All posts in “Startups”

Microsoft acquires conversational AI startup Semantic Machines to help bots sound more lifelike

Microsoft announced today that it has acquired Semantic Machines, a Berkeley-based startup that wants to solve one of the biggest challenges in conversational AI: making chatbots sound more human and less like, well, bots.

In a blog post, Microsoft AI & Research chief technology officer David Ku wrote that “with the acquisition of Semantic Machines, we will establish a conversational AI center of excellence in Berkeley to push forward the boundaries of what is possible in language interfaces.”

According to Crunchbase, Semantic Machines was founded in 2014 and raised about $20.9 million in funding from investors including General Catalyst and Bain Capital Ventures.

In a 2016 profile, co-founder and chief scientist Dan Klein told TechCrunch that “today’s dialog technology is mostly orthogonal. You want a conversational system to be contextual so when you interpret a sentence things don’t stand in isolation.” By focusing on memory, Semantic Machines’ AI can produce conversations that not only answer or predict questions more accurately, but also flow naturally.

Instead of building its own consumer products, Semantic Machines focused on enterprise customers. This means it will fit in well with Microsoft’s conversational AI-based products, including Microsoft Cognitive Services and Azure Bot Service, which are used by one million and 300,000 developers, respectively, and virtual assistants Cortana and Xiaolce.

Bail reform has a complex relationship with tech

On any given day in the United States, more than 450,000 people are behind bars awaiting their constitutionally mandated fair trial. None of them have been convicted of a crime — they’ve been accused of committing a crime, but no formal ruling of guilt or innocence has been made. That means these hundreds of thousands of people are incarcerated simply because they don’t have the financial means to post bail. 

Bail was originally designed to incentivize people to show up for their court dates, but it has since evolved into a system that separates the financially well-off from the poor. It requires arrested individuals to pay money in order to get out of jail while they await trial. For those who can’t afford bail, they wind up having to sit in jail, which means they may be at risk of missing rent payments, losing their jobs and failing to meet other responsibilities. 

Money bail is all too often a common condition to secure release from jail while a case is in progress. Cash bail systems result in leaving many people incarcerated, even though they haven’t been convicted of a crime. 

The cash bail system in the United States is one of the greatest injustices in the criminal justice system, ACLU Deputy National Political Director Udi Ofer tells TechCrunch. Bail reform, Ofer says, is a “key way to achieve” the goals of challenging racial disparities in the criminal justice system and ending mass incarceration. 

As we explored in “The other pipeline,” the criminal justice system in the United States is deeply rooted in racism and a history of oppression. Black and Latino people comprise about 1.5 million of the total 2.2 million people incarcerated in the U.S. adult correctional system, or 67 percent of the prison population, while making up just 37 percent of the total U.S. population, according to the Sentencing Project.

With a criminal justice system that disproportionately affects people of color, it’s no wonder why the cash bail system does the same. For one, people of color are 25 percent more likely than white people to be denied the option of bail, according to a pre-trial study by Dr. Traci Schlesinger. And for the black people who are given the option to pay bail, the amount is 35 percent higher on average than bail for white men, according to a 2010 study.

The national felony bail median is $10,000. For those who can’t afford it, they have to rely on bail bond agencies, which charge a non-refundable fee to pay the required bail amount on the person’s behalf. The bail bond companies, which are backed by insurance companies, collect between $1.4 billion and $2.4 billion a year, according to the ACLU and Color of Change.

And if bail bond companies are out of reach, those who are sitting in jail awaiting trial are more likely to be convicted of the crime they were charged with. The non-felony conviction rate rose from 50 percent to 92 percent for those jailed pre-trial, according to a study by the New York City Criminal Justice Agency. Along the way, leading up to the trial, some prosecutors incentivize people to plead guilty to the charges even if they’re innocent.

“It’s time to end our nation’s system of cash bail that lets the size of your wallet determine whether you are granted freedom or stay locked up in jail,” Ofer says. “Money should never decide a person’s freedom yet that’s exactly what happens every day in the United States.”

Pre-trial detention is also costly to local cities, counties and taxpayers. It costs about $38 million a day to keep these largely nonviolent people behind bars, according to the Pretrial Justice Institute. Annually, that comes out to about $14 billion to jail unconvicted people.

“The only people benefiting from bail is the for-profit bail industry,” Ofer said. “If we’re ever going to end mass incarceration in the United States, then we need to end cash bail.”

Bail reform is coming

Across the nation, bail reform has made its way into a handful of states. New Jersey’s bail reform law took effect last January; since then, its daily jail population has dropped 17.2 percent, and courts have imposed cash bail on just 33 defendants out of 33,400, according to the ACLU.

