What Finance can learn from the Evolution of Beautiful Artificial Intelligence Art; plus 14 short takes on top developments
|Jul 22||Public post|
Hi Fintech futurists --
In the long take this week, I take a journey from Michelangelo's Sistine Chapel, to the impact of Photography on Portraiture, to Artificial Intelligences building generative art on blockchain networks. We take stops along the way to see analogies to financial services, and to notice how expert craft turns to code.
The latest short takes on the Fintech bundles, Crypto and Blockchain, Artificial Intelligence, and Augmented and Virtual Reality are below.
Thanks for reading and let me know your thoughts by email or in the comments!
Enough Finance, let's talk Art. I promise it will swing back to Finance anyhow. One of the key takeaways that keeps spilling out of the conversation around Artificial Intelligence is whether humans will be augmented or replaced by software. Most of the time, these transhumanist discussions are science fiction hand-waving. But let me walk you through some tangible examples, so you can see and feel the distance traveled.
Let's say it is the 1500s, and you'd love a portrait of your family, or an image from your favorite allegory to behold once in a while. Since it is the Middle Ages, not everyone can just fire up Blade Runner (my favorite allegory) or take DSLR photos of their cats. If you are a locally powerful sovereign in the local feudal community, you might be able to hire a craftsman like Michelangelo to paint the ceiling once every twenty years. For your trouble, you will get the following masterpiece!
Of course, you are not really hiring Michelangelo alone to do all this work. He comes with a guild of apprentices and professionals that scale up the capabilities of his practice, deploying Michelangelo's style and leveraging his reputation to do illustrations all over Europe. You are welcome to analogize this to a successful investment banker doing deals all over Silicon Valley -- a craftsman with years of experience and a strong reputation using teams of underpaid interns to immortalize people in power.
However, mastery is not immune to automation. As a profession, portraiture melted away with the invention of the Camera, which in turn became commoditized and eventually digitized. The value-add from painting had to shift to things the camera did *not* do. As a result, many artists shifted from chasing realism to capturing emotion (e.g., Impressionism), or to the fantastical (e.g., Surrealism), or to non-representative abstraction (e.g., Expressionism) of the 20th century. The use of the replacement technology, the camera, also became artistic -- take for example the emotional range of Fashion or Celebrity photography (e.g., Madonna as the Mona Lisa). The skill of manipulating the camera into making art, rather than mere illustration, became a rare craft as well -- see the great work of Annie Leibovitz.
But nothing is sacred. As we moved into the Machine Age, photography was democratized through the shift to digital cameras and smart phones. This meant that the populace at large could generate endless visual imagery, and perfect their selfi skill-sets. Home photo albums grew from a few dozen mundane, poorly composed pictures, to millions of images immaculately designed and filtered for social media. Today, every single person on Tinder and Instagram is a fashion photographer. They are also their own self-immolating Madonna. It is a heavy burden.
Let's analogize back to Finance. It used to be that complex financial products, like derivatives, real estate investing, global tactical asset allocations, and foreign exchange arbitrage were built for institutions and wealthy people. Today, it is pretty much trivial to access all of these products in any corner of the world. Roboadvisors in 2019 will give you Goldman's 2006 asset allocation for pretty much free. Neobanks will provide institional-level intererest rates (almost the Fed rate!) to the smallest accounts. Crowdfunding sites have introduced real estate and high yield bond markets to regular people. And if you want to see FX arbitrage or high frequency trading at play levered up 100x, just go to Bitmex. Today, you are empowered to have the misfortune of figuring out your own investment and retirement strategy. For most people, this is as much fun as trying to make it as a YouTube influencer.
So now we come to Data. As web apps filled up with our millions of selfies, cat photos, and selfies with cats, humanity's engineers figured out one more trick. We could run thousands of simultaneous regressions across all these images to create self-managing algorithms that identified the subject matter of the content. And as the math classified what was already in the images, it also learned how to hallucinate the underlying structure of the world. By scanning through the history of human visual arts, the machine learned the styles of different artists -- and how to project them as filters on any image you decided to choose. An artist in your pocket from any moment in history!
