A.I. Increasingly Adept at Creating Images and Video
AI isn’t just for crunching data anymore. Machine learning has made great strides in analyzing diverse information, including images, but has struggled to generate realistic images on its own. Now recent strides have opened that new frontier, introducing a whole new level of accuracy, realism, to the mix – as well as potential problems.
One of the most exciting developments is an algorithm from members of Google’s DeepMind AI team. That algorithm, called BigGAN (Generative Adversarial Network), is not dissimilar from other GANs used in machine learning. Like most GANs, it leverages two models – a generative network that creates random images within masses of real images, and a discriminative network that tries to locate them. GANs are powerful, but typically limited, neural nets that require larger numbers of images in one category at a time – faces, trees, cats – in order to be effective. But backed by the computational resources of one of the world’s biggest companies, BigGAN could learn from a giant database of 14 million images in various categories at one time. The result was images that displayed an unprecedented level of realism in unparalleled variety.
AI’s newfound creativity doesn’t just extend to still images. UC Berkeley computer science researchers announced in August that they had developed a series of algorithms capable of taking video of dance moves and transposing them from one person to another – a process similar to the CGI techniques used by Hollywood to create computer-generated characters like Gollum from actual movements. Wired describes the software in action: Using 20 minutes of video, “One part of the software extracts the body positions from both clips; another learns how to create a realistic image of the subject for any given body position. It can then generate video of the subject performing more or less any set of movements.”
The implications are both exciting – AI-generated art! New, powerful creative tools! – and concerning. Software called Deepfakes, which originated via an anonymous programmer on Reddit in 2017, uses GANs to swap faces in videos. It has been used in everything from art to porn; its versatility has prompted fears that it (or similar software) could be used for more nefarious purposes.
Those fears are slowly being realized. The Guardian reported on a video “…created by a Belgian political party, Socialistische Partij Anders, or sp.a, and posted on sp.a’s Twitter and Facebook,” that depicted Donald Trump advising Belgians to withdraw from the Paris climate agreement. It drew intense negative reaction, leading sp.a to clarify that it was not, in fact, a real video. In a political climate of election interference by foreign powers, deep distrust between parties, and the term ‘fake news’ being bandied about with regularity, there is justifiable concern that algorithm-derived videos could cause real problems.
Tim Hwang, director of the Harvard-MIT Ethics and Governance of Artificial Intelligence Initiative, thinks that day is not here quite yet. He argues that amateurs are better spent using simpler means, telling The Guardian, “If you are a propagandist, you want to spread your work as far as possible with the least amount of effort,” he said. “Right now, a crude Photoshop job could be just as effective as something created with machine learning.” But that time may be coming. “…In the past, if you wanted to make a video of the president saying something he didn’t say, you needed a team of experts,” said Hwang. “Machine learning will not only automate this process, it will also probably make better forgeries.”
AI is becoming steadily more powerful, easier to use, and more effective. The breakneck pace of innovation is exciting, but with that revolution comes big questions that will need to be answered. It’s not a question of if we will see AI-created images and video on a wide scale – it’s a question of when.
Looking for Fresh Investment Ideas? See How Algorithms and A.I. Can Help
Want to invest and/or diversify your portfolio but unsure where to start? Artificial Intelligence can help! Tickeron has developed user-friendly Artificial Intelligence tools to help new and experienced investors generate investment ideas. Tickeron’s A.I. is capable of evaluating a portfolio and providing a “Diversification Score,” to tell the user how well-diversified their portfolio is. It can also generate investment ideas for a user’s 401(k) plan – even if you’re just getting started! The A.I. will give you ideas based on your risk tolerance, investment objectives, and the investment options available.
Tickeron’s new financial website is available to beginners, intermediate investors, and even experts and advisors. Explore tickeron.com today.
Contributor
Financial analyst and market blogger with expertise in equity research, fundamental analysis, and macroeconomic trends. I regularly publish coverage on individual stocks, ETFs, and sector developments — combining rigorous financial analysis with clear, engaging writing for a broad investment audience.
Advertisement