Artificial intelligence (AI) is rarely out of the headlines these days, due in large part to a steady stream of controversies from many of the major tech players. Facial recognition software is increasingly permeating society without much regard for ethics or accuracy. And crashes and near-misses blighted the recent launch of Tesla’s new Smart Summon feature, which allows drivers — in theory — to remotely beckon their vehicle in parking lots. There is a strong case for the suggestion that AI, in its current form at least, is more artificial stupidity than anything else.
Nonetheless, AI is here and here to stay. As with any transformative technology, it may take a while for AI to realize its potential. But amid the current AI hullabaloo, some budding entrepreneurs are leveraging AI and machine learning to tackle real problems and plug gaps across industries.
Against this backdrop, VentureBeat dropped in on the World Summit AI in Amsterdam last week to check out five very early-stage startups that were shortlisted to pitch their wares to a panel of judges hailing from major companies such as Oracle, Nvidia, IBM, and Accenture.
Here’s a quick snapshot of the five startups, beginning with the winner — Neuro Flash.
Neuro Flash: predicting consumer sentiment
By meshing big data, natural language processing (NLP), machine mearning, and neuropsychology, German startup Neuro Flash claims to be able to help marketers instantly discover what kinds of things consumers associate with a brand, bypassing time-consuming focus groups and questionnaires.
“We have developed a machine that can predict what you will think — and I mean literally,” claimed Dr. Jonathan T. Mall, Neuro Flash CEO and cofounder.
NLP for sentiment analysis is being used by a growing number of companies, big and small. New York-based Amenity Analytics recently raised $15 million for a cloud-based platform that leverages NLP to detect sentiment, commentary, and deceptive statements for regulatory filings and earning calls. Big companies such as Intel are also investing resources in NLP and sentiment analysis.
At its core, the Neuro Flash platform promises to “reverse engineer” consumers’ thoughts and sentiments when reading a particular piece of content, such as a website, email, or social media post. The company said it has trained its system on a swathe of publicly available digital content from the news media, Wikipedia, social media, ebooks, music, and more. “We [then] use machine learning and natural language processing to create these association networks that predict explicit and implicit [word] associations,” added Mall, a neuropsychologist turned data scientist.
After installing the Neuro Flash FlashInsight AI Chrome extension, brands can get up-to-date insights into what consumers think about their digital communications by highlighting words of phrases to see a rating of how each selection ranks against a set of parameters, covering brand values, sentiment, and so on.
Neuro Flash can also enable marketers to explore alternative wording, claims, or concepts to see how they compare and what kinds of reactions they would evoke, including any opinions or even subconscious associations the words generate.
A carmaker, for example, might want to match a new slogan with its brand values, which are “safety,” “family,” and “trust.” The company could then plug in possible slogans to see what kind of sentiment is associated with each one.
Founded out of Hamburg in 2015, Neuro Flash has yet to raise any significant outside funding, though it has gone through a number of startup accelerators over the past four years, including Microsoft Accelerator Berlin, where it was awarded mentorship and €500,000 ($552,000) in Azure cloud credits.
Neuro Flash already claims a number of big-name clients, including automotive giant Volkswagen.
ProovStation: automated vehicle inspection
French startup ProovStation is building an automated drive-through bay, which it said provides a 360-degree scan of vehicles in under three seconds — it’s all about enabling car inspections in a fraction of the time currently needed.
This could prove useful for any company that manages fleets of vehicles, from logistics companies to car rental outlets, second hand dealerships, and even manufacturers, who could use the technology to inspect new vehicles for any accidental damage.
A number of well-funded startups are employing similar computer vision smarts to detect changes to physical objects. Cape Analytics, for example, is meshing machine learning with aerial imagery to help insurance companies inspect properties from above, while ImpactVision is using AI to detect unripe or contaminated food.
The ProovStation units, which are produced at a facility in France, promise to identify any damage down to 1mm in size, while also generating an automatic estimate of the cost to fix a scratch or blemish.
ProovStation has raised around $1 million in seed funding since its inception in 2018, and the company said it plans to deploy 10 stations over the next year.
“We will deploy them with different clients to test as many use cases as possible, and start in Europe as early as the first quarter in 2020,” said ProovStation CFO Anton Komyza.
The startup is entering a potentially lucrative market — the car rental industry alone was worth an estimated $60 billion in 2016, a figure that’s projected to rise to $125 billion by 2022.
Remmedy: Pavlovian conditioning for insomniacs
“At Remmedy, we’re focusing on using AI to solve insomnia, pain, and anxiety,” said CEO and founder Mark Lazarovich.
Remmedy fits the very definition of “early-stage startup” — it was founded just a year ago, does not yet have any products on the market, and doesn’t have much in the way of publicly demonstrable prototypes. Oh, and it’s entirely bootstrapped. But at World Summit AI this week, VentureBeat met with Lazarovich, a physician based in Vermont, to get the lowdown on what seems like an interesting application of a very famous 19th century scientific theory.
Pavlovian conditioning, sometimes referred to as classical conditioning, is a mechanism whereby an animal or human can be trained through repetitive actions designed to achieve a reflex response. Russian physiologist Ivan Pavlov, who worked over a century ago, trained a dog to associate the sound of a bell with food, meaning that the dog eventually salivated whenever it heard the bell, regardless of whether there was any food present.
