Technology

MIT researchers trained an artificial intelligence using Reddit and you won&t believe what happened next. Just kidding. Of course you will. The worst things happened.

Norman, who naturally gets his name from the guest-murdering proprietor of the Bates Motel, is the &worldfirst psychopath AI,& according to its creators at MIT. First, possibly, but certainly not the last. The creation is a kind of thought experiment designed to explore how the data we use to train machine learning algorithms ultimately influences its behavior.

In this case, the deep, endless well of human misery that is the internet was used to teach poor psychotic Norman the ways of the world.

&Norman suffered from extended exposure to the darkest corners of Reddit,& the researchers state, &and represents a case study on the dangers of artificial intelligence gone wrong when biased data is used in machine learning algorithms.&

In particular, the team used &an infamous subreddit […]that is dedicated to documenting and observing the disturbing reality of death.& That information had a fairly profound impact on Normangig captioning photos (inkblots, in this case), when compared to neural networks trained on more standard data.

Here are a couple of those responses:

You get the gist, right The rest can be found over here.

&When people say that AI algorithms can be biased and unfair,& the team explains, &the culprit is often not the algorithm itself, but the biased data that was fed to it.&

M17 Entertainment, a Taipei-based live streaming and dating app group, priced its IPO this morning on the NYSE and was expected to open trading today according to their final press release. But with just a little more than two hours to go before market closing, itstill not trading, and no one seems to know why.

An interview I had scheduled with the CEO earlier this afternoon was canceled at the last minute, with the companyrepresentative saying that M17 couldn&t comment since its shares were not yet actively trading, and thus the company remains under an SEC-mandated quiet period.

M17 has had a rocky non-debut so far. Originally targeting a fundraise of $115 million of American Depository Receipts (shares of foreign companies listed domestically on the NYSE), the company concluded its roadshow raising less than half of its target, for a final investment of $60.1 million. The company priced its ADR shares at $8 each, with each ADR representing 8 shares of the stockClass A security.

My colleague Jon Russell has covered the companyrapid growth over the past three years. It was formed from the merger of dating app company Paktor and live-streaming business 17 Media. Joseph Phua, who was CEO of Paktor, became CEO of the joint M17 company following the merger. Together, the two halves have raised tens of millions in venture capital.

M17 provides live-streaming and dating apps throughout &Developed Asia&

The companymain product is a live-streaming product where creators can build their fan bases and brands. Fans can purchase virtual gifts to send to their favorite artists, and those points are proving to be extraordinarily lucrative for the company. The company, according to its amended F-1 statement, has seen tremendous revenue growth, netting $37.9 million of revenue in the first three months of this year. The company has also been able to attract more live-streaming talent, increasing its contracted artists from 999 at the end of December 2016 to 7,719 at the end of March this year.

Thatwhere the good news ends for the company. Despite that revenue growth, operating losses are torrential, with the company losing $24.8 million in the first three months of this year. The company in its statement says that it has $31.4 million in cash and cash equivalents, giving it limited runway to continue operations without a strong IPO debut.

User growth has been mostly stagnant. Active monthly users has increased from 1.5 million to 1.7 million between March 31 of 2017 and 2018. What the company has succeeded in doing is monetizing those users much better. The percentage of users paying on the platform has more than doubled over the same time period, and the value of those users has increased more than 40 percent to $355 per user per month.

The big challenge for M17 is revenue quality. Live streaming represents 91.4 percent of the companyrevenues, but those revenues are concentrated on a handful of &whales& who buy a freakishly high number of virtual gifts. The companytop 10 users represent 11.8 percent of all revenues (that$447,220 per user in the first three months this year!), and its top 500 users accounted for almost a majority of total revenues. That concentration on the demand side is just as heavy on the supply side. M17top 100 artists accounted for more than a third of the companyrevenue.

That concentration has improved over the past few months, according to the companyfiling. But Wall Street investors have learned after Zynga and other whale-based revenue models that the sustainability of these businesses can be tough.

Finally, one complication for many investors wary of the increasing use of dual-class stock issues is the governance of the company. Phua, the CEO, will have 56.3 percent of the voting rights of the company, and M17 will be a controlled company under NYSE rules according to the companyamended filing. Class B shares vote at a 20:1 ratio with Class A share voting rights.

