Mentioned in Video:
- Netflix Research: https://research.netflix.com/
- How Netflix Uses AI, Data Science, and Machine Learning – From A Product Perspective: https://becominghuman.ai/how-netflix-uses-ai-and-machine-learning-a087614630fe
- Netflix's Hidden Categories: https://mashable.com/article/netflix-search-codes
- Tesla plans to offer machine-learning training as web service with its new ‘Dojo' supercomputer: https://electrek.co/2020/09/21/tesla-offer-machine-learning-training-as-web-service-dojo-supercomputer/
- ARK Invest's Price Target on Tesla: https://ark-invest.com/analyst-research/tesla-price-target/
- #SolvingTheMoneyProblem video on Tesla and Tencent: https://www.youtube.com/watch?v=ziprHJjngFY
- Roku Leaves Rivals in the Dust: https://blog.pixalate.com/ott-ctv-programmatic-ad-device-market-share-roku-2019 https://rethinkresearch.biz/articles/roku-leaves-rivals-in-dust-claiming-machine-learning-breakthrough/
- Machine Learning at Spotify – Gustav Soderstrom | AI Podcast Clips (Lex Fridman): https://www.youtube.com/watch?v=_TWgsvF4hBQ
- Support the channel and get extra member-only benefits by joining us on Patreon: https://www.patreon.com/tickersymbolyou
What do #Tesla stock (#TSLA), ROKU, and Square (SQ) have in common? #ARKW, #ARKInvest‘s next-generation internet fund, focuses on artificial intelligence, big data, cloud computing, and more. Every company in this #ARK fund can quickly learn about YOU in order to provide a service that used to require a human expert to complete.
The world is changing.
New innovations are being delivered faster than ever before. The lines between the physical world and the digital one are boring, and so too are the lines between advanced technologies. Where one robot ends and where another begins is no longer so obvious. Where one innovation ends and the next begins is no longer obvious either. The most disruptive companies are no longer delivering single products or services. They are delivering change. And, one day, where technology ends and where we begin may not be so obvious either.
It's important to remember that ARK Invest are thematic investors actively managing funds themed around blockchain technology, DNA sequencing, energy storage, robotics, and artificial intelligence. While the graphic on the left shows these innovation pillars as separate, the one on the right shows that they are beginning to overlap more and more. This increasing overlap is what Cathie Wood is talking about when she says that these disruptive technologies are converging with each other to spawn new innovation.
And importantly, they are converging with each other to spawn new innovation.
As you can see from ARK Invest's cluster of major innovation platforms, artificial intelligence actually sits at the center of this convergence and touches the other four. In the future, I expect this artificial intelligence bubble to get bigger and bigger as companies working with every other innovation platform learn to use A.I. in new and exciting ways. The reason this matters is because a breakthrough in artificial intelligence can affect all other innovation platforms. In my video on Deep Learning, I talk about Google DeepMind's breakthrough in protein folding and mapping and how lessons learned from that might bleed into the state of the art in other seemingly unrelated areas.
Artificial intelligence, which acts as a bridge between things like online finance and energy storage and industrial robots and genome sequencing makes that possible. So if you're interested in hearing Cathie Wood's, Elon Musk's or simply my commentary on the convergence, that video covers the highlights. Whenever I'm researching a company, one thing I like to do is map it onto these charts and ask myself if I could see ARK Invest holding it, which funds it would go in and how much ARK Invest might hold. Since Tesla is ARK Invest's largest holding by far,
let's see how it would map onto these clusters. Here's how I would weigh it on here. In my opinion, Tesla is a world leader in energy storage, industrial robotics and various fields within artificial intelligence, all three of which are necessary to unlock full self-driving capabilities, launch their autonomous ride hailing network and provide the world with mobility as a service. Tesla is also the biggest holding in ARKQ, ARK Invest's Fund themed around the autonomous revolution. In my video on ARKQ, I talk about falling costs of industrial robotics, which drives the production costs and capabilities of Tesla's Gigafactories, the physical machines that build the machines.
Because Tesla builds computers on wheels and not cars, their products and services are mobile connected devices that are part of the Internet of Things. The Tesla Dojo program, which will offer machine learning training as a Web service, makes Tesla a serious player in the cloud computing space, at least for some important use cases. Tesla is the biggest holding in ARKW as well, ARK Invest's fund themed around the next generation of digital services. So this video will cover the digital side of Tesla's business and then speak to some of the other awesome companies in the fund.
The future of transport is autonomous, mobility as a service. Spanning from personal mobility, is a service where autonomous taxi platforms will make a point to point travel cheaper, more convenient and even safer to logistics and service, where autonomous electric trucks and drones will deliver goods for a fraction of today's cost. How will personal mobility as a service be cheaper and more convenient? Autonomous Taxi Networks should offer ridesharing services for about thirty five cents per mile, thanks to much higher utilization rates.
This is roughly half of what car owners pay to drive their vehicles today. Thanks to these compelling economics, consumers will likely stop buying personal cars and migrate into the autonomous mobility as a service market. And Tesla plans to release software that will enable a fully autonomous drive across the US by the end of 2017.
Tasha Keeney is ARK Invest's analyst, focusing on autonomous mobility and the mobility as a service opportunity. As you can see, Tasha has come out with a ton of great research and analysis that forms a big part of ARK Invest's investment thesis on Tesla. ARK maintains and updates a set of price targets based on 10 potential outcomes for Tesla over a five year time horizon, which I've linked in the description below. In the Golden Goose outcome, Tesla's got great margins, builds Gigafactories quickly and efficiently, and has largely solved the challenges associated with autonomous cars, successfully launching their autonomous network.
ARK Invest has four potential outcomes that include Tesla achieving full self driving and launching their autonomous ride hailing network by 2024, giving it about a 30 percent chance of happening. On the other end, we have the Black Swan cases, which include things like it going completely bankrupt or being unable to raise capital. Considering Tesla just raised an additional five billion dollars in September of 2020 and another five billion dollars in December of 2020 by selling shares at the peaks of its massive rally, my opinion is that these bottom two scenarios are a little less likely to occur than ARK Invest's table shows.
The cases where Tesla doesn't solve autonomy and launch the network by 2024 account for about 70 percent of the probabilities on this table. Discounting these two Black Swan events, I'd say that ARK Invest's 2024 price targets for Tesla account for about a one in three chance of launching its autonomous network by 2024. Tasha Keeney is a great resource for following ARK Invest's stance on Tesla, since their price target is currently largely based around full self-driving and autonomous ride hailing. I showed you that first clip to remind you that even ARK Invest can make calls that don't pan out when they expect them to.
It's important to remember that no one predicts markets, advancements in technology or crazy world events with 100 percent accuracy. The state of the art machine learning and the state of the world are much different today than when that video came out. In my opinion, Tasha Keeney and ARK Invest aren't wrong about the eventual mainstream adoption of mobility as a service. They've only been wrong about when. So let's quickly take a look at the main beats of Tasha's analysis on autonomy and then see how they've changed since that video.
Most miles driven today are done so in cities. Autonomous taxis will reduce the cost of those miles by about a factor of ten. When that happens, autonomous miles driven in electric vehicles will be substantially cheaper than miles manually driven and gas powered vehicles. That ten times reduction in cost should cause about a thirty times increase in demand. This is a pattern we see over and over between cost and demand, by the way. As costs lower to some critical tipping point, the total addressable market for a technology increases exponentially.
And because of the sharp rise in autonomous miles traveled via ride hailing, we'll see a decline in personal car ownership as well as miles manually driven.
