Inside | Real news, curated by real humans
Inside AI

Inside AI (Nov 25th, 2019)

1. Google has pulled the plug on its controversial project that paid third-party contractors to gather selfies of darker-skinned people, often targeting homeless people without their full understanding. The stated goal of the program was to gather a more diverse dataset of selfies since many facial recognition systems have shown inherent bias against darker-toned faces. According to the New York Daily News, some of the contractors went to college campuses or places like the BET awards and felt pressured to gather as many faces as possible, at times rushing through the consent form process so some participants weren't aware of how their images would be used. (The participants were also paid with "measly" $5 gift cards, reports noted). Now, Google says it has axed the practice of hiring contractors and will only gather faces on company campuses. - FUTURISM

2. Data scientist and AI enthusiast Chintan Trivedi used NVIDIA's StyleGAN to predict the design of Tesla's Cybertruck, although it failed to capture the EV's edgy, futuristic design. According to a Medium article, Trivedi fed the generative-AI model StyleGAN with the design of a normal pickup truck and a Tesla Model X. He then combined that resulting truck design with an armored personnel carrier - per Elon Musk's description - to generate the result below. Earlier this year, NVIDIA introduced StyleGAN, which is mostly used to generate realistic-looking random human faces. - MEDIUM

3. By the Numbers: Online consumers say they still prefer humans to chatbots, according to a new Intercom survey. The firm conducted two polls about chatbots, one that surveyed 500 business leaders and another that questioned 500 U.S. consumers. Here are some of the findings:

  • Business leaders said they saved an average of $300,000 this year from their chatbots, with the greatest impact on support and sales teams.
  • 35 percent of the business leaders said that the AI bots helped them close sales deals.
  • Nearly 75 percent of consumers said they always expect to encounter a chatbot on a website.
  • For quick interactions, 87 percent said they still prefer humans to chatbots. - FORBES

4. During its University of Cambridge debate last week, IBM's Project Debater managed to convince members of the audience that AI will lead to more good than harm. Speaking in a female voice during Thursday's debate at the Cambridge Union, the robot provided opening statements for both sides and even cracked a few jokes, although it still struggled with saying its sentences correctly and backing up its assertions at times. The opening remarks were based on more than 1,100 human submissions gathered beforehand, which were analyzed via a “speech by crowd” application. In its anti-AI remarks, the bot argued that companies currently can't filter out human bias adequately in datasets and AI will take that bias and "fixate it for generations.” In advocating for the benefits of AI, Project Debater said the technology will create new jobs and “bring a lot more efficiency to the workplace." At the end, the "AI for good" team narrowly won with 51 percent of the audience vote. - NEW SCIENTIST

5. Lyft’s vice president of product for autonomous technology, Taggart Matthiesen, said the average Lyft ride rating of one of the company’s self-driving vehicles is a 4.97 out of five. Matthiesen said the average score is higher than that of human-driven Lyft vehicles, although the company’s PR team later said it did not disclose driver averages to the public. The company jointly operates autonomous vehicles in Las Vegas with Aptiv. Matthiesen said part of the reason behind the high score is the excitement from riders about autonomous driving technology, as well as the comfort of the luxury vehicle used in testing - the BMW 540i. Berkeley professor Avideh Zakhor, who specializes in AI, recently noted that ride-hailing companies have an advantage in that they've already collected years of data and are always amassing more, improving their machine learning algorithms. - BUSINESS INSIDER

A version of this story first appeared in Inside Automotive. Check out today's issue for more information about how you can enter to win a free Tesla Cybertruck.

6. A new AI algorithm has ranked the best ramen restaurants in Tokyo. TDAI Lab's Wise Review analyzed more than 4,000 ramen shop reviews in the city, ferreting out those that didn't have at least 100 Google reviews, as well as those written by reviewers who didn't have at least five posts. Wise Review eliminated what it surmised were unreliable reviews as well, such as those that imitate a product's description online. The winner was Aoshima Shokudou in Tokyo's Akihabara area, which ranked 13th on Google, followed by Google's top pick, Ramen Jiro Hibarigaoka in the western part of the city. - CNN BUSINESS

7. In a new blog post, Instagram engineers explained how the social media app utilizes machine learning models to suggest content for users. It's the first time that Instagram has gone into "heavy detail" about the foundational building blocks behind its personalized content, according to Instagram software engineer Ivan Medvedev. To relate and identify accounts that are similar to each other, the company adapted the common machine learning method known as word embedding. It also has a three-part ranking funnel, developed with a custom query language and modeling techniques, that "extracts 65 billion features and makes 90 million model predictions every second," according to engineers. - VENTURE BEAT

8. Google added Explainable AI services to its cloud platform as it aims to make machine learning models more transparent. As explained recently on the Google Cloud Blog, the company's AI Explanations - as it's called - can quantify each data factor’s contribution to a specific machine learning model, which can help data scientists or developers better understand why the AI reached certain conclusions. While it can provide information about the performance (or lack thereof) of various face- and object-detection models, it still has its limitations, according to Tracy Frey, Director of Strategy for Google Cloud AI, who noted that AI Explanations reflects "the patterns the model found in the data" but doesn't "reveal any fundamental relationships in your data sample, population, or application." - BBC

9. Walmart has acquired the Israeli AI startup Aspectiva for an undisclosed amount. The startup will remain at its headquarters in Tel Aviv as it joins Store No. 8, Walmart’s in-house incubation arm that's focused on developing more innovative shopping experiences. Aspectiva's AI technology analyzes what people are shopping for online and offers them help with their purchasing decisions, based on data like what other shoppers are saying about certain products. Since its founding in 2013, Aspectiva raised roughly $4 million from investors like KDDI and Global Brain out of Japan. - REUTERS

10. In a new book "Catfishing on CatNet," a sentient AI with a love for cats befriends a teenager in need of help, according to NPR reviewer Caitlyn Paxson. While the title and premise sound a bit goofy, Paxson called it an "engaging blend of tech thriller, mystery, and teen drama" that's focused around an all-powerful AI - named CheshireCat - that attempts to help a teenager after they meet on a website called CatNet. Paxson encourages readers to try and get over the name, writing that the book "taps lightly on the concept of personhood and the ethics of artificial intelligence, but it doesn't trouble itself profoundly. There's never any question of whether or not we're on board with CheshireCat being a person. They just are." - NPR

Written and curated by Beth Duckett in Orange County. Beth is a former reporter for The Arizona Republic who has written for USA Today, Get Out magazine and other publications. Follow her tweets about breaking news and other topics in southern California here.

Editor: Kim Lyons (Pittsburgh-based journalist and managing editor at Inside).

Subscribe to Inside AI