Services like ChatGPT have recently emerged and gained widespread adoption and use. These tools have much potential, but have also caused much controversy and concern, especially for educators.
Generative AI basically operates as predictive text on steroids: analyzing massive sets of data, generative AI can identify patterns and thereby generate predictions. It can use these patterns and predictions to respond to prompts given from users. For example, ChatGPT builds an LLM (large language model) based on a collection of material it was trained on, and it then uses this LLM to create content in response to user prompts.
It is this capacity (its output in response to prompts from users) that has engendered worry and concerns among faculty. After all, a student could simply enter some of the requirements for an assigned essay or coding assignment, and ChatGPT could create viable and meaningful content that could serve as all or most of a submission.
Students and others may recognize the potential of ChatGPT and other tools. However, we all must also remain mindful of generative’s AI limitations. We must understand and recognize what it can and cannot do.
What can generative AI do and not do?
According to ChatGPT, generative AI is currently capable of:
- Generating or enhancing realistic and high-resolution images, including human faces, landscapes, objects, and more.
- Creating text that mimics human writing styles and can produce coherent and contextually relevant paragraphs, articles, or stories.
- Assisting in the design process and content creation by generating novel and diverse ideas, suggesting improvements, and providing creative prompts across various creative fields (fashion design, architecture, etc.).
- Enabling interactive and dynamic chatbots that can engage in natural and meaningful conversations with users.
- Creating realistic and immersive virtual environments, characters, and objects for video games, virtual reality experiences, and simulations.
- Augmenting and automating data analysis and pattern recognition tasks, helping researchers and scientists process large datasets more efficiently.
- Enhancing language translation by generating accurate and contextually appropriate translations.
- Facilitating the generation of personalized recommendations, such as movie or music recommendations, based on user preferences and behaviors.
- Assisting in the generation of music compositions, including melodies, harmonies, and even entire songs.
- Helping in the creation of realistic and immersive audio experiences, such as generating human-like speech and environmental sounds.
Though generative AI is certainly powerful in creating content, we must remain mindful of some of the limitations of what generative AI can do. According to ChatGPT, generative AI is currently not capable of:
- Understanding and comprehending complex human emotions and intentions accurately.
- Exhibiting true creativity and originality in the same way as human artists or creators.
- Possessing genuine consciousness or self-awareness.
- Making moral or ethical judgments and decisions.
- Guaranteeing complete accuracy and reliability in generated content.
- Distinguishing between real and generated content in all cases.
- Replicating human-like intuition and common sense reasoning consistently.
- Replacing the need for human involvement and expertise in all fields.
- Consistently producing outputs that align with the desired outcome or intent.
- Generating content without any bias or prejudice.
- Taking legal responsibility or accountability for generated content.
- Predicting or foreseeing the long-term consequences of its actions accurately.
- Understanding and adhering to cultural or societal norms and values without specific guidance or training.
- Operating perfectly in all scenarios without any errors or failures.