We think most of urban driving will go autonomous. And that's where the majority of miles occur today in urban markets. So you can think of it as a two car household becoming a one car household, a one car household moving to zero cars. So if you live in a city, chances are you won't need to own a car. And you could just take an autonomous Uber every day to work if you wanted to. So I think for the companies that are attacking this, Tesla is our largest position in the space.
Tesla has a massive data advantage. So autonomous driving is a machine learning problem. For any machine learning problem, what you need is basically massive amounts of data to solve it. Tesla has billions of miles worth of driving data that it has the option to pull from. If you look at a player like Waymo, which is Google's autonomous project, they have 20 million cumulative miles in their entire library that they've ever driven on public roads. Tesla has access to more miles than that in a single day from its customer fleet.
And so this gives it a massive advantage. We think this could be a boon to the entertainment industry. For instance, you could have more Netflix subscribers if everyone's freed up to to do whatever they want in the back seat. Of course, you could be doing work, but I mean, if sort of planes are any indication of what people would do, we think most people probably want to watch TV. So it really just it gives back time to the consumer,
And what's interesting, sort of if you look at this on an economic perspective, that's really unpaid labor.
As you can see, besides updating a couple numbers and dates, ARK Invest's thesis on Tesla hasn't changed at all. They still believe the future of point to point travel is mobility as a service and that Tesla is the front runner in achieving it. Here's another look at Tesla's massive data advantage. By the beginning of 2021, they expect to have over five billion miles of autopilot data to train their full self-driving software on, putting them
billions of miles ahead of the nearest competitor. Waymo's 20 million miles is shown with the line in red, but it's not drawn to scale because they couldn't make the line low enough. Let's also unpack what Tasha said about giving time back to the consumer. The average commute in the United States is about 30 minutes each way, and the average person spends over an hour a day already on one subscription streaming service or another. And that time spent streaming is done more and more on mobile devices.
In fact, starting in 2019, more time is spent streaming on a mobile device, than non mobile. Anecdotally, about half my audience watches me on mobile. If autonomous driving frees up your commute time, it's not unreasonable to assume that that time will go to streaming based on what people already are doing on public transportation today. If Tesla solves the full self-driving problem and launches their autonomous network, they'll effectively double the amount of time people can stream content, half of which would be inside their own vehicles.
That's a huge market advantage for Tesla for two reasons. One, it's unlikely that other competitors will come out with a competitive solution right away, letting Tesla enjoy a big first mover advantage. And two, people inside self-driving Teslas are literally a captive audience for the duration of the ride. So Tesla can offer tailored infotainment exclusive to a specific person on a specific trip, making it much more likely that that consumer will choose Tesla's solution over other entertainment options. So Tesla isn't just using machine learning and artificial intelligence to drive the car, but to decide how to entertain you while you're in it.
That's why autonomous ride hailing forms such a core part of ARK Invest's thesis and price targets on Tesla and why it's the number one holding in every fund it's in, including ARKW. Through artificial intelligence, it will bridge three of ARK Invest's five pillars of disruptive innovation and disrupt trillions of dollars of market cap in the process. Tencent, a Chinese conglomerate that owns five percent of Tesla, is a stakeholder in a lot of social media, social commerce and entertainment properties, many of which I'm sure will be available through Tesla's infotainment systems, at least in China, the world's biggest urban auto market.
Both companies are in the top 10 holdings of ARKW, ARK Invest's next generation Internet fund because they're bridging the gap between the physical world and the digital one. For the second half of this video, let's flip the script where Tesla and Tencent are using artificial intelligence to change the face of transportation in the physical world. Companies like Roku, Netflix, Spotify, Square, Teladoc and Facebook are changing the way we engage with the digital world. Each one of these companies deserves at least one full video to thoroughly cover.
Let me know in the comments below if you're interested in me spending a full video covering a single company in the future, I read every comment and your voice matters. For now, I'll cover what they all have in common, which is that they're digitally native brands that are using mountains of data and artificial intelligence to provide you with the highly targeted, highly curated set of recommendations based on what they know about you and people who behave the most like you when interacting with their brand.
