Big Ideas 2022 Report Is HereDownload Now

1. Automation Should Cause a Shift From “Unpaid” to “Paid” Labor 

While automation typically causes debates about the potential loss of jobs, ARK’s research suggests that it could transform “unpaid” labor into “paid” labor. As shown below and in our Big Ideas 2020, grocery shopping and food preparation often are “unpaid” tasks. Historically, preparing food at home has been less expensive than eating food prepared away from the home. Now, however, automation systems are increasing productivity at fast-casual restaurants, suggesting that they should be able to cut prices and maintain profitability. As a result, grocery shopping, food preparation, and clean-up should shift from “unpaid” tasks in the home to paid jobs at fast-casual restaurants, other restaurants, and their suppliers.

The trend from “unpaid” to “paid” labor is not new. Between 1950 and 2000, the number of people working on farms dropped by ~6.75 million, more than 80% of the decline in jobs classified as unpaid family workers.

ARK believes that autonomous vehicles will cause another significant shift from unpaid to paid labor. Few of the ~228 million people with licenses in the US are paid to drive today. Autonomous taxi networks could transform these “unpaid” work hours into jobs, many of which we can’t imagine right now. In the early days of the worldwide web, for example, who could have imagined the jobs created by ride hailing? Uber and Lyft could not exist without the Internet.


2. Spotify is Betting Big on Podcasting 

During the past year, Spotify has tripled the number of podcasts on its platform, now offering more than 700,000 titles. That said, it still is working on monetizing its burgeoning library.

While advertising is the primary revenue stream for most podcasts, Spotify captures very little of it. Typically, podcast hosts source and read ad spots, which means that Spotify benefits only from content that it owns.

The host-read advertising model is problematic for podcast platforms. While enabling a seamless flow between podcasts and advertisements and listener loyalty, it is difficult to scale. As a result, podcasts currently are monetizing at only $0.02 per listener hour which is roughly 15% the rate of radio.

In an attempt to bolster its podcast revenues, Spotify plans to insert targeted ads into its exclusive content, stripping out hosts’ voices and perhaps tuning listeners out during ads. As a more effective solution that also could scale, perhaps Spotify should create and train an AI program to mimic hosts’ voices, maintaining the seamless flow of host-read ads.

If this sounds futuristic, in 2019 Dessa created an AI that mimicked the voice of Joe Rogan, who hosts the most popular podcast in the world.


3. Microsoft’s 17 Billion Parameter T-NLG Is the Most Powerful Language Model Yet 

The neural nets arms race has escalated with Microsoft’s release of Turing Natural Language Generation (T-NLG), a transformer-based model with 17 billion parameters—ten times larger than OpenAI’s GPT-2 from a year ago. More parameters allow neural nets to learn, remember, and perform complex tasks.

Trained on large datasets from the internet, T-NLG has learned not only how to read and write in English, but also how to answer questions such as when World War II ended and how large the population of the United States is. Most impressive, it does so without access to structured knowledge data bases or grammar rules. The neural net learns mostly from large troves of unstructured, unlabeled data.

While Google designed Meena AI to generate human-like dialogue, T-NLG can generate human-like long form writing. In one example, T-NLG “read” a long blog on Microsoft’s goal to be carbon negative by 2030 and generated an executive summary in pristine English.

Large models like T-NLG are too expensive to deploy at scale today, but they will evolve into Siri, Alexa, and other productivity-enhancing applications in the coming years. Soon, we believe the fundamental tasks of reading and writing will be mediated and enhanced by AI.


4. Intercontinental Exchange Acquires Bridge2 Solutions 

This week, Intercontinental Exchange announced that it is acquiring Bridge2 Solutions, a company that operates 4,500 loyalty programs in the US. After the deal closes, Intercontinental Exchange’s subsidiary Bakkt will purchase the company. In an interview with Fortune, Intercontinental Exchange CEO Jeff Sprecher made comments that shed light on Bakkt’s consumer strategy: “The legacy payments infrastructure is ripe for disintermediation… Companies are always interested in buying back points at the right price.”

Based on this news we believe Bakkt’s consumer app will be more of a mobile wallet than we initially suggested. Bridge2 is testing a product called Loyalty Pay which allows users to pay for items with points from their loyalty rewards balances that can be converted into dollars at the point of sale. Facilitating a more convenient way to buy back points could help companies manage balance sheet liabilities and give consumers a more immediate source of liquidity for their points. In related news this week, loyalty program provider Fidelity Information Services announced a partnership with PayPal that allows loyalty program participants to pay for purchases using PayPal’s Pay with Rewards at millions of online PayPal merchants.

Last week, in addition to the acquisition of Bridge2, Intercontinental Exchange announced during its earnings call that it had considered acquiring eBay but has abandoned the idea. Given the consumer focus of the Bakkt app, Intercontinental Exchange was interested in tapping into eBay’s marketplace and the 180 million annually active buyers in the eBay and StubHub ecosystems. Key to the success of mobile wallets such as Cash App and Venmo has been their ability to acquire customers at a significant discount to traditional banks.

We look forward to trying the new features Bridge2 will bring to the Bakkt app when it launches!


5. Where are the Investors in SoftBank’s Second Vision Fund

This week, SoftBank reported a $2 billion operating loss for its Vision Fund in the fourth quarter, blaming unrealized losses in WeWork and Uber. While less than the $8.9 billion in operating losses incurred during the third quarter, the Vision Fund’s performance falls woefully short of the “trillion-dollar gift”, the ten-fold return on the $91.3 billion Vision Fund that Softbank CEO Masayoshi Son once promised.  Recently, Vision Fund reported $10 billion in profits cumulatively in its four year life. Given its minimum lifetime of 12 years, Vision Fund has eight years to make up the $990 billion gap.

Vision Fund 2 isn’t faring any better. Last July, SoftBank announced 13 “expected LPs” for Vision Fund 2, but they seem to have disappeared after the implosion last summer of WeWork, one of Softbank’s largest investments.  With less capital available and a new-found focus on profitability, other unicorn valuations within the Vision Funds could be challenged during their next funding rounds.

Now SoftBank is battling not only executive employee churn but also activist investor Elliot Management who is pushing for share buybacks. Buybacks could inhibit SoftBank from allocating more capital to Vision Fund 2.