/* ---- Google Analytics Code Below */

Friday, February 12, 2021

Augmented AI for Creative Invention

Have we stopped getting return from  creative investment, and can AI change that? 

‘Augmented creativity’: How AI can accelerate human invention  By Adi Gaskell   in Venturebeat. 

In 2012, economist Robert Gordon published a controversial paper in which he argued that economic growth was largely over, due in no small part to our failure to maintain the engines of innovation in recent decades.

A study from the Stanford Institute for Economic Policy Research supported his general thesis and argued that while we’re spending even more money on creativity and innovation, our returns are flatlining. And this investment is not only in dollars, as the research revealed roughly 20 times as many people work in R&D today as did in 1930.

So what gives? Why has creating things become so difficult?  Researchers from Northwestern University attempt to answer this in a paper that shows a growing percentage of today’s creation is what’s known as recombination. Indeed, 40% of all patents in the U.S. Patent and Trademark Office are not completely new works, but rather mishmashes of existing ideas bolted together.

Artificial help

Finding effective ways to combine existing ideas is by no means easy, not least because of the growing volume of published material being produced. During the first few months of the COVID-19 pandemic, for instance, around 23,000 papers had been published on the virus, with that number doubling every 20 days.

The data science community at Kaggle pulled together to provide an AI-powered literature review in an effort to make sense of the deluge of new material. Data points were harvested from a subset of papers and grouped into 17 categories, with papers then listed for each category. It may not be the most polished approach, but that’s due to the time constraints imposed by the pandemic.

Researchers at Carnegie Mellon developed an alternate method: an AI-based approach to mining the patent and research databases for ideas that could be combined to form interesting solutions to specific problems. Their system uses analogies to help connect work from two seemingly distinct areas, which they believe makes innovation faster and a lot cheaper.  ... " 

Augmented creativity

What we’re witnessing is the emergence of something called “augmented creativity,” in which humans use AI to help them understand the deluge of data. Early prototypes highlight the important role humans can, and should, play in making sense of the suggestions proposed by the AI.

OpenAI attempted to replicate this approach with the release of a music-making tool called Jukebox. While the achievement is significant from a technological perspective, the results are unlikely to threaten the livelihoods of human musicians.

Various projects have also attempted to produce new and enticing recipes by using AI to mine food composition databases and concoct interesting combinations. For instance, Google researcher Sara Robinson recently showcased her system that produced a cake-cookie hybrid. Accenture researchers prototyped a similar recipe creation tool at their Dock facility in Dublin, but with stomach-churning results.

Smarter simulation

Most of these approaches utilize huge datasets that AI mines to look for well-established yet previously untapped connections. By using general adversarial networks (GANs), the next-generation models are capable of coming up with ideas without requiring access to the underlying logic.

For instance, researchers from Nvidia and the University of Toronto recently showcased a GAN that had been trained to simulate games by observing the screenplay alongside the actions of the human player in real time. The system was able to learn the best strategies simply from watching the gameplay as it unfolded and didn’t require any access to the game logic whatsoever.

“In addition, GameGAN is able to disentangle static and dynamic components within an image, making the behavior of the model more interpretable and relevant for downstream tasks that require explicit reasoning over dynamic elements,” the researchers explain. “This enables many interesting applications, such as swapping different components of the game to build new games that do not exist.”

No comments: