminal justice (n = 4), and other majors (n = 32).

Device and Software Ownership and Use

• Eighty percent of participants owned at least one AI-enabled device, and 72% reported

using AI at least once per day. Apple Siri was the most commonly preferred device (n =

35), followed by Amazon Alexa (n = 23).

• Most participants reported being very or fairly comfortable using AI (59%).

• When studying, AI was primarily used for practice testing and information retrieval.

• Of the participants, 40 of them reported using AI as a tool to study while 60 reported that

they did not.

Materials

• Free-recall pre-test, practice test , and post-test that included 30 basic trivia questions,

generated through AI (i.e., What is a group of toads called?).

• A demographic test was administered after the study period, which asked participants

their age, race, academic school year, GPA, and feelings on the use of AI for studying

and other related tasks.

• Participants in the generative AI condition were assigned to either Gemini or ChatGPT

4.0 and used it as an audio tool replicating the IoT device on a computer running Mac

OS to study.

• Participants in the IOT conditions were assigned to either an Amazon Echo Dot Fire or a

Google Home Mini Smart Speaker to study.

• All study sessions happened in real time over a standard wi-fi network.

Procedure

• Participants were voluntarily recruited via the university online experiment scheduling

system (SONA) and in class solicitations.

• Participants were provided with informed consent.

• Participants were instructed to complete a pre-test with 30 questions on general information

from a variety of fields. Participants had up to 10 minutes to complete this section, and once

finished, they moved to the practice test with a study partner.

• Participants were randomly assigned and instructed to use their study partner (Human-Study

Buddy, Gemini, Alexa, ChatGPT, or Google Home) for up to 30 minutes to learn the

answers to the trivia questions.

• Researchers gave a standardized overview of how to interact with each study partner using

their voices. Learners could use their practice time to answer practice questions and

memorize/review answers. No request to record answers physically on the recall sheet was

made to the learners.

• After the practice test, the participants filled out a demographic questionnaire which asked

participants their age, race, academic school year, GPA, and feelings on the use of AI for

studying and other related tasks.

• Participants were given the post-test and were not allowed to use their designated study

partner or review previous test sections.

• At the conclusion of the study, participants were thanked and given a debrief of the study’s

purpose.

Results

• To analyze the results of data collection, a 5 X 3 Factorial ANOVA was used.

• A significant main effect of studying was found, F(1, 95) = 757.61, p < .001, η²ₚ = .89.

Participants were able to answer significantly more questions correctly during the study

condition with their partner (M = 22.12, SD = 5.70) than they were during the post-test (M

= 13.82, SD = 5.75) and pre-test (M = 0.67, SD = 0.91) conditions. The post-test showed

significant gains over the pre-test condition supporting the hypothesis that studying with a

partner has a positive effect.

• A significant main effect of Device was found, F(1, 95) = 744.16, p < .001, η²ₚ = .89. This

supported the hypothesis that not all study partners are equally effective.

• Both main effects were qualified by a significant interaction effect, F(4, 95) = 9.86, p <

.001, η²ₚ = .29. All conditions were equivalent at pretest (M = 0.67, SD = 0.92)). A

Student–Newman–Keuls post hoc test scores indicated that IoT devices, Google Home (M

= 11.20, SD = 4.56) and Amazon Alexa (M = 9.20, SD = 3.41), significantly improved

performance and did not differ from each other. Human partners (M = 15.50, SD = 6.20)

and AI tools Gemini (M = 16.60, SD = 3.87) and ChatGPT (M = 16.60, SD = 6.12) yielded

greater gains than IoT devices and did not differ from each other.

Discussion and Implications

• Generative AI and human study partners produced equivalent improvements in learning

and recall. Participants who studied with ChatGPT, Gemini, or a human performed

significantly better on the post-test than those who used Assistive AI IoT devices like

Amazon Alexa and Google Home.

• IoT showed lesser improvement on the post-test in learning and recall.

• The interactivity of tools/devices appear to drive stronger outcomes. Generative AI tools

support multi-step reasoning, elaboration, and conversational engagement which closely

copies/imitates humans; whereas IoT assistants are optimized for brief, task-oriented

responses.

• Supports social learning theory. Humans learn best through social interaction (De Felice

et al., 2022). AI may partially replicate this benefit through its conversational scaffolding

but, AI does not completely replicate human connection.

• Post-COVID, student socialization patterns shifted and access to tutors/SI sessions

became limited or may still be limited, AI became a flexible, accessible supplemental

study partner.

• Accessibility and equity implications. AI tools may reduce barriers for students who

cannot attend tutoring sessions, struggle to find study groups, or prefer low-pressure, self-

paced interactions.

• Generative AI is not a replacement for human interactions; it is an assistive tool. AI is

effective as a learning support tool. However, over-reliance may reduce critical thinking

and academic scaffolding without the proper prompting. Substituting AI output for

authentic engagement may weaken learning processes.

• Rather than restricting AI outright, institutions should teach students how to use AI

strategically, encourage evidence-based prompting, follow up on the information

gathered from AI, and emphasize AI as academic support, not a substitution.

Limitations

• Data from 2025 suggest that 86% of college students use AI to study, which contrasts

with our data set, which suggests only 40% use it to study (Campbell University, 2025).

IoT data was collected in 2022, before generative AI rolled out and became accessible to

the masses.

• The participants were limited to verbally asking the provided questions and recording

responses; no open-ended dialogue or follow-up clarification was permitted.

Future Directions

• Investigate the difference in learning of complex/specific academic content between AI

and humans, assessing performance on application and formula-based questions.

• Investigate overuse and misuse of AI, including effects on critical thinking skills,

impacts on social learning behaviors, and performance on academic and employment

tasks.

• Assess the difference between socializing with an AI versus a human. Are participants

more fulfilled when having a one-on-one conversation with ChatGPT or a human

study buddy.

Southeastern Psychological Association, New Orleans LA, 2026

Department of Psychological Science