artificial intelligence

ARTIFICIAL INTELLIGENCE: Proof that we have been training AI fakes to stab us in the back

In the 1933 film Duck Soup, actor Chico Marx is famously known to have asked, "who ya gonna believe, me or your own eyes?" Fairly meaningless in the 30s, but today, it's more relevant than ever. Let us explain. We know how the ever-expanding capacities of computing power and algorithm efficiency are leading to some pretty wacky technology in the realm of computer vision. Deepfakes are one of the more terrifying outcomes of this. A deepfake can be described as a fraudulent copy of an authentic image, video, or sound clip, which is manipulated to create an erroneous interpretation of the events captures by the authentic media format. The word 'deep' typically refers to the 'deep learning' capability of the artificially intelligent algorithm trained to manifest the most realistic version of the faked media. Real-world applications being: Former US president Barack Obama saying some outlandish things, Facebook founder Mark Zuckerberg admitting to the privacy failings of the social media platform and promoting an art installation, and Speaker of the US House of Representatives Nancy Pelosi made to look incompetent and unfit for office.

Videos like these aren’t proof, of course, that deepfakes are going to destroy our notion of truth and evidence. But it does show that these concerns are not just theoretical, and that this technology — like any other — is slowly going to be adapted by malicious actors. Put another way, we usually tend to think that perception — the evidence of your senses (sight, smell, taste etc.) — provides pretty strong justification of reality. If something is seen with our own eyes, we normally tend to believe it i.e., a photograph. By comparison, third-party claims of senses — which philosophers call “testimony” — provide some justification, but sometimes not quite as much as perception i.e. a painting of a scene. In reality, we know your senses can be deceptive, but that’s less likely than other people (malicious actors) deceiving you.

What we saw last week took this to a whole new level. A potential spy has infiltrated some significant Washington-based political networks found on social network LinkedIn, using an AI-generated profile picture to fool existing members of these networks. Katie Jones was the alias used to connect with a number of policy experts, including a US senator’s aide, a deputy assistant secretary of state, and Paul Winfree, an economist currently being considered for a seat on the Federal Reserve. Although there's evidence to suggest that LinkedIn has been a hotbed for large-scale low-risk espionage by the Chinese government, this instance is unique because a generative adversarial network (GAN) -- an AI method popularized by websites like ThisPersonDoesNotExist.com -- was used to create the account's fake picture.

Here's the kicker, these GANs are trained by the mundane administrative tasks we all participate in when using the internet on a day-to-day basis. Don't believe us? Take Google’s human verification service “Captcha” – more often than not you’ve completed one of these at some point. The purpose of these go beyond proving you are not a piece of software that is unable to recognise all the shopfronts in 9 images. For instance: being asked to type out a blurry word could help Googlebooks’ search function with real text in uploaded books, or rewriting skewed numbers could help train Googlestreetview to know the numbers on houses for Googlemaps, or lastly, selecting all the images that have a car in them could train google’s self-driving car company Waymo improve its algorithm to prevent accidents.

The buck doesn't stop with Google either, human-assisted AI is explicitly the modus operandi at Amazon’s Mechanical Turk (MTurk) platform, which rewards humans for assisting with tasks beyond the capability of certain AI algorithms, such as highlighting key words in an email, or rewriting difficult to read numbers from photographs. The name Mechanical Turk stems from an 18th century "automaton" or self-playing master chess player, in fact it was a mechanical illusion using a human buried under the desk of the machine to operate the arms. Clever huh?!

Ever since the financial crisis of 2008, all activity within a regulated financial institution must meet the strict compliance and ethics standards enforced by the regulator of that jurisdiction. To imagine that a tool like LinkedIn with over 500 million members can be used by malicious actors to solicit insider information, or be used as a tool for corporate espionage, should be of grave concern to all financial institutions big and small. What's worse is that neither the actors, nor the AI behind these LinkedIn profiles can be traced and prosecuted for such illicit activity, especially when private or government institutions are able to launch thousands at a time. 

maxresdefault.jpg
54646.PNG
800.jpeg
captcha_examples.jpg
amazon-mechanical-turk-website-screenshot.png

Source: Nancy Pelosi video (via Youtube), Spy AI (via Associated Press), Google Captcha (via Aalto Blogs), Amazon MTurk

ROBOADVISORS & INVESTING: Robinhood's latest $8bn valuation means that scale players need to wake up

There’s no such thing as a free lunch in life, but there are such things as free trades on Robinhood. What Chime did with banking, Robinhood has done with trading. Their massive 4 million active user base is enviable to every other Fintech. So then it's no surprise that the firm is estimated to be valued at $7-8 billion, following a $200 million fund raise with existing investors. Founded in 2013 by two former Stanford University roommates, Baiju Bhatt and Vlad Tenev, with the goal of  building a brokerage service that democratized access to the financial system -- specifically, stock trading and its significant barriers to entry (costs, fees, and minimum capital requirements). Since it's launch, millennial investors -- an elusive audience to traditional financial services firms -- have flocked to the service to trade stocks, options, cryptocurrencies and exchange-traded funds, at low-to-no fees.

Such success stems from the app's ability to earn fees via indirect channels such as marginal interest, lending, a $6 per month premium product called Robinhood Gold -- offering up to $1,000 of margin to trade with, and lastly, rebates from high-frequency trading and payment order flow. Here, third-party market makers, such as Citadel Securities, Two Sigma, and Virtu, pay Robinhood a rebate for processing trades on the app's behalf, apparently to offer better execution quality and prices. Whilst that sounds noble, it must not be forgotten that such a non-transparent practice -- as noted by CNBC -- could encourage brokers to send orders to market makers that offer the most generous rebates, and not necessarily the ones who offer the best prices for stocks. However, this is likely not to be the case as Robinhood's leadership has stressed that "we don’t take rebates into consideration when we choose which market maker will execute your orders. Also, all market makers with whom we work have the same rebate rate". Last year Bloomberg reported that Robinhood made in excess of 40 percent ($69 million) of its 2018 revenue from payment order flow.

