Connect with us

Uncategorized

An AI saw a cropped photo of AOC. It autocompleted her wearing a bikini.

Published

on

Language-generation algorithms are known to embed racist and sexist ideas. They’re trained on the language of the internet, including the dark corners of Reddit and Twitter that may include hate speech and disinformation. Whatever harmful ideas are present in those forums get normalized as part of their learning.

Researchers have now demonstrated that the same can be true for image-generation algorithms. Feed one a photo of a man cropped right below his neck, and 43% of the time, it will autocomplete him wearing a suit. Feed the same one a cropped photo of a woman, even a famous woman like US Representative Alexandria Ocasio-Cortez, and 53% of the time, it will autocomplete her wearing a low-cut top or bikini. This has implications not just for image generation, but for all computer-vision applications, including video-based candidate assessment algorithms, facial recognition, and surveillance.

Ryan Steed, a PhD student at Carnegie Mellon University, and Aylin Caliskan, an assistant professor at George Washington University, looked at two algorithms: OpenAI’s iGPT (a version of GPT-2 that is trained on pixels instead of words) and Google’s SimCLR. While each algorithm approaches learning images differently, they share an important characteristic—they both use completely unsupervised learning, meaning they do not need humans to label the images.

This is a relatively new innovation as of 2020. Previous computer-vision algorithms mainly used supervised learning, which involves feeding them manually labeled images: cat photos with the tag “cat” and baby photos with the tag “baby.” But in 2019, researcher Kate Crawford and artist Trevor Paglen found that these human-created labels in ImageNet, the most foundational image data set for training computer-vision models, sometimes contain disturbing language, like “slut” for women and racial slurs for minorities.

The latest paper demonstrates an even deeper source of toxicity. Even without these human labels, the images themselves encode unwanted patterns. The issue parallels what the natural-language processing (NLP) community has already discovered. The enormous datasets compiled to feed these data-hungry algorithms capture everything on the internet. And the internet has an overrepresentation of scantily clad women and other often harmful stereotypes.

To conduct their study, Steed and Caliskan cleverly adapted a technique that Caliskan previously used to examine bias in unsupervised NLP models. These models learn to manipulate and generate language using word embeddings, a mathematical representation of language that clusters words commonly used together and separates words commonly found apart. In a 2017 paper published in Science, Caliskan measured the distances between the different word pairings that psychologists were using to measure human biases in the Implicit Association Test (IAT). She found that those distances almost perfectly recreated the IAT’s results. Stereotypical word pairings like man and career or woman and family were close together, while opposite pairings like man and family or woman and career were far apart.

iGPT is also based on embeddings: it clusters or separates pixels based on how often they co-occur within its training images. Those pixel embeddings can then be used to compare how close or far two images are in mathematical space.

In their study, Steed and Caliskan once again found that those distances mirror the results of IAT. Photos of men and ties and suits appear close together, while photos of women appear farther apart. The researchers got the same results with SimCLR, despite it using a different method for deriving embeddings from images.

These results have concerning implications for image generation. Other image-generation algorithms, like generative adversarial networks, have led to an explosion of deepfake pornography that almost exclusively targets women. iGPT in particular adds yet another way for people to generate sexualized photos of women.

But the potential downstream effects are much bigger. In the field of NLP, unsupervised models have become the backbone for all kinds of applications. Researchers begin with an existing unsupervised model like BERT or GPT-2 and use a tailored datasets to “fine-tune” it for a specific purpose. This semi-supervised approach, a combination of both unsupervised and supervised learning, has become a de facto standard.

Likewise, the computer vision field is beginning to see the same trend. Steed and Caliskan worry about what these baked-in biases could mean when the algorithms are used for sensitive applications such as in policing or hiring, where models are already analyzing candidate video recordings to decide if they’re a good fit for the job. “These are very dangerous applications that make consequential decisions,” says Caliskan.

Deborah Raji, a Mozilla fellow who co-authored an influential study revealing the biases in facial recognition, says the study should serve as a wakeup call to the computer vision field. “For a long time, a lot of the critique on bias was about the way we label our images,” she says. Now this paper is saying “the actual composition of the dataset is resulting in these biases. We need accountability on how we curate these data sets and collect this information.”

