Opex AI Roundup — January 2019
by Gabriella Runnels and Macon McLean
The Opex AI Roundup provides you with our take on the coolest and most interesting Artificial Intelligence (AI) news and developments each month. Stay tuned and feel free to comment on any you think we missed!
___________________________________________________________________
AI from Start to Finnish
While America has trouble funding its public schools, Finland has somehow managed to create and administer a free online AI education course for its citizenry.
The program is designed to educate the nation on meaningful practical applications of artificial intelligence. First unveiled in May of 2018, this program was released as a massively open online course (MOOC) targeted toward Finnish residents but available to anyone. It’s now being rolled out nationally, and after commitments from 250 Finnish companies to train their staffs with this course, over 1% of the entire population of Finland will participate.
After the precipitous decline of former national corporate powerhouse Nokia, Finland’s profile in the tech community suffered, and their economic outlook became murkier as well. Understanding that Finland doesn’t have the resources of a world superpower to compete in pure AI research, the course’s focus on real-world implementation of AI is designed to give them a corner of the AI world they can call their own, as well as serve business needs in both the short and long term.
Dog the Balance Hunter
I don’t know about you, but my robot dog doesn’t seem to walk normally at all. Thank goodness for the bright minds at the Swiss Federal Institute of Technology in Zurich, who have recently pioneered a reinforcement learning-based method to help teach robot dogs to learn from their locomotive mistakes without any manual intervention.
Training these policies on real, nuts-and-bolts robots is expensive due to the high cost of equipment and reduced ability to iterate quickly. However, this team was able to train a reinforcement learning policy in simulation, and then transfer the model to the real robot with excellent results.
Some speculative uses for these dog-like robots include serving as manual labor in difficult terrain, inspecting unsafe underground caves and tunnels, destroying the Rebel base on Hoth, and haunting my dreams.
Something Wiki This Way Comes
With 6,000 pages and 550 chapters, Machine Learning — The Complete Guide is one large textbook. Try to cite it for a high school research paper, though, and your teacher may not be pleased — this book is a Wikibook, composed entirely of Wikipedia articles. This “complete guide” is an incredibly thorough resource on all things machine learning, but given its size, it’s a bit impractical to read.
That’s why researchers from Ben-Gurion University of the Negev decided to build an algorithm that would use artificial intelligence to automatically pare down the textbook to a reasonable length. Objectively evaluating the quality of the final version is difficult, since editing is a somewhat subjective process. Maybe future versions of this book will include a chapter on best practices for evaluating textbook-editing algorithms, but for now, the results seem pretty promising.
Why So Spurious?
The scientific method can be reduced to a few simple steps: ask a question, form a hypothesis, gather data, analyze the results, and draw a conclusion. This process has successfully guided much of humankind’s inquiry into the natural word for hundreds of years. However, with the recent explosion in the volume and velocity of data generated in nearly every imaginable field, the asking and answering are often flipped. “It is appealing to form the hypothesis based on the data,” argues UNC Chapel Hill statistics professor Kai Zhang. “The order of building the hypothesis and seeing the data has reversed.”
Zhang warns that relying too much on the data to ask the question can lead to finding spurious correlations disguised as genuine relationships (one such spurious correlation suggests that the divorce rate in Maine is linked to margarine consumption, and alternate would-be associations frequently include “number of Nicolas Cage films released by year” as a correlate). He also calls out the tendency of scientists, however well-meaning, to “manipulate the data to produce the most publishable result.” With enough effort, a data scientist could probably contort a sufficiently large dataset to “prove” any number of wildly untrue hypotheses. Our unprecedented access to huge amounts of data certainly gives us power, but it’s a power we must use responsibly.
Dropping the Dropout Rates
Student retention is a struggle for most American universities, with dropout rates now twice as high as they were in the sixties. Georgia State University in Atlanta is no exception, where nearly half of all students fail to graduate within six years. While most other colleges have resigned themselves to these bleak statistics, Georgia State has decided to take action.
A few years ago, GSU began tracking and examining hundreds of different student metrics, including factors as wide-ranging as class attendance and debt payments. They’re using this data in predictive models to identify the students who are most at risk of attrition, and when a student pops up on their radar, an adviser on staff personally reaches out to provide help. The program has reportedly “… improved outcomes with low-income, first-generation and minority students,” which is a significant achievement given that students belonging to these groups are some of the most likely to withdraw from school.
That’s it for this month’s roundup! Check back in February for more of the most interesting news and developments in the AI community (from our point of view, of course). Happy New Year!
_________________________________________________________________
If you liked this blog post, check out more of our work, follow us on social media (Twitter, LinkedIn, and Facebook), or join us for our free monthly Academy webinars.
