The latest round of far-out AI advances
Looking for ways AI can innovate your business and inspire your team? Here are some of the latest ways data scientists are using machine learning, computer vision and other AI to help us live on alien planets, watch better movies and even (maybe) smell better.
1. 'AI Anchors' who read the news
Can a bot be as engaging than your favorite cable news anchor? Not just yet. But China’s state-run Xinhua news agency is giving it a try anyway. They’ve paired up with tech firm Sogou to create a puppet-like newsreader
(modeled after a real news anchor) that matches lip movements and facial expressions to the stories it reads aloud.
Sogou isn’t revealing exactly which AI technologies they used to develop their anchors, but it looks like they’re using a combination of machine learning, speech synthesis tech and some facial modeling and recognition software.
Why it could work: Xinhua says that using AI anchors saves money, especially during breaking 24-hour news cycles.
Why it might fail:
Like many android animations, the anchors have some underlying uncanny valley
issues that may prevent them from gaining popularity.
2. AI-designed fragrances
Smell ya later, Old Spice: An algorithm might design your next favorite fragrance. Computer scientists have paired up with fragrance manufacturer Symrise to create Philyra, a system that uses machine learning algorithms to sift through hundreds of thousands of formulas and raw materials, comparing it with historical data to identify potentially appealing combinations.
Named after the shape-shifting Greek goddess of perfume, Philyra
has already developed two fragrances aimed at Brazilian millennials (see how specific it is?) with more to come.
Why it could work: Symrise is a 200-year-old company with a treasure trove of data for machine learning to mine; the program could unearth some great, unusual scents that people might never think of.
Why it might fail: In the end, machines don’t have noses. An experienced perfumer can identify something intangibly appealing that an algorithm might miss.
3. An AI physicist for alien worlds
A pair of researchers at MIT has developed an “AI physicist
” that can theorize about the physical laws of imaginary universes. The machine learning involved is much more complicated than, say, teaching an algorithm to recognize an elephant by feeding it thousands of different pictures of elephants.
For physics, researchers need partition data into smaller subsets; this “divide and conquer” approach provides simpler theories that the AI can remember and apply to future scenarios. Next, researchers created 40 mystery worlds governed by laws of physics that vary from one location to another; the AI was able to figure out 90 percent of their new physical laws.
Why it could work: This subdivided machine learning methodology also has implications in our own world, such as better understanding the complex data sets of climate change.
Why it might fail: The AI struggles to understand the laws of physics when the environments become more complicated - which another planet would be.
4. Machine movie critics
For a while now, the movie industry has enthusiastically used AI techniques to improve their bottom line, such as using machine learning to dig through movie scripts
and look for future box office successes.
Now, by looking at the trailers of hit movies from the past, 20th Century Fox is using computer vision to analyze movie trailers
and predict what films people will want to see. Fox paired up with Google’s Advanced Solutions Lab to create Merlin Video
, which analyzed the trailer for the movie Logan to isolate elements that its target audience might respond to.
Why it could work: Merlin correctly identified some other superhero movies that shared the same audience as Logan, such as Doctor Strange and Batman vs. Superman.
Why it might fail: Some of the patterns the AI noticed in Logan (such as “forest” or “light”) matched up to the trailer of a very different movie: Tarzan. AI still trails behind the expertise of movie buffs for now.