Additionally, Robinhood is planning a U.K. launch to muscle-up against the likes of challenger broker Freetrade -- a London-based twin of Robinhood, and challenger bank Revolut -- who has indicated its intention to offer a free trading platform in the near future. The interesting aspect here is that Robinhood has been desperate to become a full-service bank, with evidence of this coming from last year when the company ended up with egg on its face after announcing its intentions to launch savings and checking accounts with 3% interest rates (30 times the U.S. national average) - despite not being FDIC insured (which is illegal). All too soon after this discovery was brought to regulator's attention, the product was rebranded as a "cash management program" and references to deposit protection were swiftly removed. Yet, the pursuit continues, as the company's second attempt has recently been made via an application for a bank charter in Push-to-Offer Traditional Banking Services with the Office of the Comptroller of the Currency (OCC).

Lastly, there are rumors that Robinhood is expecting a much bigger round of funding later this year, which could value the company at over $10 billion. This, coupled with the success of the company's latest commission-free crypto trading app, U.K. expansion, and launch of its full service bank, should make scale players in the industry such as Schwab, E-Trade, M1 Finance, and Fidelity fairly nervous. From zero-fee index funds, to zero-fee trading of single stocks. Fee-free trading apps like Robinhood, Vanguard, and FreeTrade have initiated a pricing war between scale players and themselves. So long as the strategy to fight this war remains: platforms and marketplaces who cross-sell products with the aim to retain customers and lock them into a sales cycle, this tech-enabled price war will squeeze margins down to zero. Last one to the bottom is a rotten egg.

12165165.PNG

Source: Robo-Advisors with the most AUM (via Roboadvisorpros)

download51651.png
Rowady3-1024x520.1555604603433.jpg
Phone-array-LTC-BCH.png

Source: Robinhood (via Bloomberg), Robinhood Gold (Robinhood Blog), CNBC (article), Robinhood Crypto (Robinhood Blog)

ARTIFICIAL INTELLIGENCE: Amazon's new wearable edges us closer to a reality of emotionally manipulative financial institutions

In the past, we have touched on how a specific device that you use for conversational interface interactions will be locally better at understanding you -- rather than some giant squid-like monster AI hosted on Amazon Web Services. But, what if the conversational interface device is the friendly avatar to such a terrifying AI monster that possesses the ability to emotionally manipulate its user? Well, Isaac Asimov eat your heart out, Amazon are reportedly building an Alexa-enabled wearable that is capable of recognizing human emotions. Using an array of microphones, the wrist-worn device can collect data on the wearer's vocal patterns and use machine learning to build models discerning between states of joy, anger, sorrow, sadness, fear, disgust, boredom, and stress. As we know, Amazon are not without their fair share of data privacy concerns, with Bloomberg recently disclosing that a global team of Amazon workers were reviewing audio clips from millions of Alexa devices in an effort to enhance the capability of the assistant. Given this, we can't help but think of this as means to use the knowledge of a wearer’s emotions to recommend products or otherwise tailor responses.

Let's step back for context. Edge computing is the concept that there are lots of unique distributed smart devices scattered throughout our physical world, each needing to communicate with other humans and devices. Two layers of this are very familiar to us: (1) the phone and (2) the home. Apple has become a laggard in artificial intelligence -- behind Google on the phone, and behind Amazon and Google at home -- over the last several years. Further, when looking at core machine learning research, Facebook and Google lead the way. Google's assistant is the smartest and most adaptable, leveraging the company's expertise in search intent to divine meaning. Amazon's Alexa has a lead in physical presence, and thus customer development, as well as its attachment to voice commerce. Facebook is expert in vision and speech, owning the content channels for both (e.g., Instagram, Messenger). We also see (3) the car as developing a warzone for tech companies' data-hungry gadgets.

Looking back at financial services, it's hard to find a large financial technology provider -- save for maybe IBM -- that can compete for human attention or precision of conversation with the big tech firms (not to mention the Chinese techs). We do see many interesting symptoms, like KAI - a conversational AI platform for the finance industry used by the likes of Wells Fargo, JP Morgan, and TD Bank; but barely any compete for a relationship with a human being in their regular life. The US is fertile ground for this stuff, because a regulated moat protects financial data from the tech companies. Which is likely to keep Big Tech away from diving head first into full service banking, but with the recent launch of Apple's AppleCard we are starting to see vulnerabilities in that analogy. So how long can we rely on the narrative so eloquently put by Chris Skinner"the reason Amazon won’t get into full service banking is because dealing with technology is very different to dealing with money; furthermore, dealing with money through technology is very different to dealing with technology through money"? Also, how would you feel about your bank knowing when you are at your most vulnerable?

54654654.PNG
65465.jpg

Source: Bloomberg Article, KAI Platform (via Kasisto)

ROBO ADVISORS: Robo-advisors are winning but leaving cash on the table

We will keep this brief. In a recently updated, “Robo-Advisors with the Most AUM” the top 5 robo-advisors, consisting of three Fintechs and two Incumbents, remained in the same position as last year, although each of them have seen gains in Assets Under Management (AUM) and the number of accounts. Yet, the jury is out as to whether gathering assets or gathering users are good measures of success -- we wrote about it here.

A lot of digital wealth management innovation targets people who have been excluded from the traditional wealth management business because the amounts they have to invest are too small for the economics of traditional wealth management to work. So the strategy is to target this opportunity by getting to the consumer, earn them loyalty with at least one good service, perhaps free, and then lock them into a full financial services relationship. The expected outcome of this is to see a reduction in the number of these individuals and/or the assets they hold -- Unadvised assets - the liquid cash in real wallets and check & savings accounts.

Daily fintech's Efi Pylarinou, has done the heavy lifting on this, finding unadvised assets in the US, EU, and UK to be around $14.5 trillion, $13.7 trillion, and $3 trillion respectively. Surprisingly, each of these on average have experienced growth of 9% over the past 3 years. Such findings point to the fact that, since their inception, robo-advisors have had none or a negligible impact on unadvised assets. Although unadvised assets are impacted by all innovations in Fintech, robo-advisors are more likely to be the ones that incentivise you to split up with your cash to some degree in hopes of generating returns with very little friction/costs. And if this is a direct result of trends in monetary policy, public markets, and human behavior superseding the digitization of capital markets, when should we expect the reversal to occur?

sdfdsf.PNG

Source: Robo-Advisors with the most AUM (via Roboadvisorpros)

ARTIFICIAL INTELLIGENCE: Synthesia prove that not all deep fakes are malicious, but for those that are, is Blockchain the answer to spotting them?

