1. Moving Towards Storing Data in DNA
A startup called Catalog Technologies has built a prototype system for storing data in DNA more cost effectively than previous experimental methods. Researchers have demonstrated that it is possible to convert data into a series of nucleotides in synthetic DNA, however this process is slow, laborious, and expensive. Catalog’s system relies on a school-bus-sized machine that encodes data by combining short, premade strands of DNA into longer stands, which Catalog says is more cost effective and significantly faster rather than synthesizing entire strings from scratch. Catalog expects their system to be able to write one terabyte worth of DNA per day by the time it is complete next year.

2. AI Can Create Realistic Images From Scratch
Researchers at Deepmind have developed an AI system called BigGAN that can generate realistic looking images from scratch. BigGAN uses a machine learning technique called generative adversarial networks that has one neural network generate an image while another neural network evaluates it, creating a feedback loop to make the image as realistic as possible. Image generation with generative adversarial networks is not new, however the researchers were able to make significantly more realistic-look images than previous efforts by relying on large amounts of computing power, allowing their models to be more complex.

3. Monitoring the Environment with Machine Learning
Researchers at Stanford University have developed a machine learning system that could help environmental authorities prioritize investigations into agricultural facilities most likely to be at risk of illegally polluting. Federal and state environmental agencies are responsible for regulating over 300,000 facilities but do not have the resources to inspect even 10 percent of them per year. The researchers developed their model using historical inspection data to predict whether a facility is likely to fail an inspection based on characteristics like location, facility characteristics, and inspection history. Using this method, the researchers claim inspectors could identify two to seven times as many violations as existing approaches.

4. Thought-Reading AI Can Help Quadriplegics
Researchers at Ohio State University and applied science company Battelle have developed a system that uses brain implants and AI to help a quadriplegic man control his hands. The researchers implanted electrodes in an area of the man’s motor cortex that controls his right hand that could send signals to a computer based on the brain activity they detect. Then, a machine learning system translates millions of these data points into movements and controls a robotic sleeve that can move the man’s arm and hand accordingly. The system is effective enough to allow the man to play the video game Guitar Hero.
 
5. Farming Without Human Workers
Robotics company Iron Ox has opened an 8,000 square foot hydroponic farming facility in San Carlos, California that can produce leafy greens five times more productively than a traditional farm thanks to automation. Software developed by Iron Ox called “The Brain” monitors data throughout the farm, such as temperature and nitrogen levels, and directs human workers and systems of robotic arms and movers to tend to the crops.

6. Building Smart 3D Printers to Build Sturdier Parts
The U.S. Office of Naval Research (ONR) has launched an initiative to develop 3D printers that use AI to learn and make decisions about how to print objects that can hold up to extreme stress, such as parts for use in spacecraft and airplanes. Traditional manufacturing is subtractive, meaning large component materials are cut down to size and combined, whereas 3D printing is additive, meaning component materials are constructed in layers. This means that the structure of 3D-printed materials on the molecular level could be substantially different than subtractively manufactured materials, significantly affecting performance. ONR’s initiative will attempt to develop printers that use AI to print objects and autonomously alter their structure to achieve the same performance specifications as traditionally-manufactured parts.

7. Fitbit Data Helped Catch a Killer
San Jose Police have charged a man named Anthony Aiello with the murder of his stepdaughter with the help of data from her wearable Fitbit fitness tracker that pinpointed her time of death. Aiello said his daughter was alive and with him at 3:20 pm the day she died and claimed that he was not with her when she died. However, data from her Fitbit indicated her heart rate spiked significantly at 3:20 pm before stopping at 3:28. This fact, combined with other forensic evidence, was enough for authorities to determine Aiello was his stepdaughter’s killer.

8. AI Could Help People Find Jobs
Baidu has developed a machine learning system that can interpret job seekers’ resumes and pair them with job postings that ask for the skills they have. The system, called the Person-Job Fit Neural Network, analyzes language in resumes to identify what skills a person has, such as by linking the phrase “product development procedure” with project management experience, and identifies jobs in a database that require those skills.

9. Predicting Cognitive Decline with AI
Neuroscientists at the University of Toronto have developed an AI system that can analyze genetic, clinical, and MRI data to predict whether a person’s cognitive faculties will decline, leading to Alzheimer’s disease, within five years. The scientists trained their system on data from over 800 people with no, some, or advanced cognitive impairment due to Alzheimer’s disease. Detecting cognitive decline early to provide early treatment is the most effective way to delay or reduce the symptoms of Alzheimer’s disease.

10. Another Quantum Computing Service Goes to the Cloud
Canadian quantum computing company D-Wave has launched a service called Leap that allows users to access quantum computing via the cloud in real time. Users can develop quantum computing algorithms and use Leap to process the equations and send back the results. D-Wave estimates that one minute of compute time using Leap would be enough for a user to run between 400 and 4,000 jobs.