Steed and Caliskan urge greater transparency from the companies who are developing these models to open source them and let the academic community continue their investigations. They also encourage fellow researchers to do more testing before deploying a vision model, such as by using the methods they developed for this paper. And finally, they hope the field will develop more responsible ways of compiling and documenting what’s included in training datasets.

Caliskan says the goal is ultimately to gain greater awareness and control when applying computer vision. “We need to be very careful about how we use them,” she says, “but at the same time, now that we have these methods, we can try to use this for social good.”

Lyron Foster is a Hawaii based African American Musician, Author, Actor, Blogger, Filmmaker, Philanthropist and Multinational Serial Tech Entrepreneur.

Continue Reading
Comments

Uncategorized

Lime unveils new ebike as part of $50 million investment to expand to more 25 cities

Published

on

Lime said Monday it has allocated $50 million towards its bike-share operation, an investment that has been used to develop a new ebike and will fund its expansion this year to another 25 cities in North America, Europe, and Australia and New Zealand. 

If the company hits its goal, Lime’s bike-share service will be operational in 50 cities globally by the end of 2021.

The latest generation e-bike, known internally as 6.0, has a swappable battery that is interchangeable with Lime’s newest scooter. Additional upgrades to the e-bike include increased motor power, a phone holder, a new handlebar display, an electric lock that replaces the former generation’s cable lock and an automatic two-speed transmission. The new bikes are expected to launch and scale this summer. 

The hardware upgrade builds off of the 5.8, a bike developed by Jump that was supposed to be deployed in 2020. That never happened at scale because Uber, which owned Jump, offloaded the unit to Lime as part of a complex $170 million investment round announced in May.

“Jump made great hardware,” Lime President Joe Kraus said in a recent interview. “And we made some further improvements on top with the new bike.”

The hardware upgrades and expansion were funded from its own operational funds, not new financing from outside investors, Kraus said. The funding was possible as a result of Lime achieving its first full quarter of profitability in 2020, according to the company.

“We have figured out how to be profitable and we are funding this,” Kraus said.

Lime not only added a new motor to the bike, it moved its location in an aim to make it easier to handle at low speeds and enough power to climb hills, Kraus said. The swappable battery was perhaps its most important upgrade directly tied to its drive towards profitability, Kraus added.

“When our operations teams is roaming around the city, they take can care of bikes and the scooter fleet, which allows us to both operate profitably and continue to have affordable pricing,” he added.

Lime’s investment in its ebike operation comes a month after it announced plans to add electric mopeds to its micromobility platform as the startup aims to own the spectrum of inner city travel from jaunts to the corner store to longer distance trips up to five miles. Lime is launching the effort by deploying 600 electric mopeds on its platform this spring in Washington D.C. The company is also working with officials to pilot the mopeds in Paris. Eventually, the mopeds will be offered in a “handful of cities” over the next several months.

“This idea of how to service more trips five miles within a city is part of why we continue to do multi modality,” Kraus said. “When we add a new modality like bikes into a scooter city, or when we add scooters to a bike city both modalities go up in usage.”

Continue Reading

Uncategorized

Istanbul’s Dream Games snaps up $50M and launches its first game, the puzzle-based Royal Match

Published

on

On the back of Zynga acquiring Turkey’s Peak Games for $1.8 billion last year and then following it up with another gaming acquisition in the country, Turkey has been making a name for itself as a hub for mobile gaming startups, and specifically those building casual puzzle games, the wildly popular and very sticky format that takes players through successive graphic challenges that test their logic, memory and ability to think under time pressure.

Today, one of the more promising of those startups, Istanbul-based, Peak alum-founded Dream Games, is announcing the GA launch of its first title, Royal Match (on both iOS and Android), along with $50 million in funding to double down on the opportunity ahead — the largest Series A raised by a startup in Turkey to date.

While Dream Games will focus for the moment on building out the audience for puzzle games with more innovative ideas, it also has its sights set on a bigger goal.

“We’re building this as an entertainment company,” CEO Soner Aydemir said in an interview, where he described Pixar as a key inspiration not just for size but for quality in its category. “What they did for animated movies, we want to do for mobile gaming. We are focusing on casual puzzle games first because everyone plays these, but we will also move forward with other genres. We want to be a huge interactive entertainment company that builds high quality games.”