Last week we touched on how convolutional neural networks can be easily duped using nothing more than a computer-generated "patch" applied to a piece of cardboard (here). This week we want to keep the theme of neural networks alive, only this time addressing the fascinating topic of deep fakes. We have discussed this before (here), touching on how hyper-realistic media formats, such as images and videos, can be faked by a model where one algorithm creates images and another accepts or rejects them as sufficiently realistic, with repeated evolutionary turns at this problem. These algorithms are known as generative adversarial neural networks (GANs). Initially GANs were used in jest to make celebrities and politicians say and do things they never (here), over time, however, their sophistication has prompted more malicious use cases. Evidence of such malicious intent is reportedly coming from China in which GANs are used to manipulate satellite images of earth and/or provide strategic insight to manipulate the Chinese landscape to confuse the image processing capabilities of adversarial government GANs. Think about it, GANs, much like in our cardboard patch example, can be fooled to believe that a bridge crosses an important river at a specific point. This, from military perspective, could lead to unforeseen risk exposure to human lives, similarly so, in the context of open source data used by software to navigate autonomous vehicles across a landscape. Such malicious use cases of GANs have resulted in the concern of government entities such as The US Office of the Director of National Intelligence who explicitly noted deep fakes in the latest Threat Assessment Report (here). China has gone one step further, recently announcing a draft amendment to its Civil Code Personality Rights to reflect an outright ban on deep fake AI face swapping techniques. Currently, GANs dedicated to counteracting deep fakes are purely reactionary to those dedicated to creating them, but we are seeing novel solutions harnessing blockchain technology come from the likes of Amber - who protect the integrity of the image/video data via "fingerprinting" -- a sequenced cryptographic technique applied to bits of data associated with a single frame/image, which flags any manipulation to the original file.

But let's end this on a good note shall we. An AI-driven video production company called Synthesia used GANs to "internationalize" a message delivered by football icon David Beckham to raise awareness around the Malaria Must Die initiative. Synthesia's GANs were trained on Beckham's face so that 9 different malaria survivors could deliver their message through his avatar in their mother tongue. The resultant campaign has over 400 million impressions globally, and provides insight into the evolution of digital video marketing, corporate communications, and advertising which leverages GANs to reduce production costs and improve engagement.

deepfakes.PNG
amber.PNG
tesla ridesharing app.PNG

Source: arxiv.org (Deep Video Portraits Report), AmberVideo, Malaria Must Die (via Youtube)

INNOVATION & PAYMENTS: Tesla entering the autonomous vehicle "space race" does not bring us closer to a Utopian future, yet

It's difficult to ignore the utopian dream of riding shotgun in a fully autonomous vehicle whilst chuckling at the seemingly prehistoric ideas of road rage, congestion, and side-mirrors. Yet, upstarts dedicated to making this dream a reality ingest massive amounts of venture funding with little return. Take transportation-on-demand app Uber -- who recently raised $1 billion for its Advanced Technologies Group (ATG) from Softbank, Toyota, and auto-parts manufacturer Denso (here). The aim of the investment is to accelerate the development and commercialization of automated ridesharing services, especially given that the company blames the bulk of its estimated $702 million net loss this quarter on costs attributed to human drivers (here). Question is, how sophisticated the software has become since the 2018 incident in which a driverless Uber vehicle struck and killed a pedestrian? Interestingly, Alphabet-backed and Uber-rival Waymo, boasts racking up over 10 million miles worth of autonomous driving data as a hedge against such fatal incidents. Up until last week, Waymo prided itself as the only upstart to have launched a dedicated commercial driverless car service (Waymo One). Enter electric-vehicle giant Tesla -- who promised an all-electric, 1 million car fleet of self driving Tesla taxis by the end of 2020. Some, of which, will come from existing Tesla's on the road -- which will be used as autonomous taxis when their owners do not need them. This is noteworthy because Tesla has amassed over 1 billion miles worth of 'Autopilot' data, which was used to build their latest custom-designed artificial intelligence driving chip -- claimed to allow Tesla's to pilot themselves. The only missing pieces to the puzzle are (1) regulatory approval for such vehicles to legally operate and (2) "feature-complete" software to prevent any life-threatening incidents, both of which are assured to be ready for 2020 year end launch. 

Whilst there's no doubt that we have a "space race" type scenario between digital transportation upstarts: Waymo, Uber, and now Tesla -- all competing to arbitrage a phone's GPS to deliver custom mobility solutions with greater precision and experience than a human transaction can. There is concern around the impact that autonomous taxis will have on the existing infrastructure, especially what they will do in-between customers: park, go home, or drive around aimlessly. All of these have significant congestion implications. Such implications could incentivise upstarts aimed at offering an aggregated view of transportation options available to customers, such as CityMapper -- whose latest subscription offer 'Pass' -- exemplifies how to take this one step further by building an instantiated financial product on top of abstracted digital infrastructure (here). Until then we will continue to dream.

uber merger.jpg
waymo.PNG
tesla ridesharing app.PNG
uber vs tesla.jpg
imgonline-com-ua-twotoone-PfYb0jil3tIIBLu-625x1024.jpg

Source: BusinessWire (Uber's Advanced Technologies Group $1 billion), Waymo, Techcrunch (Tesla Ridesharing App), Techcrunch (Uber vs. Tesla), Gizmodo (Citymapper Pass)

ARTIFICIAL INTELLIGENCE: Cornell students break convolutional neural networks using cardboard