This is roughly where we are with financial services Artificial Intelligence today. Take for example a digital lender that uses AI to maintain an underwriting model for personal loans. European core-banking company Temenos just bought a startup called Logical Glue to build exactly this feature inside its broader platform. By studying past human judgment in lending, literally teaching the machine through supervised learning on existing data sets, the software is replicating the "style" of a prior underwriter. Perhaps the underlying picture will change -- but the overall vibe, and hopefully the associated default rate, will stay a masterpiece. Lending Club charge-offs below illustrate the point by analogy.
But the story does not end here. New media art has been thriving over the last several decades, as programmers learned to write code that visualized beautiful and complex algorithms, which in turn could be rendered by increasingly more powerful hardware and websites. Artists like Holger Lippmann created generative systems that would yield infinitely variable patterns, fractals, and other gorgeous designs that take your breath away. Still, these are highly engineered, mathematical outcomes. Like the case of Michelangelo earlier, they are shaped by an expert craftsman using modern tools. You can compare Lippmann to the billionaire mathematician financiers behind Two Sigma, or D.E. Shaw, who code the quantiative fractal investment strategies to beat markets.
Today's innovations of Artificial Intelligence and Blockchain are opening up a new frontier for the machines, and their implications for our creativity. Neural networks -- the math that powers the style transfer I referenced earlier -- can now create their own generative outcomes based on the millions of images of photos and visual artifacts that the Web has fed them. Take for example a project called Ganbreeder, which allows users to traverse mathematically between different types of objects in a visual space (e.g., an image could be half fish, a quarter truck, and a quarter castle). The network can be used to hallucinate forward music covers, landscapes, or human portraits. The below illustrations were not done by a human hand, other than the coding of course.
These infinite realistic, abstract, or Surrealist images are not all equally valuable. Some are gorgeous and should be saved, while others are redundant and uninteresting. This is where blockchain networks add their magic. Check out this great article on Artnome about the attempts to make scarce and commercial digital art by tokenizing its ownership. If the machine authors a beautiful thing -- or perhaps if you discover it during your travels through its alien landscapes -- how do we record provenance and property rights? How do we value, exchange, gift, create reproductions, or destroy the thing itself?
Here are a few projects working on the issue: SuperRare, KnownOrigin, Portion, RareArtLabs, DigitalObjects, Crypto Punks, Dada.NYC, CurioCards, Pixura, and Freeport. By tieing each piece of digital art to a blockchain collectible, scarcity allows for economic activity between human beings.
Determining what is valuable can be human work, as in the case of Ganbreeder. Or perhaps it is the work of mathematical algorithms as well! One such example is a social media experiment called Archillect, which scrapes the social web for highly unusual images with strong user engagement. This has earned its Twitter account over 1 million followers, and I recommend you check out the results here. Perhaps in Finance, this would be a highly successful momentum trading bot, targeted at the most discussed companies on social websites. Or perhaps it is the trade-copy Fintechs like Covestor or Gimmer.
We are getting a bit long in the tooth with this entry, so let me land us in the natural conclusion of these developments. Gene Kogan, one of the pioneers in creative AI, has just started work on a generative neural network that makes digital art, and bundles it with a blockchain-based marketplace. Another similar project is Artonomous from Simon de la Rouviere (using a simpler procedural engine), yet to launch. Participants in the training of the AI artist get economic rewards, the artist's outputs are saved as crypto collectibles, and there is potential to leverage all the decentralized financial services that exist on the Ethereum blockchain.
You could, for example, try to use generated art objects as collateral to get margin loans from MakerDAO. Or, perhaps, you would exchange the works using decentralized exchanges. When the economic value of such artwork is trivial, the concept is not particularly compelling. But once you realize that traditional paintings (with a record of a sale of about $450 million) will follow a similar route, things become interesting.
In the financial services world, Numerai is the main comparable that comes to mind. That team runs machine learning competitions on a common data set, using the winning algorithms to generate alpha in the stock market, and pays out participants in a proprietary cryptocurrency. But I think such an approach is too greedy -- why should one hedge fund get to monopolize the benefits of all that math? Far more interesting would be a Decentralized Autonomous Organization that has a participatory rewards model, such as 20% of carry, for high quality AI-based trading algorithms that pass a certain threshold of quality. If such an open source projects does appear, it may be possible for it to accrue massive returns to scale. That is, if it isn't stomped into the ground by regulators first.