The ACLU itself is working on bail reform in 38 states, including California, where Ofer says he is optimistic reform will happen this year. Right now, a pre-trial release bill, Senate Bill 10, is up for consideration in the Assembly. The bill argues California should ensure people awaiting trial are not incarcerated simply because they can’t afford to pay bail. The bill also advocates for counties to establish pre-trial services agencies to better determine if people are fit to be released.

The bill, introduced by Senators Bob Hertzberg and others, is backed by the ACLU and Essie Justice Group, an Oakland-based organization that advocates for actual justice in the criminal justice system.

“Today we have a system that allows for people to be released pre-trial if they have enough money to afford their bail,” Essie Justice Group founder Gina Clayton tells TechCrunch. “Everyone else is required to sit inside of a cage without any way out.”

Essie Justice Group works mostly with and for women who have incarcerated loved ones. Often, the only way out for people is help from family or a plea deal, Clayton says.

“When we see people making the bail, we see that women are going into tremendous debt and are also beholden to an industry that has time and time again been cited and known to practice in quite an incredibly despicable way in terms of coercing and harassing their customers,” Clayton says. “When we think about who are the people who know about what’s going on with bail, it’s black and brown women in this country.”

For the past two years, Essie Justice Group held an action around Mother’s Day, with the goal of bailing moms out of jail or immigration detention. Last year’s action led to release of 30 women.

Photo via Essie Justice Group

Can tech help?

The short the answer is maybe. Earlier this month, Google banned ads for bail bonds services, which Clayton says is the largest step any corporation has taken on behalf of people who have loved ones in jail. But while tech can help in some ways, Clayton has some concerns with additional for-profit entities entering the criminal justice system.

“There are definitely tech solutions that I’m very against,” Clayton said, but declined to comment on which ones in particular. “I will say that my energy around this doesn’t come from an imagined place. I’m seeing it happen. One of the things we’re seeing is companies who are interested in bail reform because they see another opportunity to make money off of families. Like, ‘let this person out, but have them, at a cost, check in with people I hire to do this fancy but expensive drug testing three times a week, pay for an ankle shackle or bracelet and GPS monitoring.’ I think the companies that are making money off of those types of things are the ones we need to be wary of.”

There is, however, one for-profit company that immediately jumped to Clayton’s mind as being one doing actual good in the criminal justice space. That company is Uptrust, which provides text message reminders to people regarding court dates.

“I think that is a really great addition to the landscape,” Clayton says. “The reason I’m a proponent of theirs is because I understand their politics and I know what they won’t do, which is take it a step further or get involved with getting incentivized to add on bells and whistles that look less like freedom for people but more revenue for them.”

Uptrust, founded by Jacob Sills and Elijah Gwynm, aims to help people make their court dates. While the movies like to depict flight risks and people skipping town ahead of their court dates, failure to appear in court often comes down to a lack of transportation, work conflicts, not receiving a reminder, childcare or poor time management, Sills tells TechCrunch.

That’s where the idea came to humanize the system a bit more, by enabling public defenders to more easily connect with their clients. Uptrust is two-way in nature and reminds people on behalf of the public defender about court dates. Clients can also communicate any issues they may have about making it to court.

“If the public defender knows the client has an issue, they can usually get court moved,” Sills says. “But if they don’t have the information, they’re not going to lie on behalf of clients.”

Because public defenders don’t make much money, Uptrust doesn’t charge very much, Sills says.

“But they really care about the client and one of the things we saw with this was we needed to change the whole front end of the system to be less adversarial and more human,” Sills says.

In addition to text reminders, Uptrust enables public defenders to assist with other needs clients may have.

“A lot of stuff around bail reform is around risk assessment rather than need assessment,” Sills tells me. “But we saw a lot of these individuals have needs, like helps with rides, child care or reminders.”

Public defenders who are invested in the care of their clients can remind them via Uptrust to do things like ask for time off work or schedule child care.

For the end-user, the client, Uptrust is all text-based. For the public defenders, Uptrust offers a software solution that integrates into their case management systems.

Since launching in the summer of 2016 in California’s Contra Costa County, the court appearance rate improved from 80 percent to 95 percent, Sills says. To date, Uptrust has supported 20,000 people with a five percent FTA rate.

“As we improve product, if we can get [the FTA rate] down to 3 percent, you really can start taking that data and pushing forth major policy change,” Sills says.

Uptrust’s goal is to shift from risk assessment to needs assessment and ensure people are supported throughout their interactions with the criminal justice system.

“Our view is in terms of bail reform, we need to make sure there’s not a proliferation of things like ankle monitors and whatnot,” Sills says. “For us, success is really being a subcontractor to the community as well as working with the government. I think there’s a huge risk in bail reform as it relates to technology because people see it as a big business opportunity, If a company replaces the government, they may not have the community’s best interest in mind. So it’s important to keep in mind they have the community’s best interest in mind.”