Let's take it from the top. Artificial intelligence is a science like mathematics or biology. It studies ways to build intelligent programs and machines that can creatively solve problems, which is something traditionally left up to humans. Today we have mostly narrow artificial intelligence, which means a given program is designed to do one task. I'm talking about applications like chat bots that try to resolve the most common issues with the service before routing you to real human tech support. I'm talking about automated labeling systems that tag images based on their contents,
like what color, size or even brand the logo is in it, or going back to Tesla's example, automatically labeling frames in a video based on their contents, like if there are pedestrians, obstacles, road signs, traffic lights and so on in them. Artificial intelligence is used to personalize and streamline manual tasks like finding a meeting slot that fits everyone's schedule, optimizing budgets, planning the best path for a robot to take across a warehouse floor, or the best path for a last mile delivery vehicle to take to fulfill its route in the cheapest way.
The companies in ARKW all rely on different areas within artificial intelligence to provide you with some service that an expert human used to have to do for you. Spotify, for example, started out by curating playlists for people acting like that one friend everyone has that always makes the best playlists for every occasion.
Spotify has, maybe you can correct me, but over fifty million songs, tracks and over three billion playlists. So the million songs and three billion playlists, 60 times more playlists. What do you make of that?
Yeah, so the way I think about it is that from a statistician on machine learning point of view, you have all these if you want to think about reinforcement learning, but you have this state space of all the tracks and you can take different journeys through this through this world. And these I think of these as like people helping themselves and each other, creating interesting vectors through the space of tracks. And then it's not so surprising that across, you know, many tens of millions of kind of atomic units, there will be billions of paths that make sense.
And we're probably pretty, quite far away from having found all of them. So kind of our job now is users. When Spotify started, it was really a search box that was, for the time, pretty powerful. And then I'd like to refer to it as this programing language called playlisting, where if you as you probably were pretty good at music, you knew your new releases, you knew your back catalog, you knew your Stairway to Heaven, you could create a soundtrack for yourself using this play, listening to all that's like metaprogramming, language for music to soundtrack your life and people who are good at music.
It's back to how do you sell the product for people who are good at music. That wasn't actually enough. If you had the catalog and a good search tool, you can create your own sessions. You could create really good soundtrack for your entire life, probably perfectly personalized because you did it yourself. But the problem was most people, many people aren't that good at music. They just can't spend the time. Even if you're very good at music, it can be hard to keep up.
So what we did to try to scale this was to essentially try to build you can think of them as agents that this friend that some people had that help them navigate this music catalog. That's what we're trying to do for you.
How can playlists be used as data in terms of machine learning and just to help Spotify organize the music?
So we found in our data not surprising that people who play listed a lot, they retain much better. They had a great experience. And so our first attempt was to playlist for users. And so we acquired this company called Tunigo of editors and professional playlisters and kind of leveraged the maximum of human intelligence to help build these vectors through the track space for people and that broaden the product. Then the obvious next. And we you know, we used statistical means where they could see when they created a playlist,
how did that playlist perform? You know, they could see skips of the songs, they could see other songs perform, and they manually iterated the playlist to maximize performance for a large group of people. But there were never enough editors to playlist for you personally. So the promise of machine learning was to go from kind of group personalization, using editors and tools and statistics to individualization.
Roku has figured out how to do the same thing, but with advertisements. Specifically, it uses machine learning to put ads and content in front of the most relevant audiences possible. It's so good at matching ads and content to people that Roku only users are twice as likely to consider ads on Roku than on any of its competitors. That's real money. That's also why Roku is the second biggest holding in ARKW because its content in ad matching machine learning algorithms are largely responsible for its meteoric share price increase from 2019 through 2020 – algorithms that its competitors like Amazon Fire, Apple TV, Sony's PlayStation and Microsoft's Xbox aren't easily able to reproduce.