The same principle can be applied to other situations — such as improving sleep habits. At first, Remmedy is targeting a common condition known as “situational insomnia,” which affects people when they sleep in unusual situations, such as hotel rooms, or shift workers who work staggered hours.
For seven days, the user wears a sensor across their chest to detect heart-rate variability, data that can be used to spot patterns leading up to the moment the wearer falls asleep. The sensor sends this heart-rate variability data to the company’s servers via the user’s smartphone, and when the key sleep pattern is detected, a scent is issued from a bedside diffuser.
“The diffuser subjects you to a mist of a specific smell,” Lazarovich said. “It’s a proprietary scent that has been developed for us. And after seven nights of this happening, you become conditioned to associate this specific smell with your normal falling asleep pattern.”
After a week — in theory, at least — the user no longer has to wear the body sensor. Instead, they’re given little scent sticks that resemble a pack of gum, which they stick to their pillow. Each stick is effectively a scent-delivery device, which supposedly evokes a feeling of sleepiness and enables the user to nod off more easily.
Remmedy told VentureBeat that it expects to have its first working devices in the public domain in the next six months. While it’s far too early to pass judgment on the credibility or efficacy of Remmedy’s product, it does fit into a broader trend of technology to help people improve their sleep — the global sleep aids market was pegged as a $50 billion industry in 2016, and is expected to grow to $80 billion by 2022. Apple, for example, is expected to join the long list of tech companies to have launched a sleep tracker, while other startups are focusing on more novel ideas, such as beds that rock you to sleep.
In the future, Lazarovich said the same conditioning techniques could be applied to pain management to reduce reliance on drugs, by training the body to react to smells in a similar way as it does to opiates. However, that would require regulatory approval since it would be used in a medical environment.
Pixyle.ai: visual search for fashion retailers
Pixyle.ai is one of a number of companies using computer vision to improve visual discovery. Founded in 2018, and with just $77,000 in grants and seed-funding to its name, Netherlands-based Pixyle.ai provides an API that lets any fashion brand integrate image-based search into their digital properties. So a user could upload a photo of a dress that they like, for example, and the online retailer would find similar-looking items.
The underlying technology isn’t particularly novel. Pinterest has been leaning on computer vision-powered search for a number of years already, while Google Lens offers a pretty powerful demonstration of how machines are learning to see. And European fashion retail giant Asos offers built-in visual search functionality.
However, the Dutch startup is looking to set itself apart by focusing on visual search for third-party fashion-focused retailers. The idea is that powering visual search for clothes and accessories, including jewelry, saves retailers having to build their own dedicated visual search smarts.
“AI technology is only as good as the data it is trained on,” noted Pixyle.ai founder and CEO Dr. Svetlana Kordumova. “So that’s why I put a lot of focus on creating a massive database and knowledge base for fashion.”
Kordumova said the company has processed some 200,000 images into 300 categories and trained its systems on such attributes as colors and patterns. “With this technology, and the data we have created, we are now able to detect even the tiniest objects in images — like, for example, earrings.”
EmitWise: carbon footprint monitoring
“I really think this planet is dying,” said EmitWise cofounder and CEO Mauro Cozzo. “And if we don’t rush and figure out how to save it, we are freaking screwed.”
Earth’s climate crisis was a recurring theme throughout Cozzo’s pitch for his fledgling startup EmitWise, which is setting out to help big businesses monitor, manage, measure, and ultimate reduce their carbon footprint.
“We’re trying to accelerate the transition to a global carbon neutral society by producing the data that is required to figure out how to do that,” Cozzo said.
EmitWise’s software integrates with companies’ various systems, including enterprise resource planning (ERP) tools, and uses data from business documents, such as procurement sheets, in addition to publicly available information. It then uses machine learning to aggregate and analyze this data, with a view to continuously auditing the carbon footprint of the company’s internal operations and its entire supply chain.
The platform aims to help companies comply with emissions policies, as well as giving them something to “improve their brand image,” given that climate change is becoming a key concern for consumers and companies globally.
Cozzo founded EmitWise just four months ago and said the fledgling startup is currently working on pilots with two large enterprises.
While the impending climate catastrophe is stirring broad concerns, there remains a deep-rooted belief that solutions to the problem will likely have to be anchored in profitability.
“It looks like we might be able to start doing business in a different way — sustainability might start being profitable,” Cozzo said.
There are countless other AI startups out there, with 2019 already shaping up to eclipse all previous years for AI startup funding. Indeed, recent data from the National Venture Capital Association showed that 965 AI-related U.S. companies raised north of $13 billion in VC funding for the first nine months of the year, compared to the $16.8 billion plowed into 1,281 companies in the whole of last year.
The five shortlisted AI startups at World Summit AI offer a glimpse into some of the hot areas of the moment: NLP and computer vision were common themes, while extracting meaning from big data was also big — and an area VCs continue to invest in heavily.
But over and above all that, the age-old adage that for technology to succeed it has to fix a real problem was as evident as ever. Neuro Flash offers tools to help marketers understand consumer sentiment without the resource-intensive process of carrying out focus groups; ProovStation is pushing to help automotive companies identify vehicle damage in seconds rather than minutes; Remmedy wants to help people get to sleep; Pixyle.ai aims to help large fashion companies keep pace with advancements in visual search technologies without having to build the technology from scratch; and EmitWise is setting out to encourage big companies to be more proactive in cutting carbon emissions.
Together, they serve as a reminder that AI applications need to offer companies genuine value. There is no point investing in automation just to keep up with the cool kids.