Similar to Uptrust, a tech organization called Appolition works by operating within the confines of the system. Appolition, founded by Dr. Kortney Ryan Zieger, enables people to funnel their spare change into the National Bail Out fund. As of April, Appolition has facilitated over $130,000 to go toward bail relief. Ziegler was not available for comment for this story.

Promise, on the other hand, aims to provide an alternative to the cash bail system. In March, Promise raised a $3 million round led by First Round Capital with participation from from Jay-Z’s Roc Nation.

The idea is to offer counties and local governments an alternative approach to holding people behind bars simply because they can’t afford bail. With Promise, case managers can monitor compliance with court orders and better keep tabs on people via the app. GPS monitoring is also an option, albeit a controversial one.

Let’s say you get arrested and end up having a bail hearing. Instead of asking you to pay bail, the public defender could suggest a pre-trial release with Promise. From there, Promise would work with the public defender and your case manager to determine your care plan.

“It’s clear that our values are about keeping people out of jail,” Promise CEO Phaedra Ellis-Lamkins told me on an episode of CTRL+T. “Like, we’re running a company but we fundamentally believe that not just it’s more cost effective but that it’s the right thing to do.”

Instead of a county jail paying $190 per day per person, Ellis-Lamkins said, Promise charges some counties just $17 per person per day. In some cases, Promise charges even less per person.

It’s that for-profit model that worries Clayton.

“Whenever you bring in the for-profit ethos in a criminal justice space, I think we need to be careful,” Clayton says.

She didn’t explicitly call out any companies. In fact, she said she doesn’t feel ready to make a judgment on Promise just yet. But she has a general concern of tech solutions that “dazzle and distract system actors who we really need to hold accountable and see operate in more systemic, holistic ways.”

Solutions, Clayton says, look like social safety nets like hospitals and clinics instead of jails.

“If we want to really move ourselves away from this path we’ve been on,” Clayton says, “which is towards normalizing state control of people then we should be really careful that our system that once looked like slavery to Jim Crow to mass incarceration doesn’t then become tech surveillance of all people.”

La Belle Vie wants to compete with Amazon Prime Now in Paris

French startup La Belle Vie announced a new funding round of $6.5 million earlier this week (€5.5 million). Julien Mangeard, Thibaut Faurès Fustel de Coulanges, Louis Duclert, Kima Ventures and Shake-Up Factory participated in the founding round.

Online grocery shopping is becoming quite competitive in Paris. You can order groceries from Amazon using Amazon Prime Now. And all the traditional supermarkets are launching or relaunching services to order and receive groceries within a couple of hours — Carrefour Livraison Express, Franprix’s mobile app, etc.

But all those services aren’t necessarily designed for this kind of offering. With Franprix’s app for instance, a rider is going to pick up your groceries in the nearest store and bring them to you. With Amazon Prime Now, Amazon has a big warehouse in the North of Paris filled with Kindles, books and tomatoes.

La Belle Vie wants to focus exclusively on your groceries and optimize all the steps. It starts with a big inventory. La Belle Vie sells you basic groceries, organic stuff, meat, fish and vegetables. Last year, the company acqui-hired 62degrés to sell fresh prepared meals too.

La Belle Vie has developed all its tools from scratch, including its ERP, a warehouse management service and a delivery management service. In 2017, the startup generated $3.5 million in sales (€3 million) in sales.

With this funding round, the company plans to launch a second warehouse in Paris and new cities, starting with Lyon. But the best part is that you can order croissants without going to the boulangerie — finally a croissants-as-a-service startup.

zGlue launches a configurable system-on-a-chip to help developers implement customized chipsets

The complexity and cost of packing an array of sensors and power inside a small amount of space has opened the door to a wider and wider variety of use cases for internet-connected devices beyond just smart thermostats or cameras — and also exposed a hole for getting those ideas into an actual piece of hardware.

So there are some startups that are looking to address this hole by providing developers a path to creating the customized chipsets they need to power those devices. zGlue is one of those, led by former Samsung engineering director Ming Zhang.  The company’s chiplets are built around the kind of system-on-a-chip approach that you’ll see in most modern devices, where everything is in a single unit that reduces some of the complexity of moving processes around a larger piece of hardware — shrinking the space constraints and allowing all these actions to happen on a device, such as a smartphone. As more and more IoT devices come online, they may all have varying form factor demands, which means companies — like zGlue and others — are emerging to address those needs.

“From the developer point of view, think of us as a system that is not different from any thing else on the market, user-interface-wise,” Zhang said. “It is just smaller in size, faster in time to market, and flexible — customizable by individuals rather than just by Apple and Qualcomms. [We’re] democratizing chip innovation so it is no longer [a] privilege of Fortune 500 companies.”