Just to be clear, Roku owned 60 percent of the programmatic ad market share at the end of 2019. Three times more than the next best device, which was the Amazon Firestick. Roku has an insane lead. The company I want to focus on now, though, is actually Netflix. For two reasons. Most of you watching this will probably have an active Netflix account or at least have many hours of experience with the service. And because Netflix actually uses artificial intelligence at every single step of their service, we can relate some of their use cases back to other companies.
It is well known that Netflix uses recommendation systems for suggesting movies or shows to its customers. Specifically, it uses a combination of two pretty intuitive approaches used to automatically and dynamically recommend content. Let's walk through them quickly because even a high level understanding of them is enough to frame how many other companies in ARKW are using artificial intelligence. The first technique is called collaborative filtering. The idea is to cluster users together in a way that minimizes the differences between people in a given cluster and minimizes the overlap between clusters. Said in English,
the idea is to group very similar people together while trying to make each group as distinct as possible. In the Netflix example, it groups users together by things like viewing habits, ratings, genres you tend to like and dislike, favorite actors, average view duration, the device you use, the times of day you typically watch, and tons of other factors it can infer from the way you interact with the service down to how long you view a trailer before deciding to watch that piece of content or move on. With collaborative filtering,
Netflix looks at the content that people in your cluster enjoy the most, and if you haven't seen it yet, it will recommend that to you. One way to think about collaborative filtering is to imagine that everyone with the most similar taste to yours got together and generated a list of movies. Your recommendations are the movies on that list that you haven't seen yet. The other way is called content filtering, which groups content together by similarity instead of users and offers you content similar to the content that you already enjoy.
Netflix is very, very good at grouping content together in creative ways. If you've been using it a long time, you may remember some of their crazy specific content categories. Critically acclaimed emotional Father Son dramas, cynical comedies featuring strong female leads, dysfunctional online dating stories featuring nerdy Ginger YouTubers. Wait what? Just like in collaborative filtering, Netflix wants to cluster movies so that movies within a cluster are very similar, while movies in different clusters are not similar to each other.
One way to think about content filtering is by just saying if you liked movie X, then you'll probably like movie Y. Liked Gladiator, you'll like 300. Liked Pulp Fiction, you'll like Drive. Liked the ending of Game of Thrones, you'll like Twilight. Combined recommender systems are exactly what they sound like, they somehow combine the results of content based and collaborative recommenders to generate these lists for you. Before we dive into some of Netflix's lesser known but equally interesting machine learning applications.
Let's apply this stuff to the other companies in ARKW. Let's trade out Netflix's content for items you'd buy in a store. That's Amazon. Let's trade out movies for songs or playlists that we think you'd love. That's Spotify. Lets trade out movies for stores you shop at or items you'd purchase or services you'd probably qualify for and people like you already use. That's the ads and sponsored content all over the Internet. I'm definitely oversimplifying here, but if you treat Amazon as the Netflix of stuff and Spotify as the Netflix of audio, your investment research wouldn't be worse for it.
Here's one more example, just to nail the point home. Tesla knows a lot about you, the passengers similar to you, how long the duration of your trip is, the stores you'll pass on your route and the content you enjoy. If you called Tesla the Netflix of point to point travel because it tries to provide the most enjoyable possible experience for you from point A to point B., I'd say you're really starting to see the bigger picture. OK, back to the Netflix of movies, which is Netflix.
Apart from recommendations, there are many other areas in which Netflix is using data science and machine learning. Every movie recommended by Netflix comes with associated artwork. That artwork isn't the same for everyone. It's personalized to you. Each movie gets a whole little portfolio of art created for it. And depending on what Netflix thinks about your tastes and preferences, it will choose the art that it thinks you're most likely to click. Check out these examples for Stranger Things and Pulp Fiction.