I've done a couple of talks that were published last week. You may be interested in the following -- just click the banners.
Charles Schwab in Talks to Buy USAA Wealth-Management, Brokerage Units. Leading trading firms are moving from capital markets access for self-directed investors to offering wealth management and financial advice. What is the longer term trend? There is no value in retail trading, and there is no value in ETF manufactuing -- unless you own the customer and have fantastic scale. I'll throw in a couple more links into this category: Merrill Edge Mulls Subscription Pricing, Voya Adds Robo To Help Advisors Bid For Younger, Less Affluent Clients.
N26, the online bank backed by Peter Thiel, is now worth $3.5 billion. I could link to their P&L and make comments about how 3.5 million customers each valued at $1,000 of enterprise value is phychotic. But then I would be a tinfoil hat crazy persona yelling into the wind. We wouldn't want to be the crazy ones -- leave that to Softbank.
Plaid Rolls Out Liabilities Tool To Track Student Loan Debt. This is interesting in the context of data aggregation -- especially as lending moves from a product destination to a point of sale feature. I'd expect to see a SynapseFi clone product here soon enough. Or you know, more of Goldman: Goldman Sachs Invests €25M in Raisin.
Real Estate Fintech Roofstock Tops $1.6 Billion in Transactions. Three years to generate a pretty meaningful amount of democratized investment in single family homes. I like to think about what impact this is having on retail asset allocations. And of course, how much more powerful this is once tokenized.
Cryptocurrency Crimefighter Chainalysis Becomes First Blockchain Company To Make Next Billion-Dollar Startups List. Interesting data points about the company -- $53 million in venture funding, $8 million in 2018 revenue. That's not Binance's $250 million of trading revenue per quarter, but you know, it's a bit higher quality as far as turnover goes.
Mike Cagney’s Figure Technologies Completes $20 Million STO on Provenance Blockchain with Help of Dechert. So in addition to raising $120 million of traditional VC equity for Figure, the company sold $20mm of its proprietary Hash tokens (which are treated as securities). Any competent entrepreneur that is not using blockchain building blocks today is leaving opportunity on the table. In a similar vein, SDX Will Execute Digital-Asset Trades. The Switzerland stock exchange plans to launch digital asset trading within 12 months, using R3 Corda technology. Another comparable initiative is Singapore’s Stock Exchange Backs New Ethereum Security Token Platform, build by ConsenSys on Ethereum. This stuff will only accelerate.
Tether USDT Is Launching on a ‘Pure PoS’ Blockchain, Algorand. Tether is becoming a native cash sweep in pretty much every meaningfully adopted blockchain, including ones that have not launched yet. As long as Asian investors participate in crypto, and the legal trouble remains "unsolved", this will keep moving forward. I'd love to see cross-chain DAI -- maybe it exists already!
Temenos Pushes Into Artificial Intelligence with Acquisition of Explainable AI Platform. £12 million for a company that provides a neat feature for banks -- out of the box machine learning-based loan underwriting software. I sort of think this is more about checking the box, and then charging banks integration fees to set up on their proprietary data, but at least there's an AI box now.
RiskGenius announces Series B, partnerships with trio of world’s largest insurance carriers. These guys use natural language processing to help commercial carriers (e.g., Liberty Mutual) to review policies and underwrite more quickly. It's a good success story, reasonably priced and useful.
Data governance in the age of AI: Beyond the basics. Nice, long read on how to get things done -- those things being actually implementing AI in large, stubborn insurance organizations. Short answer is Human Leadership.
Carsharing Platform, Turo, Raises $250M And Becomes A Unicorn. Here's your AirBnB for cars, everyone. As cars become smart, we don't necessarily "own" them -- that's too broad a brush. Rather, we know when property rights are being applied to them, and by whom.
The VR Revolution Is Alive and Well, It's Just Not Ready For You. A nice recap of the state of the VR market today. While shipments are at a healthy 10 million-ish per year, it still feels like there hasn't been a dent in consumer adoption. This article makes the same point I have, which is that early VR is for enterprise (and government).
Microsoft Uses HoloLens 2 to Demo Multi-Language Speaking Avatar That Looks Just Like You. It is going to be trivial to render high-resolution high-detail avatars of ourselves speaking in various languages and styled across different environments. This is one the first steps.