Earlier this year we touched on how the the digitization of the human animal continues unopposed, with symptoms all over. China is a great example of a sovereign infatuated with this more than any other. Harnessing sophisticated machine vision software and swarms of CCTV cameras to strengthen the sovereign-imposed social human constructs of law, power, culture and religion. Leveraging apps to do its dirty work, such as Chinese firm Megvii, maker of software Face++ that has catalyzed 5,000 arrests since 2016 by the Ministry of Public Security. Pretty scary stuff. But, as with any software, there are always ways to break it, and it seems as though the folks over at Cornell University have figured out a creative way to deceive a convolutional neural network. Using computer-generated "patches" that can be applied to an object in real-life video footage or still frame photographs to fool automated detectors and classifiers. The main use case is to generate a patch that is able to successfully hide a person from a person detector i.e., an attempt to circumvent surveillance systems using a piece of uniquely printed cardboard which faces the camera and covers some aspect of the subject's body. The accuracy of machine vision stems from the software's ability to break the image up into various filters and pixels, comparing it to thousands of digested images, and using statistics to generate a probabilistic classification of what is presented in that image. Given this, it is clear why such a rudimentary solution could fool such sophisticated neural networks. Your move China. (READ MORE)

f938b80d-c28a-4237-b2b9-eeb9031d6eaa.png
c537769f-d774-4053-9627-a404433bb4ce.png
e1a0e6ad-8d35-4912-a1b2-53ad6ecd9d30.png

INSURTECH: Softbank's $300 million double-down on digital insurer Lemonade

Last week we saw Softbank double-down on its backing for Lemonade - the renter's insurance company built for Millennials. In its Series D funding round led by Softbank and supported by Allianz, General Catalyst, GV, OurCrowd, and Thrive Capital, the poster child of disruptive InsureTech innovation, raised an additional $300 million. This latest cash injection, coupled with revenues of $60 million in 2018 and potential $100 million in 2019, puts the company at an estimated $2 billion valuation, and is set to help fuel further growth in the US and with expansion into Europe. We will remind you that Lemonade uses artificial intelligence and analytics to replace the front-office function of incumbent carriers. Simply, their mobile app can chat with users and onboard them without much human involvement. Last year, this was personified in an attempted smear ad run by competitor - StateFarm, who ridiculed the usage of bots and technology in insurance, mentioning “a knockoff robot created by a rival insurance company.” Needless to say that the digital insurer took that lemon and made...well...lemonade - sponsoring the ad across social media, essentially because it promoted Lemonade's AI tech. Last year, we mentioned that Softbank's portfolio of millions of American financial services companies with modern technology stacks and cool brands, spread across different verticals, requires only one of them to be a Goldman Sachs. Could this news be a sign?

Imagen 1 high(1).jpg
quartely_sales-1024x536.png
ad005fe7-7eb1-4c62-97a6-65c11271f04c.jpg
802d93e3-2137-41c2-88a4-6710929af1de.png

Source: DigitalInsuranceAgenda (Lemonade), Lemonade (2018 Results), Youtube (Lemonade - StateFarm Ad), Twitter (Daniel Schreiber)

BIG TECH: YouTube, Facebook and NVIDIA powering hyper-realistic human avatars

The digitization of the human animal continues unopposed, with symptoms all over. Chinese firm Megvii, maker of software Face++ that has catalyzed 5,000 arrests since 2016 by the Ministry of Public Security, is looking for an $800 million IPO. The other champion of public/private surveillance, Facebook, is working a virtual reality angle. The company is improving the technology used to model rendered avatars of human faces, which can then be displayed across virtual environments. Using multi-camera rigs and hours of facial movement footage, Facebook is building neural networks that learn how to translate realistic facial muscle movement into models. The Wired article linked below is worth exploring for the videos alone, and the uncannily realistic motion these animation possess.

One of our recurring points is that frontier technologies -- AI, AR/VR, blockchain, and IoT -- appear disparate now, but are intricately connected. Take for example the new feature from Google called YouTube Stories. Similar to SnapChat and Instagram, video creators can apply 3D augmented reality overlays to their faces. While this technology looks like virtual reality rendering, it is primarily a machine vision (i.e., AI) problem to anchor rendered objects to a human face realistically. To do this Google provides a developer library called ARCore, not to be confused with Apple's ARkit. Human video avatars can be further extended and customized with code -- the twenty first century version of personal branding.

Another take on the same issue comes from generative adversarial neural networks (GANs). We've discussed before how hyper-realistic images and videos can be faked by a model where one algorithm creates images and another accepts or rejects them as sufficiently realistic, with repeated evolutionary turns at this problem. Highlighted below is a recent software release from NVIDIA, where a drawing of simple shapes and lines is rendered by a GAN into what appears to be a hyper-realistic photo of a landscape. We can imagine a similar approach being applied to the output generated by Facebook's avatars, which still border on creepy, to ground the outcome in reality. Little details, like a reflection of a cloud on water, are hallucinated by GANs automatically, based on massive underlying visual data. Expect these digital worlds to become increasingly indistinguishable from reality, and to spend way more time living in them for the years to come.

c0552a77-9b22-4f32-8c09-36f2ca71f5ac.jpg
water.PNG

Source: SCMP (Face++), Wired (Facebook Avatars), NVIDIA (GAN drawing)

BIG TECH: The macro-scams of Fyre and Theranos & the micro-scams of Google and Facebook

Spoiler alert: Fyre Festival ended up being a securities fraud that cost investors $27 million dollars, left hundreds of workers unpaid and emotionally ravaged, and negligently put attendees in dangerous conditions. Even Blink-182 cancelled their performance! Another spoiler: Theranos ended up being a securities fraud costing investors (including Betsy DeVos!) $700 million, leaving hundreds of workers unpaid and pushing at least one to suicide, negligently putting users of the product in dangerous medical circumstances. In both cases, the founders were young and narcissistic, optimizing the story-telling about their company over delivering on the promised expectations. Billy McFarland used Instagram supermodels to sell a false vision. Elizabeth Holmes leveraged the Steve Jobs black turtleneck and VC group think to do the same.