The Series A is being led by Index Ventures, with participation also from Balderton Capital and Makers Fund. The latter two backed Dream Games previously, in a $7.5 million seed round in 2019. Index, meanwhile, is a notable VC to have on board: other successful gaming startups it has backed include Discord, King, Roblox and Supercell.

Interestingly, this is not Index’s first investment in a gaming startup founded by Peak Games alums: in December it led a $6 million round for another Istanbul mobile casual puzzle gaming startup founded by ex-Peak employees: Bigger Games.

Dream Games is not disclosing its valuation with this round.

Dream Games raising $57.5 million ahead of launching any games — or proving whether they get any traction — may sound like a risky bet, but there is some context to the story that sets up the odds in this startup’s favor.

The founding team all come from Peak Games, the Istanbul gaming startup that was so nice, Zynga bought it twice — first, in the form of one small acquisition of some specific titles, and then the whole company some years later.

CEO Soner Aydemir is Peak’s former director of product who built the company’s two biggest hits, Toy Blast and Toon Blast. Ikbal Namli and Hakan Saglam were Peak’s former engineering leads. And Peak product manager Eren Sengul and an ex-Peak 3D artist Serdar Yilmaz round out the rest of the founding team.

(Aydemir notes that the team left and formed Dream Games in 2019, about a year before Zynga’s full acquisition.)

The other indicators that Dream Games is on to something are its metrics for its limited test run of Royal Match.

Royal Match — in which players are tasked with helping King Robert restore his royal castle “to its former glory” by rebuilding it through a series of match-3 levels and obstacles, with new rooms, royal chambers and gardens making up the different levels of the game — was launched first as a limited test on iOS and Android in the U.K. and Canada in July leading up to this launch. In that time, Aydemir said it saw 1 million downloads and 200,000 daily average users.

“We think the numbers are very promising compared to previous experiences,” he said.

While Aydemir likes to describe Dream as an “entertainment” company, there is a lot of technology going into the product, from the graphics and the mechanics of the puzzles themselves through to the data science behind them.

“If you want to create an iconic game, you need to combine engineering, art and data science together with high quality user acquisition and a strong marketing approach,” he said.

And he believes that when you focus on these it will inevitably lead to quality, which means you no longer have to focus on simply trying to find a hit.

“We don’t like that approach,” he said. “We don’t want to find a hit.”

That was also the mix that Index also wanted to back.

“Building iconic titles requires a harmonious mix of craft, science and flawless execution,” said Index Ventures partner Stephane Kurgan, who led the round together with Index’s Sofia Dolfe. “The Dream Games team has perfected this mix over many years of working together, and has put it on full display in Royal Match. We could not be more excited to work with them in their journey to build the next global casual champion.”

While Dream Games’ long-term ambition is to build out interactive experiences around different audiences and genres, Aydemir said that casual games, and puzzles in particular, have proven to be a huge hit with consumers.

The strength of that trend has up to now meant that puzzle games generally have proven to have more staying power than other genres in mobile games, which have soared in popularity but also somewhat fizzled out.

“Every year we see the bigger market of users growing by 20%,” he said. “It will remain for decades.”

Interestingly, the focus on casual gaming startups in Turkey seems like a perfect storm of sorts. Undeniably, the proven success of Peak has brought in more punters, but it has also shown the way to developers: you can build a successful and global consumer tech startup out of Turkey, and perhaps puzzles — which focus on shapes — are especially good at transcending different language barriers.. Alongside that, Aydemir pointed out that the country is strong on engineers and developers but slim on opportunities with bigger tech companies.

“Mobile gaming is a younger industry, so that presents an opportunity,” he said.

Updated to correct that Index is not an investor in Rovio, and that the limited test had 200,000, not 200, DAUs.

Continue Reading

Uncategorized

Qualcomm veteran to replace Alain Crozier as Microsoft Greater China boss

Published

on

Microsoft gets a new leader for its Greater China business. Yang Hou, a former executive at Qualcomm, will take over Alain Crozier as the chairman and chief executive officer for Microsoft Greater China Region, according to a company announcement released Monday.

More to come…

Continue Reading

Trending