Emerging use cases for generative AI

Generative AI refers to machine learning models that can generate new content like text, code, audio, video, and images rather than simply classify or label existing data. Powerful generative algorithms are rapidly advancing, unlocking exciting new applications across industries while raising potential risks. As generative models become more capable and accessible, organisations in Australia and worldwide are exploring innovative ways to leverage these AI systems across diverse use cases.

Content Creation

One of the most popular emerging uses of generative AI is automating various forms of content creation to increase productivity.

Text generation - Models like GPT-3 can generate long-form articles, stories, reports, ads and other text content given just a prompt. This can significantly augment human writing. Tools like Copy.ai and Rytr utilise generative AI to translate core ideas into draft text content for blog posts, social media, emails and more based on custom prompts.

Art and design - DALL-E, Stable Diffusion and other image generation models can rapidly synthesise original digital artwork, images, logos, posters, book covers and illustrations from text prompts. This boosts visual content creation for marketing and advertising materials. Architects and interior designers also employ AI to generate design concepts.

Video - AI synthesis platforms like Synthesia, Metaphore and Hour One can generate custom video content using text prompts and a few images of a person. The AI generates lip-synced talking head videos, allowing fast video creation at scale without filming.

Audio - Models like ElevenLabs AI Voice and ReplicaStudios synthesise human-like voice from text to automate audio generation for dialogue, narration, podcasts and more. This expands creative possibilities in entertainment industries.

3D modelling - AI model Maker aims to revolutionise 3D content creation by enabling users to generate 3D scenes, objects, avatars and animations through text and sketch inputs. This simplifies and expands 3D digital worldbuilding for gaming, VR and the metaverse.

Personalisation and Recommendation

Leveraging user data and past behaviour, generative AI can deliver highly personalised content catered to individuals across applications:

Personalised shopping - Clothing companies like Stitchfix employ AI models to recommend apparel suited to customer style based on past preferences and transactions. Music and video streaming services like Spotify and Netflix use recommenders to suggest new media catered to user taste.

Dynamic product generation - Brands like Chubbies use generative AI to automatically generate numerous tailored clothing product variants based on core designs. This allows unique personalised products.

Individualised education - AI tutoring systems like Century analyse student strengths and weaknesses to generate personalized lessons and practice materials optimized for individual learners. They can adapt in real-time based on student responses.

Targeted advertising - Generative algorithms can create endless tailored ad combinations integrating elements like customer demographics, search keywords and purchase history data to display ads matching individual interests and needs.

Personalised news - Apps like SmartNews and Feedly use AI to curate and recommend news articles on topics of interest to each reader based on past engagement and readings rather than generic trending stories.

Conversational Agents

Advances in natural language processing enable generative AI to hold seemingly natural conversations:

Customer service chatbots - Models like Clara from Anthropic can engage directly with customers to handle inquiries, orders, technical support and more, providing 24/7 automated self-service.

Intelligent assistants - Alexa, Siri and Google Assistant incorporate generative algorithms to interpret speech, generate fluent responses, and complete tasks through conversational interactions.

Companion chatbots - Apps like Replika craft emotionally responsive conversational personas that offer users friendship and mental wellness support. However, transparency regarding their limitations is important.

Gaming NPCs - AI can generate dynamic dialogue and behaviour for non-player game characters tailored to each interaction, improving immersion and replayability. This assists open world and interactive narrative games.

While offering many conveniences, responsible design of conversational agents is crucial to avoid deception and ensure clarity that users are interacting with an AI.

Data Augmentation

Since gathering massive datasets can be challenging, generative models can artificially expand limited training data by creating plausible new synthetic samples:

Autonomous vehicle training - AI can generate millions of simulated traffic scenarios with appropriate labels to safely train self-driving algorithms beyond what real world data provides.

Synthetic patient data - Healthcare AI uses generative adversarial networks to create fictional labelled medical scans, lab tests, disease records and more. This improves model accuracy with diverse training data and preserves patient privacy.

Computer vision training - Models like Tencent's GauGAN paint realistic landscape images from segmentation maps that provide labelled training data to improve landscape classification.

Voice assistant training - Generative models synthesise large labelled datasets of diverse speech examples with varying accents, voices and backgrounds to enhance speech recognition.

Drug discovery data - AI generates molecular structures and compounds with desired chemical properties to expand datasets for training models that predict promising new drug candidates.

Code Generation

Automating coding stretches developer productivity and enables customised software:

Autocomplete - Models like TabNine, GitHub Copilot, and Kite provide intelligent code autocompletion to accelerate development workflows. The AI suggests context relevant syntax and entire function blocks as coders type.

Code generation - Services like Anthropic's Claude can translate natural language requests into running code. This allows rapid prototyping and MVP development with minimal manual coding.

Bug finding and repair - Startups like DeepCode analyse codebases to automatically flag bugs, security flaws, and suggest fixes. This automates code reviews and debugging.

Personalisation - AI can customise templates to generate tailored websites, analytics dashboards, reports, and data pipelines adapted to each user's specific needs and use case.

Maintenance - Legacy systems often lack documentation. AI models can analyse code to add comments, fill documentation gaps, and flag dependencies to ease maintenance.

While AI coding assistance shows much promise, developers must remain vigilant regarding any flaws and biases inherited from the model's training process. Augmented coding combines the flexibility of generative AI with human oversight.

In summary, rapid advances in generative models are opening up a wealth of possibilities across content creation, personalisation, conversations, data generation, and software applications. However, like any technology, it also carries risks if deployed without diligence. Maintaining transparency, evaluating bias mitigation, and keeping a human in the loop will be critical as generative AI sees increasing real-world adoption in the coming years.