This stuff is so easy in retrospect -- to point fingers and throw the stone. Having spent a lot of time in the early stage ecosystem, we can tell you that all founders have these devils inside them. These are the devils that let you take the risk, tell the story and defend your tribe (e.g., see Elon Musk). The issue is that these particular people could not and did not execute -- and any reasonable person in their situation would know enough to stop marketing and selling lies. We can look at crypto ICOs to date and say the same thing. Surely the people who raised over $30 billion globally, and burned nearly all of it, sold us a falsehood. Some -- like John McAfee or Brock Pierce -- had to know what was up. Or did they, perhaps believing in a zeitgeist change tilting the axis of human industry? 

The issue is asymmetric information and intent to profit from that asymmetry. When someone sells us a broken car claiming it works great, they are selling a "lemon" -- something the US protects against with "lemon laws" that remedy damages from relying on false claims. Let's shift from these obvious macro lemons, to the invisible micro lemons sold by Facebook and Google. It was revealed that Facebook was -- in the worst case -- paying 13+ year olds $20 per month to install a research app that scraped all their activity (from messages to emails to  web) and provided root permissions to the phone, misusing Apple-issued enterprise certificates. Facebook should not have been able to create these apps for anyone other than its employees on internal apps (e.g., bug testing new versions). But it did, and got its access revoked by Apple immediately.

Google did a version of this too, exchanging gift cards for spying on web traffic. As yet another example, Google's employees refused to help the company build a war-drone AI for the US Department of Defense. So instead, Google outsourced the work to Figure Eight (a human-in-the-machine company), hiring gig economy workers for as little as $1 per hour for micro-tasks like identifying images (teaching drones to see). These workers had no clue what they were doing -- and we imagine that some would exhibit the same ethical concerns that Google employees did in refusing the work. In all these tech company examples, the lemon is the un-revealed total cost. Compared to Fyre and Theranos, where we pay billions, and get nothing in return, here we are given $1 an hour or $20 per month (i.e., nothing), but we lose our privacy, agency and humanity (i.e., everything). 

cbef13a2-5fb2-44be-a753-14836463401b[1].jpg
441ca0eb-3a96-4cec-ab2b-5f6ab911ddec[1].jpg
b83d1247-1bc8-4dd7-b1d3-1d0cfab382c9[1].png
3a1235ce-7bd7-4f7e-bc74-1fd2b1b46a11[1].jpg

Source: Wired (Project Maven), The Intercept (Google project maven), Gizmodo (Google micro-tasks), TechCrunch (Facebook, Google, Apple), Wikipedia (FyreTheranosLemon Law)

INSURANCE: Porsche and Mile Auto to cut premiums 40% using AI for pay-per-mile insurance.

Insurance is the holy grail for Artificial Intelligence and the Internet of Things in finance, because it requires a messy interaction with the physical world, rather than living merely in a spreadsheet, database, or blockchain. To this end, we like the news of Porsche partnering with Mile Auto on pay-per-mile insurance. There is a reasonable demand-side argument: owners of Porsches don't drive the car as a primary automobile, and would prefer to only pay insurance for the time they are actually on the road. The second argument is even more fun -- owners of Porsches don't want to be tracked via GPS or a black-box by something like Cambridge Mobile Telematics ($500MM from Softbank) or Metromile ($90MM from VCs) because they are fancy and private people. No tracking please!

How does the thing work? You pay a cheap base rate to Mile Auto, and once in a while take a picture of the speedometer's reading in the app. The picture is translated to numbers via a machine vision algorithm, and your per-mile variable insurance rate is calculated on the spot. The company claims this will lead to a 40% reduction in premiums for the average user. For what it's worth, we hear that the growth of renter's insurer Lemonade is similarly fueled by people who are forced to get coverage (e.g., by the landlord) but are looking for the most discounted, easy to manage product. What does that mean? It means that the low risks self-select out of the insurance pool, driving up the price for unsophisticated non-techies that don't drive a Porsche.

Let's take the argument to an absurd extreme. On the developer website Programmable Web, there are 59 separate APIs that developers can use to build insurance apps and connect into underwriting engines and carrier capital. From Clearcover (affordable car insurance in your app!) to Haven Life (term life insurance on any website or application!) to Lemonade, OCBC Materntity, Qover and a plethora of others, developers have real choice in how to weave these more digital insurance products into the attention black holes in your phone. What happens when the tech-forward customer considers only these options, and the conservative customer considers only insurance sold by agents and direct mailing? Could there be a bifurcation of risk profiles that fundamentally injures the risk-pooling function of the industry? Perfect information about risk collapses the value of hedging. Half of us will know and live in a predicted future, while the other half will pay for the ignorance.

dc33b340-6e91-464e-ae0c-2e4cdb64883a[1].png
cb8107d5-26fd-4ee4-89d6-b66aac565087[1].png
5b9fd9a3-e19d-4096-bf5e-cea629568e49[1].png
a9eebfbd-05f1-4bc7-8e9e-29a0fea81f6e[1].png

Source: PR Newswire (Porsche), Company websites, Programmable Web (Insurance)

ARTIFICIAL INTELLIGENCE: "Financial Deadbeats" map is the worst things about Chinese Fintech

In our continued amazed gawking at the Chinese fintech landscape, we bring you the following. There is now a feature within WeChat, one of two channels for all mobile chat communication, to show a map of "financial deadbeats" around you. That's right -- a shaming visualization of people who are in financial trouble, like some sort of public sex offender list. We link to the article below, and assume that it is true despite how preposterous the whole thing seems. 

Offenses that could land you on the blacklist include serious ones like being the founder of a digital lender that collapsed with 12 million unpaid accounts, and trivial ones like being a single mother embroiled in a divorce proceeding. Once you are on the list, not only will your full name and financial information be public entertainment on this app, but access to credit, commerce and university admission could be revoked. To add insult to injury, a special ringback tone is added to the "discredited" person's mobile phone, alerting any potential caller about your poor financial management skills.

We add to this soup the idea of algorithmic bias exhibited by AI based on training data. We've covered this issue in the past, but point to Rep. Alexandria Ocasio-Cortez (D-NY) recently bringing it up into mainstream conversation. From propaganda bots to algo-racism, these arcane issues are starting to concern the broader Western polity. So when you combine historical training data reflecting past social and economic biases with social media enforcement systems, dystopia calls. One of the most important financial innovations in the West was bankruptcy, allowing entrepreneurs to fail and start over. This normalization of financial wipe-out led to an equilibrium with higher risk-taking and innovation. It is chilling to see technology being used, with potential for error and misuse, to stifle that spirit. Based on the US personal bankruptcy data below, you can see that 6 out of 1000 people would be guilty according to WeChat, skewed in large part to minority populations. No thanks. 

