Discover more from PATENT DROP
PATENT DROP: Amazon will teach you AI
Plus: Oracle’s AI-generating chatbot, Ford listens in
Happy Thursday and welcome to Patent Drop!
Today, we’re checking out a filing from Amazon detailing its plans for an AI learning center; Oracle’s tech for an AI chatbot to build other AI models; and two patents from Ford that show its intent to listen to your road rage.
Let's go for it.
Amazon’s AI tutor
Amazon may want to demystify AI for the tech novice.
The company is seeking to patent what it calls a “hands-on artificial intelligence education service.” Amazon’s platform is essentially an AI sandbox for beginners, aiming to help people with little to no prior knowledge in machine learning and AI to “familiarize themselves with and experiment with various aspects of complex machine learning techniques including generative artificial intelligence.”
Here’s how it works: The service includes a number of user interfaces and modules breaking down different AI concepts that guide users toward creating their own customized AI models. Amazon said this would be implemented as part of a “suite of services of a provider network,” likely referring to its AWS division. This would also include a machine learning service that could execute and run the models.
The education service would focus on sub-fields of AI that are “exceptionally complex,” Amazon noted, specifically citing generative AI. Amazon said that the “creative potential” of generative AI makes it an “especially attractive entry point to introduce users to machine learning concepts.”
Amazon said this service could help break down barriers to entry for AI newcomers in a number of contexts, including health scientists, writers, graphic designers and musicians. Using the platform could allow the layperson to “become conversant with artificial intelligence methodologies, and even initiate the training of customized models which can then be used to produce real-world results,” the company noted.
Amazon has been public about its AI push in recent months. The company vowed to invest $100 million in what it calls the AWS Generative AI Innovation Center, and news broke on Tuesday that the company has put together a team to work on its “most ambitious” large language models, led by Alexa head scientist Rohit Prasad, according to AI Business. The company has also sought several AI-related patents, including a system for rapid model development.
Amazon still has some catching up to do compared to frontrunners like Microsoft, Google and Open AI. However, Amazon is specifically treating AI as a support technology for the company’s business priorities, said Irusha Peiris, retail analyst and portfolio manager for William O’Neil.
“Amazon seems to be doing this kind of in tandem with what they were doing with AWS,” said Peiris. “They’re saying, ‘We'll make it very simple for you just to take advantage of all the resources that we have, and we'll keep developing tools to make it easier and easier for you to do whatever you want.’”
Given that AWS is one of the company’s major golden tickets, raking in $21.4 billion from the segment in the first quarter (16% of the company’s total revenue for the period), it makes sense that Amazon would pair its AI work with its cloud offerings. AWS already offers Amazon Bedrock, a service for startups to build and scale generative AI apps.
Peiris doesn’t expect Amazon’s AI lag to last long. “Over the next year or so we'll probably start seeing Amazon make more noise,” he noted. “They already have all these business relationships with most of the larger companies, so once they come up with something that's a really good value-add, a lot of those companies are probably going to start taking advantage.”
Oracle’s BYOD (Bring Your Own Data)
As long as you bring the data, Oracle wants to make machine learning development as easy as typing in a chatbot.
The company seeks to patent a system that uses an intelligent assistant to “enable a user to generate a machine learning system.” Oracle’s system allows a user to automatically create an AI-based system using a chatbot that translates natural language commands into a “structural representation of a machine learning solution.”
“In this way, a user can work with artificial intelligence without being a data scientist,” Oracle noted in its filing.
Think of it like Build-A-Bear for machine learning models: In chatting with the user, the system ascertains where the user’s data is stored, what problem they’re hoping to solve, and what kind of predictions or performance they want. Oracle noted that this interaction can be “aural, textual, or through a graphical user interface.”
The system then generates a prototype AI model by picking out the proper building blocks from a “library of machine learning applications” and locating the necessary data from a user’s database, as well as recommending how a user can best deploy that model.
Oracle’s system also can monitor the output of the new models after the fact, allowing for feedback and adjustments. The model can be “trained, tested, and compiled for export as stand-alone executable code.” Plus, for those repeat users, the platform can create user profiles and make recommendations based on historical user preferences, as well as use previously generated models to build news ones.
Oracle isn’t the only company looking to helping people understand their data better with little to no data science experience. JPMorgan Chase is seeking to patent a similar invention for “no-code machine learning” that creates an executable custom code based on a user’s specification and data. Even Amazon’s system discussed above offers similar guidance to its non-techy clients. The only difference is that Oracle is using a chatbot to do so.
