Unleashing the generative capabilities of AI to enable individuals and organizations to express themselves better in different creative ways, even if they lacked the capabilities or manpower for it in the past
What if we told you this pitch is written by AI? It seemed the final frontier; where technology would automate our repetitive, mind-numbing tasks we would find our new forte as humans in creativity – an area where AI could never match us. Turns out that generative, creative AI systems produce increasingly spectacular results in areas as diverse as images, video, audio, text, art, products, medicines, games, program code, and test data ... the list is endless. When done well, AI becomes a powerful, inclusive technology, enabling many more people to express themselves effectively, raising both the individual and corporate Creativity Quotient (CQ). Now there’s a creative machine.
Viswanathan Rajeswaran Expert in Residence
- Generative, creative AI is based on the concept that – given enough training data and the right machine learning approaches – an AI system can not only detect patterns in said data, but can also produce new, synthetic ones out of the same.
- Generative Adversarial Networks (GANs) let two neural networks work together: the “generator,” attempts to produce realistic data, the “discriminator” assesses how plausible it is, and a feedback loop creates increasingly realistic, synthetic results.
- Auto-regressive language models such as GPT-3, Google Switch Transformer and Megatron-Turing build on hundreds of billions of parameters and huge ‘piles’ of internet text to generate convincing, high-quality text, including program code.
- Many creative AI systems are based on pre-trained models - they only need to be properly ‘prompted’ to generate results. Training creative models can therefore consume many computing resources, using them for creative purposes consumes much less.
- Generative, creative AI has a multitude of potential applications, from the design of software to interiors of houses and fashion, but also the creation of text, music, medicines, video, audio, books, art, and even test data.
- AI21 Labs’ online WordTune AI application uses generative language models to help people write better, richer, more varied pieces of text. It is currently exploring how to enable dyslexic people to express themselves more effectively.
- The open source GAN Zoo is a rapidly growing directory of all named GANs, covering application areas as diverse as text generation, medical imaging improvement, malware attack detection, even those creating passable works of art.
- Sogeti’s Artificial Data Amplifier (ADA) enabled a Swedish government agency to build and test systems and models on generated, synthetic personal citizen data, without ever having any real personally identifiable information involved.
- Iktos and Facio Therapies collaborate to apply Iktos’ AI-driven structure generation in one of Facio’s drug discovery programs, aiming to expedite the identification of potential pre-clinical candidates and to identify suitable novel chemical matter.
- The ability to deal with the increasing scarcity of human resources and lack of specialized skills thanks to augmentation by creative AI in generating, creating, and transforming all sorts of different content and assets.
- Inclusion of more people who can express their creativity, where they did not have the capabilities, skills or means to do so before. There is also the potential to unleash hidden creative power in the company’s (historical) datasets.
- Generating de-personalized, synthetic data from “real” data to address privacy, quality, fairness, bias and availability concerns of training and test data used within the organization.
- Exploring models, approaches and scenarios that would otherwise be too time-consuming or complex for humans to cover or to comprehend, for example in life science and other scientific research areas.