59ca575a-39a6-4f36-a60c-89c9ccfd9a01[1].jpg
f1102030-8e90-4366-8bcf-31e14f4b774b[1].gif
0f2b8c5a-2d85-479c-9ab1-239e47436d17[1].png

Source: Abacus News (deadbeat map), Independent (deadbeats), Vox (algo-racism), On bankruptcy normalization and bankruptcy zip codes

ARTIFICIAL INTELLIGENCE: Evolution of Creative AI and WeChat's Payment Score

One ongoing, false refrain is that machine learning does not generate creative outcomes. Increasingly, this is proven wrong by the technologists and artists playing with the technology. What started several years ago as "neural style transfer" (i.e., transferring Picasso's visual DNA to any photo) has moved on to BigGAN, which is a machine learning algorithm to manufacture images that appear realistic but are made from machine hallucination. Notably, artists are playing not just with the realistic versions of these hallucinations, which you can see below, but with the "latent space" in between. This mathematical term for interpolation is filled with abstract, surprising, and surreal outcomes. Our takeaway from these results is both (1) that machines will be far more precise in understanding and approximating humans than we assume, and (2) that machines will be far better at creativity that we assume.

Fitting a financial product to a ranked "perception" of a human being matters -- especially when it is done at a scale of a billion people. Tencent's WeChat is running a new initiative called "WeChat Pay Score", which is analogous to the Alipay's "Sesame Credit", both of which (we expect) flow to the Chinese government to make up the national social credit score. Sesame Credit looks at 5 dimensions: safety, wealth, social, compliance, and consumption from over 3,000 specific data points collected by the app. The WeChat version is collecting data on how users chat on the messenger, what they read and buy, where they travel, and how they run their life in general. These combined attributes grant access to perks, like waiving bank account minimums.

Listen, in a massive nation where a large swath of the population doesn't have traditional financial data or bank accounts, machine-learning based estimates of credit-worthiness are a life saver. Not every economy comes with a FICO score and legacy credit agencies (though the Equifax breach wasn't particularly kind to incumbents).  But they key question comes back to the two picture sets below. Do the machines see us like those perfectly generated, accurate pictures of people? Or like the surreal goo in abstraction? The former means distributed access to well-suited financial products, while the other is a Black Mirror nightmare.

8779262f-e84c-4f5f-b44d-760b434eb989[1].jpeg
7c2e985e-bd99-4c14-9acc-4524618f4782[1].png
17bad1d6-62d4-495d-ba05-3dfbed2985d7[1].jpg

Source: Medium (GANs), Joel Simon (GANreeder), TechCrunch (WeChat)

INSURTECH: Rage Against the Machine and $500MM telematics Softbank investment

Let's start off with the ridiculous, and get more ridiculous. SoftBank has a lot of money to invest in category killing fintech businesses, and one of the latest such players is Cambridge Mobile Telematics, which just received $500 million from the investor. What is it? A widget attached to a car windshield, and then used to collect data about the quality of a particular driver -- from speeding to breaking. This data is then tied to the purchasing of insurance, where "good" drivers have access to lower cost financial products. This is an interesting, and pioneeing, example of how edge computing will create orders of magnitudes more digital data that then feeds the manufacturing of finance. 

A sneaking suspicion in the back of our minds is that driving data is really good for training robots how to drive. Meaning, Google and the rest of the big tech companies are all running experiments with self-driving cars on the road to collect driving data. Something simple from a telematics device certainly is not equivalent to major machine vision and radar data. But it does paint a straight line towards how self-driving car insurance should be priced. Let's repeat that. If a widget in a car tells you insurance prices based on driving performance and you combine that with an AI car, you could compare humans and machines on an apples to apples basis.

The ridiculous part is the human response to tech-first transportation companies. In London, Chinese bike-sharing company Ofo is pulling out of the city because people steal and destroy their untethered bikes. In California, aspiring freedom fighters keep throwing scooters from Bird and Lime into oceans, lakes and rivers. Public service employees are straining to fish out these venture capital funded wonders out of the water. In Phoenix, self-driving Waymo cars are getting their tires slashed and assaulted by gun-wielding road-ragers (Mad Max style, we assume). All that to say that the human element in this story is allergic to being entirely prodded, measured, and automated away. Can politics catch up with SoftBank's Vision Fund, which could build Trump's wall 20 times over? We hope so.

d2233e7a-7e6d-4bc2-811d-f7328d39763e[1].jpg
d68567b1-b834-4a08-b05f-7ad055f5a3d6[1].jpeg

Source: DigIn (Softbank), Gizmodo (Ofo), Slate (Bird), Business Insider (Waymo)

2019 FINTECH PREDICTION: Government and Enterprise Platforming, led by AI and Mixed Reality

Source: Images from Pexels,     2019 Keystone Predictions Deck

Source: Images from Pexels, 2019 Keystone Predictions Deck

Over the last decade, consumer tech has undergone a cycle of platform building, user aggregation, data mining, and value extraction, resulting in GAFA monopolies. Exhaustion with Facebook and the adjacent issues of privacy and radicalization, in our view, will lead to problems building new splintered consumer attention platforms for AI, AR/VR and other new media ground up.  This implies that consumer platforms based on new technologies will be much more long-tail oriented, serving niche markets with very strong fit. Communities may be passionate, but smaller.

Enterprise tech lags retail adoption by, give or take, 5 years. Similar platforming has not fully penetrated on the enterprise side -- Salesforce is not yet the AI monopoly we should all fear, and Open Banking is barely a fizzle. Therefore, we expect increasing data transparency, aggregation and monetization to occur in enterprise underwritten by venture capital investors. As an example, augmented reality adoption and economics will be driven primarily by municipalities, utilities, large industrial manufacturers, and the military. Similarly, artificial intelligence at scale (and its meeker cousin Robotic Process Automation) are to be directed largely at the workflows and manufacturing processes of large corporates. Dont' get us wrong -- consumer AI is extremely important -- but within Financial Services, the scope for this in the corporate world is even larger.