All of this is to say that while the company has had some dizzying growth in its revenue numbers over the past 24 months, that success is moderated by some significant challenges in revenue concentration that will have to be a top priority for M17 going forward. Why the company priced and hasn&t traded remains a mystery, and we have reached out for more comments.

Panda has built the next silly social feature Snapchat and Instagram will want to steal. Today the startup launches its video messaging app that fills the screen with augmented reality effects based on the words you speak. Say &Want to get pizza& and a 3D pizza slice hovers by your mouth. Say &I wear my sunglasses at night& and suddenly you&re wearing AR shades with a moon hung above your head. Instead of being distracted by having to pick effects out of a menu, they appear in real-time as you chat.

Panda is surprising and delightful. Italso a bit janky, created by a five person team with under $1 million in funding. Building a video chat app user base from scratch amidst all the competition will be a struggle. But even if Panda isn&t the app to popularize the idea, itinvented a smart way to enhance visual communication that blends into our natural behavior.

It all started with a trippy vision. Panda18-year-old founder Daniel Singer had built a few failed apps and was working as a product manager at peer-to-peer therapy startup Sensay in LA. When Alaska Airlines bought Virgin, Singer scored a free flight and came to see his buddy Arjun Sethi, an investor at Social Capital in SF. Thatwhen suddenly &I&m hallucinating that as I&m talking the things I&m saying should appear& he tells me. Sethi dug the idea and agreed to fund a project to build it.

Panda founder Daniel Singer

Meanwhile, Singer had spent the last 6 years FaceTiming almost every day. He loved telling stories with his closest friends, yet Applevideo chat protocol had fallen behind Snapchat and Instagram when it came to creative tools. So a year ago he raised $850,000 from Social Capital and Shrug Capital plus angels like Cyan (Banister) and SecretDavid Byttow. Singerset out to build Panda to combine FaceTimelive chat with Snapchatvisual flare triggered by voice.

But it turns out, &video chat is hard& he admits. So his small team settled for letting users send 10-second-max asynchronous video messages. PandaiOS app launched today with about 200 different voice activated stickers from footballs to sleepy Zzzzzs to a &-&%!#& censorship bar that covers your mouth when you swear. Tap them and they disappear, and soon you&ll be able to reposition them. As you trigger the effects for the first time, they go into a trophy case that gamifiesvoice experimentation.

Panda is fun to play around with yourself even if you aren&t actively messaging friends, which is reminiscent of how teens play with Snapchat face filters without always posting the results. The speech recognition effects will make a lot more sense if Panda can eventually succeed at solving the live video chat tech challenge. One day Singer imagines Panda making money by selling cosmetic effects that make you more attractive or fashionable, or offering sponsored effects so when you say &gym&, the headband that appears on you is Nike branded.

Unfortunately, the app can be a bit buggy and effects don&t always trigger, fooling you that you aren&t saying the right words. And it could be tough convincing buddies to download another messaging app, let alone turn it into a regular habit. Apple is also adding a slew of Memoji personalized avatars and other effects to FaceTime in its upcoming iOS 12.

Panda does advance one of technologyfundamental pursuits: taking the fuzzy ideas in your head and translating them into meaning for others in clearer ways than just words can offer. Itthe next wave of visual communication that doesn&t require you to break from the conversation.

When I ask why other apps couldn&t just copy the speech stickers, Singer insisted &This has to be voice native.& I firmly disagree, and can easily imagine his whole app becoming just a single filter in Snapchat and Instagram Stories. He eventually acquiesced that&Ita new reality that bits and pieces of consumer technology get traded around.I wouldn&t be surprised if others think ita good idea.&

Itan uphill battle trying to disrupt todaysocial giants, who are quick to seize on any idea that gives them an edge. Facebook rationalizes stealing other apps& features by prioritizing whatever will engage its billions of users over the pride of its designers. Startups like Panda are effectively becoming outsourced R-D departments.

Still, Panda pledges to forge on (though it might be wise to take a buyout offer). Singer gets that his app won&t cure cancer or &make the world a better place& as HBOSilicon Valley has lampooned. &We&re going to make really fun stuff and make them laugh and smile and experience human emotion& he concludes. &At the end of the day, I don&t think thereanything wrong with building entertainment and delight.&