The company’s first product is called the zOrigin, a “chip-stacking” product that aims to allow developers to embed the sensors and processes necessary for their devices. Stemming from an ARM 32-bit core processor (meaning it can handle more complex and precise calculations), the first launch costs $149 for the wearable and development board and can include pieces like a Bluetooth radio, accelerometers, and other necessary features.

zGlue’s chipsets have embedded memory, which is an increasingly common approach to try to reduce the number of trips going from the actual processing power to where the information is stored. Those trips cost power, speed, and can restrict the scope of use cases for internet-connected devices. Zhang said the chiplets are packaged closer together — literally reducing the space that information has to cross — in order to speed it up, though that of course carries consequences when it comes to heat constraints these processors can have.

“That’s the price to pay for the continuation of Moore’s law, as it has in the past 40 years,” Zhang said. “Heat dissipation in our system is not going to be any worse than a conventional system. In fact, with the silicon substrate in place, it’s easier to conduct heat compared to a conventional package or board substrate.”

As a kind of templated approach, zGlue is geared toward helping developers produce a custom setup that the can implement into devices that may require a wide set of sensors. The company says it looks to help developers go from a design to a prototype in a few weeks, and then reduce the turnaround time from a prototype to production in “weeks or months,” depending on the complexity and volume.

While this is one example of trying to get a prototype chip out into the wild, there are a few others as well. Si-Five, for example, offers developers a way to prototype custom silicon for their specific niches based on the hardware and IP the startup has. The goal there is to offer both a prototype flow and the ability to graduate into a production flow, allowing developers and companies to get products out the door that require custom silicon. Si-Five hardware is based on the RISC-V architecture, an open-source instruction set for silicon, and the company most recently raised $50.4 million.

Zhang, too, said RISC-V offers some potential, especially in its own scope. “RISC-V is a great tool to build small, fast, and low power IoT applications,” he said. “The nature of open source makes it more available to more people. We welcome and embrace RISC-V to join the family of ‘MCU’ chiplets supported by our technology.”

When it comes to inference — the machine learning processes that happen on the hardware to execute some kind of action, like image recognition, based on trained models — Zhang said the chipsets would support it, but he would not comment further. There is a blossoming ecosystem around custom silicon that looks to speed up inference on devices like cars or IoT devices, which is geared toward reducing the space and power constraints of those chips while also running those processes much more quickly. Companies like Mythic have raised significant venture funding in order to build that kind of hardware.

WARD is an app for placing fantasy bets on esports games

Prediction markets, such as those that exist in the realm of fantasy sports, have taken off amongst consumers in the last few years. But fantasy sports have yet to make much of a play in one of the hottest areas online right now, namely esports. And it’s a big market.

Fantasy esports have been thriving across international markets. In 2017, more than 360 million viewers watched League of Legends alone, significantly overtaking the Super Bowl viewership. By 2020, the esports industry is estimated to be worth more than $1.5 billion, with the target audience being 21-35 years old. But quite how to take advantage of this arena has been a conundrum.

Now a new startup thinks it has the answer. What if you could create a live predictions market around esports as it happens?

That’s the aim of WARD, a startup out of Berlin that has created a “pick and predict” real-time prediction smartphone game, where players can win real prizes.

Billed as a fantasy esports game that provides a second-screen real-time experience for tournaments, WARD has now secured a $600,000 seed round. The backers are Impulse VC, SmartHub and a number of European angel investors. The seed investment will be used to build out the product, but also to expand in the lucrative markets of Asia and the U.S.

So how does it work? Well, fans who watch a championship or a specific esports game can predict their version of in-game events in real-time. So, for example, in the League of Legends game, a user can make a prediction about who will spill “first blood” or which team will destroy the first tower in the game, and so on.

For every prediction users make correctly, they are awarded points. Users who acquire the most points can top the leaderboard and go on to win prizes. These can include headphones, tickets for championships and signed merchandise such as team jerseys. Of course, this depends on the partner paying WARD to be featured. But, the more the user wins, the better prizes they get and the bigger the brand uplift for the team or sponsor.

Kirill Belov, managing partner at Impulse VC, says he was “stunned by the WARD technology, team and global vision. It is our first funding in the esports industry, and WARD is one of the best platforms to expand the scope of further investments.” High praise indeed.

WARD has so far run beta tests in Berlin based around the European League of Legends Championship Series, but the official launch is set for June 2018 with an aim to attract 3 million downloads by the end of the year.

The plans also include global expansion to Asia and adding new esports disciplines, such as Overwatch, CounterStrike (CSGO) to the app.

You can download the app here.