The corollary is that the pricing pressure that started in consumer Fintech -- roboadvice (150 bps to 25 bps) or in remittance (600 bps to 10 bps) -- will spill over into B2B banking, money movement, insurance, treasury management and product manufacturing. An inevitable outcome is pressure on profit margins as prices equilibriate. For those companies that are able to re-design operations using a digital chassis, they will be able to compete on the margin with Fintech unicorns. Those that are not should exit, or retreat into more bespoke, relationship-driven business lines. 

ARTIFICIAL INTELLIGENCE: Morgan Stanley, Yext and Chinese AI-first Apps.

A point is not enough. It takes two points to make a trend-line, at least in a two dimensional space. One of the muscles we try to flex often is to connect points in different sectors and themes to see the limits of the possible. Let's contrast the following: (1) Morgan Stanley partnering with Yext for financial advisor business pages, and (2) Andreessen Horowitz' commentary on Chinese consumer artificial intelligence applications on a path to capture the hearts of teenagers everywhere. Disparate, funky, and painfully obvious.

About ten years ago, "hyper-local" became a venture catchphrase. News would go from being general to local, video would go from main-stream to niche, and so on, contextualized by the GPS in our pockets. Yext is a company that won one of the battles for hyper-local content by building the retail knowledge graph that gets printed on Google Maps. Simply, if you see a business listing for a laundromat on your Maps app, likely the app provider is licensing local data from Yext. This data then scales up into pre-made business websites, analytics, and customer funnel conversion. Morgan Stanley inked a partnership with this scale content manager to give their 15,000 financial advisors a digital presence. Controlling and printing out that content at scale, with embedded compliance and into every Google/Apple phone, is hard and smart. And perhaps physical presence is the main value of a human advisor.

Now for Chinese AI. Unlike Americans, with their hand-wringing about privacy, choice, and human agency, Chinese apps don't care. The next generation version of Instagram and Snapchat is called TikTok, and the storied venture firm Andreessen celebrates them for taking away any human choice in what content a user would see. The algorithm is not a search support tool, it is the only and ultimate arbiter of where your attention goes. And it tends to make kids happy (unlike Youtube, which generally makes them into Twitter trolls). 

So let's mesh these things together. A financial services version of TikTok with a Yext overlay would be an app that is tied to the physical world, perhaps through Augmented Reality or just simple Maps, that would decide for you which financial provider to find. It would know that you still want to talk to a person for that emotional connection, and would find one that's closest geographically and a best-fit emotionally -- a two factor optimization problem for an AI. Yext financial advisor reviews, combined with a Morgan Stanley risk/behavioral client questionnaire could do this. Thus the TikTok aspect kicks in, with the human in the loop simply being a form of physical content marketing, gaming the algorithm with a meatspace presence. 

98edf5ec-17a0-419f-95db-a4e668a4d7e5[1].jpg
3c004f7a-4647-426c-9c4d-95eb75713c85[1].png

Source: Finextra (Yext), Andreessen Horowitz (AI apps), FactorDaily (App downloads), 

ARTIFICIAL INTELLIGENCE: Apple trying to catch up in conversational interfaces through privacy

Apple acquired Silk Labs, an AI startup with significant tech pedigree, whose tagline is to "embed instant cognition into your next product". We have to respect the science fiction marketing, of course. But we also respect that the machine learning solutions from this company allow machine vision, sound recognition and natural language processing to be done locally on a particular device. That means that a specific device that you use for conversational interface interaction will be locally better at understanding you -- rather than some giant squid-like monster AI hosted on Amazon Web Services. And of all the tech companies, Apple is the most credible in its claim to protect your privacy on the iPhone, with such an acquisition potentially powering other edge-computing / Internet of Things products.

Edge computing is the concept that there are lots of unique distributed smart devices scattered throughout our physical world, each needing to communicate with other humans and devices. Two layers of this are very familiar to us: (1) the phone and (2) the home. Apple has become a laggard in artificial intelligence -- behind Google on the phone, and behind Amazon and Google at home -- over the last several years. Further, when looking at core machine learning research, Facebook and Google lead the way. Google's assistant is the smartest and most adaptable, leveraging the company's expertise in search intent to divine meaning. Amazon's Alexa has a lead in physical presence, and thus customer development, as well as its attachment to voice commerce. Facebook is expert in vision and speech, owning the content channels for both (e.g., Instagram, Messenger). We also see (3) the car as developing warzone for tech company gadgets.

Looking back at financial services, it's hard to find a large financial technology provider -- save for maybe IBM -- that can compete for human attention or precision of conversation with the big tech firms (not to mention the Chinese techs). We do see many interesting symptoms, previously covered in our Augmented Finance analysis, like AllianceBernstein building an AI-based virtual assistant for bond traders, but barely any compete for a relationship with a human being in their regular life. The US is fertile ground for this stuff, because a regulated moat protects financial data from the tech companies. Is there room for a physical hardware financial assistant in your home? How much of your financial life would you delegate to some*thing* that decides how you should live it?

368895aa-acb4-4856-87e2-119bef79f727[3].png
073ac488-611b-44c4-86ff-67641d64844a[1].png
29e979c0-e59a-44a5-b7b0-318bc1c10945[1].jpg

Source: Silk Labs, Bloomberg (Bond Bot), TechCrunch (Google to be nicer if you say please), Autonomous NEXT (Augmented Finance), USA Today (car AI), Voicebot (Install Base)

ARTIFICIAL INTELLIGENCE: Facebook's Lasso clones Asia's TikTok to grab 500 million users

ByteDance is a $75 billion AI-powered Chinese attention gathering machine. Their marquee application TikTok -- a frankenstein formed from the combination of Vine-like videos and the acquisition of Musical.ly -- boasts 500 million users, and is currently ranked the #6 free app in China and #7 in the US. That position is ahead of Facebook (surely angering comic book supervillain Mark Zuckerberg), Snapchat and Messenger, having achieved this result in mere months since launch in the US market this past July.

