The Unreal #TechVision Enterprises are becoming architects of the then be shared with academic researchers and interact with, saving valuable time and effort 129 unreal world. As they push AI into more and other organizations. This allowed them and enabling new and unique experiences. collaborative and creative roles, they are blurring to innovate faster and in ways they never could 65% the lines between what’s real and what isn’t. have alone to create an algorithm that helps Consider OpenAI’s Codex – trained on both Advances in generative AI, and generative clinicians determine when patients should be code and prose – that helps programmers of global consumers lack 132 adversarial networks (GANs) in particular, are given drug treatment or sent to the ICU. write code in multiple languages. If asked in confidence in recognizing or making the creation and use of synthetic data plain English to create a bouncing blue ball, for identifying deepfake videos or that is incredibly realistic – while being unreal – American Express also uses synthetic data instance, it can create the code to do just that. synthetic content. possible. Gartner predicts that by 2030, most to improve their fraud detection – including Although Codex only produces the right code 128 of the data in AI models will be synthetic. more, and more unique, instances of fraud 37% of the time according to OpenAI estimates, 130 This has major implications for businesses. than normally occur to bulk up their algorithm. that doesn’t prevent it from being a valuable And MOSTLY AI is a company tackling AI bias partner for developers, saving them from having 131 First, synthetic data is being used to train by promoting fairness in synthetic data. to look up reference materials and providing AI models in ways that real-world data This means that if someone wants to train an blocks of code to build on. practically cannot or should not. This realistic AI model to learn that the same percentage yet unreal data can be shared, maintaining of women as men make over $50,000 a year, the same statistical properties while protecting because that’s what’s fair, then they can adjust confidentiality and privacy, and it can also the synthetic data to reflect that – and any If this isn’t convincing enough, a 2019 Accenture be made to have increased diversity and to models using the data won’t have that real-world survey found that 84% of C-suite executives counter bias, thus overcoming the pitfalls bias built in. believe they need to leverage AI to reach their of real-world data. growth objectives, and 75% believe that if they Second, synthetic data is being made realistic don’t scale AI in the next five years they risk For example, in the early days of the pandemic, in important ways too – to be more human- 127 going out of business entirely. Clearly, AI will Israel’s largest hospital used a platform created like for creation and interaction. As AI is be intrinsic to the future of business – if it’s by the Israeli startup MDClone to create democratized, it definitionally becomes more not already. synthetic COVID-19 patient data, which could human-like to make it easier for anyone to use Introduction // WebMe // Programmable World // The Unreal // Computing the Impossible 61