Here are five important AI stories from the week.
Led by Netflix and its recommendation engine prowess, more of the Hollywood film industry is relying on algorithms to decide who to cast and which movies stand to make the most money (and among which demographic and countries). Rather than just betting on a star actor or plot line, studios are now able to process lots of data on how movies have performed in the past to gauge which movies will be the best bet going forward. Among the front-runners in this Hollywood AI space are Cinelytic and ScriptBook.
The Dirty Little Secret Behind Google’s Duplex (The New York Times)
A year ago, Google introduced Duplex, a voice bot to call and book reservations at restaurants. Duplex dazzled the AI world with its near-human-like conversations skills, everything from the occasional pause to the umms that are so common in human speech. This article, however, calls into question just how advanced Duplex is, revealing that even Google relies very heavily on “humans in the loop” to achieve its restaurant booking service. Sometimes Google’s Duplex uses a bot; other times it relies on humans.
Data Moats By Themselves Are Not Sufficient (Andreesen Horowitz)
Companies with network effects, such as Facebook and LinkedIn, have thrived; for these businesses, the more users that join, the stronger the network becomes, increasing the company’s valuation. According to Andreesen Horowitz, data companies do not experience the same virtuous cycle; acquiring more and more data becomes harder the more data that you have. Acquiring and using more data has diminishing returns, not increasing returns like in the case of network effects. You only need data sufficient enough to solve the core business problem at hand. After that, time and effort are better spent on verticalization, go to market, and hiring world-class talent.
NLP to Solve Help-Desk Tickets (ZDNet)
NLP startup Moveworks rethinks how IT tickets should be resolved, leveraging natural language processing and automation to replace work that would typically have been dealt with by IT or help desk staff. The company leverages work by Google and the latest advances in using Transformers (a relatively new type of neural network architecture) to solve NLP problems.
A Deep Dive into OpenAI’s “Too Dangerous to Release” AI (FloydHub Blog)
A few months ago, OpenAI released its latest language model, GPT-2, which leveraged a massively large Transformer network. This excellent read goes deep into what makes the GPT-2 language model so special (and promising for the future of NLP).
More Stories Worth Reading and Watching…
25 Machine Learning Startups to Watch In 2019 (Forbes)
The Latest in Natural Language Processing (Medium)
25 Best Datasets for NLP (Lionbridge)
Almost Unsupervised Text to Speech and Automatic Speech Recognition (Microsoft)