TikTok engages teenagers with personalized content driven by ByteDance’s proprietary machine learning algorithms, emoji video commentary features, Snapchat-like augmented reality renders, and glitchy filters. Creators on the platform have the chance to make viral content, which is distributed at scale and mass-targeted at consumers by a machine. Using AI this way is a growing strength for Chinese companies. It is also a strength of recently beleaguered Facebook, which is fighting back by launching a clone called Lasso. The app features nearly identical gesture features, structures, endless content feeds, and hashtag groupings for browsing. The main differences lie in (1) video creation, where TikTok offers up to 60 second videos compared to Lasso’s 15 seconds, and (2) TikTok’s ability to customize content using filters, music, and lenses, which far outweigh the limited selection of Facebook's Lasso.
 
Two conclusions of note. First, Facebook has defended their turf before, and succeeded. For photos, it outright bought Instagram. For video stories, it failed at buying Snap but succeeded at building the feature into Instagram. For messaging, it bought Whatsapp and built Messenger. We wouldn't count it out in this case either. Second, these attention companies exist to deliver advertising and form consumer preference functions. In China, data about customer preferences already informs access to financial services, such as credit, payments and investing. In the US, increasingly Facebook is seen as a conduit for opinion manufacturing to the highest advertising bidder, with such data still a step away from being included in a financial underwriting decision. Yet as tech solutions and norms are exported between global jurisdictions, we expect that line to increasingly bend.

bc46086e-b155-4367-8e38-38e966cd99f6[1].jpg
4d8a5b9e-11da-4112-97fd-cb84ec0a8f6a[1].jpg

Source: TikTok, Lasso, Slate on Facebook's Recent Allegations

ARTIFICIAL INTELLIGENCE: Self-driving cars and self-speaking news anchors inch us closer to dystopia.

Financial services regulators have been so hard on crypto currencies, roboadvisors, digital lenders and payments companies. It's as if that money is a life or death situation! But getting a permit to drive a robot car on a public road without a human being holding the wheel -- not a problem in California. Waymo, which is the Google car spinout, has been given the green light to put 40 autonomous cars on the road. This is already happening in Arizona, with 400 users that can get into a robot car via an app around Phoenix.

3dc3f300-8696-4d51-ae39-a4581f08f17d[1].jpg

We don't want to be alarmist, of course. Statistically, these machines are likely much better than humans at driving -- they are just more likely to make mistakes that humans would think are preventable. The same process took place in regards to machine vision, with early prototypes making classification mistakes between cats and dogs; now, such algorithms can tell apart the difference between hundreds of breeds. So we hope to see similar progress as driving and visual data is incorporated into autonomous car systems. We'd be remiss not to mention our white paper on the topic, which models out how the insurance industry may lose its lunch when cars don't crash. On the other hand, we note that the DMV required a $5 million bond to put a self-driving car on the road, so the risk is still wildly unknown.

fd5430d6-f02d-4862-8d0e-321cf43a30af[1].png

In a more sinister move, China's state-owned news agency recently launched "composite anchors", which is a machine vision version of a news anchor that can be manipulated with text. Here's how it works. You shoot dozens of hours of video of a person speaking, and then spin up neural networks that can (1) manufacture sounds similar to the target's speech and (2) manufacture video resembling the human making that speech. Presto -- just type in whatever into a command box, and your generated anchor will say it, in any language you would like. Given the recent video editing experiments that the White House supported in relation to denouncing a journalist, we are acutely terrified of how this can impact the attention economy. Not to mention the implications for selling a human likeness for endless manipulation. 

ad628a3f-c9da-4f6b-8136-24ca748f5c44[1].png

Source: SF Chronical (Waymo), South China Morning Post (AI Anchor)

ARTIFICIAL INTELLIGENCE: From BMO's Chatbot to a full virtual avatar

Let's paint the progression. Finn.ai, a chatbot company that integrates banking services into Facebook Messenger and other chat channels, launches Bolt for the Bank of Montreal. The project took 10 months to private label and deploy to BMO's 12 million customers, covering 250 questions at launch. As the app gathers more information about what customers ask, its usefulness grows and it becomes an increasingly relevant channel for customers to ask their financial institution informational questions.

bd55987e-762a-40c2-97ba-066f2fd58f5f[1].png

By the end of 2017, 40 million smart speakers were installed worldwide, with 2018 projected to land at a 100 million install base. People are getting more than one instantiated smart assistant -- littering their home with several Echo dots. And, reportedly, Alexa lives in 3,000 others types of smart-home devices -- giving this bot army 45,000 skills, from Spotify to financial conversations. Google and Apple are working to catch up to these numbers, releasing eerily realistic robo-conversationalists like Google Duplex that can answer telemarketing calls and book hair salon appointments. And maybe cancel your financial subscriptions. What's perhaps most important is that an answer to a voice query is 40-80% fulfilled by the "featured snippet" at the top of a search engine's list, according to Martech. That means no more long tail of any kind, full stop.

c12bc0cc-7c56-4e78-93a0-656a2d02b04d[1].png

Eventually, it is no longer enough for our avatars to be disembodied functions powered by a retail product recommendation engine. From virtual worlds and into augmented reality, agents will take on hyper-realistic rendered physical bodies. A recent Intel whitepaper describes how machine learning is now being combined with a modeling of animal and human bodies under a physics simulator to quickly and realistically build CGI for games and movies. Instead of an artist using intuition to draw the perfect frame, machines build skeletons with physical properties, connect bones with digital ligaments, fire up virtual muscles, and package the final version in the species of choice. Machine learning algorithms add realistic, generalized movement to the equation. They can also add speech, appearance and function. And maybe even help the UBS chief economist look a little bit more lifelike!

f0b0778c-0de0-4ee2-8466-25dc54c01e18[1].png

Source: Finn.ai BMO case study, Fast Company (Amazon AI), The Atlantic (Smart Speakers), Martech Today (40% answers from featured snippet), Intel (Rendering CGI), Financial Review (UBS CIO)