Applications and Challenges of Generative AI

Generative AI can employ deep-learning models to create new content, including text, images, audio, videos, and software code, based on the data it receives and that which it was trained on. Nowadays, people can utilize various forms of generative AI to fit specific domains and tackle diverse tasks. However, given the diverse output it can produce, ensuring reliability has currently become a crucial issue, with new platforms ardently seeking to avoid the abundant and often absurd malfunctions that were once nearly ubiquitous in AI operations. Below are some popular generative AI applications and challenges related to building trust in generative AI.

Applications of Generative AI

Text: ChatGPT, Gemini, Claude

As the foundation for many generative AI models, text-based platforms are regarded as the most advanced and reliable. These tools can provide contextual answers to given topics and information based on training datasets. They can also can automatically generate scenarios, poems, reports, and more.

Image: Leonardo AI, Midjourney, DALL-E

Image generation models can generate realistic or creative images from text, sketches, or other pictures. They create new visuals by using computer vision, machine learning, and deep learning techniques. Midjourney provides images from text input.

Music and Audio: AIVA, SOUNDFUL for music, ElevenLabs, TypeCast for speech

Music generators create new pieces by learning musical elements, such as sheet music, rhythm, or melody. Audio generators are also often used in text-to-speech, allowing them to create more human-like voices.

Video: RUNWAYML, Kaiber, Pictory

By utilizing a combination of various AI technologies, such as image generation, audio synthesis, and natural language processing, video generators can turn text, videos, photos, and music into stunning videos. These models can be used for making movies, advertising, video games, etc.

Challenges of Generative AI

Data Dependency: Generative AI models are heavily reliant on huge amounts of data for being trained. The training data may introduce biases, inconsistencies, and inaccuracies. Moreover, they can reinforce stereotypes, spread misinformation, and result in output that does not always reflect the real world. For example, Amazon abandoned its AI-driven hiring project in 2017, due to the AI recruiting tool showing biases against female applicants. Google temporarily paused Gemini’s people-generating feature in February after facing criticism from consumers over inaccuracies in depictions of historical figures including altered groups like Nazi-era German soldiers appearing as people of color.

Privacy Concerns: AI data collection can violate privacy and increase the risk of fraud and cybercrime by enabling AI tools to infer personal details and create fake media. For example, deepfake technology can manipulate images and videos, leading to identity theft, misinformation, or defamation. In response to growing concerns about this issue in South Korea, the Ministry of Education on Aug. 28 pledged to take a strong stance against deepfake pornography crimes in schools.

Legal Issues: Generative AI models may unintentionally infringe on copyright by using intellectual property without permission. This can lead to legal disputes, as distinguishing AI-generated output from original content can be difficult presenting challenges in the determination of responsibility and accountability. On Dec. 27, 2023, The New York Times company sued OpenAI and Microsoft for the unauthorized use of their articles to train ChatGPT.

By Kim Young- Ji, Editor