The filing also signals that major companies are starting to catch on to the low-code and no-code development movement, or the trend toward developing apps, products and tools with as little coding as possible. In December, a Gartner study found that the low-code and no-code development sector could grow by 20% this year, with the uptick attributable to the growing demand for customizable automated workflows and analytics.
And with the rise of machine learning as a way to gain insight from data stockpiles, companies may look to low-code and no-code development as a way to take advantage of their data without needing to hire a ton of AI experts.
Data is at the heart of Oracle’s business. The database management and cloud services company already works with a number of Fortune 500 companies, including Siemens, Cisco and Zoom. But Oracle has stiff competition in the data services market from Amazon, Google and Microsoft.
Oracle already offers some AI infrastructure integrated into its cloud services, and has sought to patent an AI-enabled system for cleaning up large databases. But with its competitors stepping up their AI game in more ways than one, it makes sense that Oracle is following suit and continuing to build.
Ford hands you the mic
What you scream at that truck that just cut you off during rush hour should stay between you and your car.
Ford is seeking to patent a system for “anonymizing speech data” that’s collected by a voice recognition system in a vehicle. This system removes “speaker-identifying characteristics” from speech data collected from in-car voice commands. It then uses machine learning to generate a “random vector,” or randomized data in place of the previously identifiable characteristics, to apply to the speech data.
Essentially, the manufacturer’s system obfuscates personally identifiable information that may be used when giving commands, replacing it with a random, unidentifiable sampling of speech data. Ford noted that the data becomes anonymous while still preserving non-identifiable characteristics, such as content, tempo, volume, pitch and accent.
While Ford may be keeping your road rage anonymous, the company may also want to use your voice to sell you stuff. The company filed a patent application for a system for “providing targeted content to users.”
The method uses location data, audio data and other sensor data to recommend a “product, service or entity” based on a user’s intent, such as when they speak a command to an in-car digital assistant.
Think of it like search engine ads for your car: When you search for or ask something, targeted ads generally float to the very top if they’re related to the query. Ford is doing the same thing, just surfacing those targeted ads in its vehicles. For example, if you ask your car’s digital assistant to take you to a gas station, it may pull up options to direct you to gas station chains that have paid to rank higher.
Ford’s patent activity has shown that it wants to pack its vehicles with features. The company files to patent tons of new and inventive features practically every week, seeking to lock down everything from biometric eye scanners to EV charging reservation system to a vehicle curfew enforcement feature. Adding systems that both keep your personal data safe and use that data to give you recommendations as you drive just adds to its ever-growing pile.
But as Ford continues growing its EV footprint — with plans to make half of the vehicles it sells electric by 2030 — additional features that rely on close user tracking could be part of its plan to compete with the EV whale that is Tesla. The industry leader’s vehicles are commonly known to be packed with features that you wouldn’t think to add into a car, and often use internal sensors for rider tracking purposes. (Check out our recent edition on Tesla’s patent for physical user tracking for a deep dive into that can of worms.)
The second patent in particular also opens up a new potential revenue stream for Ford: digital advertising within its GPS system.
Advertisement within mapping can be a lucrative bet. For example, while Google doesn’t break out how much its Google Maps ads make specifically, the service has hundreds of millions of monthly active users. Apple has also reportedly considered adding advertisements into its own mapping app last year. By adding in digital ads into its own proprietary systems, Ford could stand to make some extra money off the user data it’s collecting.
Some other fun patents we wanted to share.
Adobe wants you to get your best angles. The company is seeking to patent a system for generating enhanced digital images from a “burst of digital images.”
Google wants to know when you’re on the go. The company is seeking to patent a system for determining what physical activity you’re performing using sensor data, and if you are “consuming media” such as music or video while you do so.
Meta wants to fill out your virtual world. The company wants to patent a system using “scalable parallax” to render avatars, objects and environments in the distance.
What else is new?
Waymo is self-driving its way to Austin, Texas. The Alphabet-owned autonomous vehicle company will start offering rides there this fall, making it the fourth major U.S. city the company has expanded to.
Alibaba opened Tongyi Qianwen, its large language model, to developers, potentially pitting the Chinese tech giant against Meta’s Llama 2.
Agence France-Presse, the French international news agency, is suing X, formerly Twitter, in a copyright case to potentially secure payment for its news content.