Which stranger things thumbnail are you most likely to click? For me, it's the top right one and then maybe the bottom left. If you're a Ghostbusters fan, maybe you'd get served the one in the middle right. What about these Pulp Fiction thumbnails for me is probably the top one because I'm a huge fan of Kill Bill and Netflix knows it because I smashed the thumbs up button on that movie. That means the Netflix you're seeing when you log in looks totally different from mine, especially if you enjoyed the ending of Game of Thrones.
Artwork is a big factor in what movies you decide to watch. Just like YouTube thumbnails, Netflix is trying to maximize views across all of their content. The only difference is that Netflix controls all of the thumbnails and wins no matter what you watch. Square, Facebook, Spotify, Roku, Tencent, Apple, Shopify, Pinterest, Amazon, Alibaba, they're all personalizing their content and ads to maximize the chance that you click. The same store or vendor or artist or studio will create many different images for many different product and demographic combinations, all in an effort to get you to click. This type of problem,
personalizing previews to increase clicks is called conversion rate optimization. Every company cares about it across the board, and every investor presentation worth its salt talks about the work going into optimizing every step of the company's conversion funnel. Automatically adding information to and pulling information from images and videos is an area of artificial intelligence called computer vision. Netflix also uses computer vision techniques to annotate and pull the best frames of each movie out so that graphic artists don't have to go through movies frame by frame.
Netflix might have different algorithms that look for human faces, or unique objects or explosions or rapid changes in brightness or motion blur or whatever. Hmm. Automatic labeling and extraction of information from video frames. Does that sound familiar? That's right. Tesla might have different sets of algorithms for looking for people, signs, lines, debris, sidewalks, other cars and so on in their camera content. Understanding and labeling images is a lot more important for Tesla to get right, because getting it wrong could mean a traffic accident.
There are also way more combinations of things that can occur while driving in the real world than in movies that have continual structure from start to finish. Mostly. Tesla's also need to make these correct decisions in real time as the car continues to move. That's why Teslas need something like Project Dojo and dedicated onboard computers, while Netflix can freely test new ideas and then adjust when the results aren't up to par. Netflix has eight or nine major areas of research that are all worth checking out, and I'll leave a few links to their research blogs and websites in the description below.
There's actually one more thing that's really important to the ARKW discussion. And that's you. Yes, you. Almost every company in ARK Invest's Next Generation Internet fund has clever ways of labeling and segmenting their products and services and putting them in front of the right person once they know enough about that person. But if they're not a massive company like Google or Amazon or Netflix, how can they possibly know enough about you to do this type of targeting with enough confidence?
Well, that's where the whole conversation about you and your data and your privacy comes in. Every social network works hard to extract information from pictures and statuses and text and metadata to understand who you're with, where you are and what you're doing. Your pattern of life affects your buying decisions. Facebook has a service called Lookalike Audiences, where businesses can upload a list of their best customers and Facebook will find more people that behave like them. That essentially means you can have Facebook help you with collaborative filtering by finding new people to put into each customer group that you care about.
Facebook's success isn't really as a social media platform, it's as an advertising data broker. And Facebook isn't the only one with this massive amount of data on you. Amazon is the world's largest database of product attributes and information to connect users to similar products, even if those products come from different stores. This means Amazon's data is a great data set to help other businesses with content labeling and organization, as well as understanding what users map best to each product, label and category.
That's essentially the content filtering side of the coin. Google Chrome, Google Search, Android OS and YouTube all supply Google with different bits and pieces of your behavior and interests across multiple devices and so on. Strange, isn't it, how the biggest companies in the world are the same ones that somehow provide their digital products to you for free? Here's the most important lesson you can learn when it comes to the next generation of the Internet. If you're not paying, you are the product.
So join my Discord community. You don't have to pay for it. This is Ticker Symbol: You, my name is Alex, reminding you that the best investment you can make is in you.
If you want to comment on this, please do so on the YouTube Video Here