In the 21st century, data has evolved into a digital currency. To protect their customers' privacy, telecommunication companies are implementing more storage and accessibility safeguards, with the industry expected to invest USD 36.7 billion 1per annum in artificial intelligence (AI) software, hardware, and services by 2025. But choosing the most effective solution is not an easy call. After all, building AI models and protecting your users’ data is generally difficult, costly and complex. However, a team of intrapreneurs at Ericsson ONE may have come up with a breakthrough solution: Snapcode.AI. Discover more here.
In one sentence, what is this new solution?
Snapcode.AI is a novel and disruptive solution that uses a blockchain trusted data layer to stop the transport and storage of data.
Why is this solution needed?
As 5G services mature with new infrastructure and platforms emerging, enterprises need to rethink their application architecture with a completely different strategy. They need a portable, secure with a trusted data layer, database or platform agnostic, that’s able to run on-premises or in the cloud. However, the development of a single AI model can cost between USD 20K to 1 million – until now.
Developing an AI model with Snapcode.AI will significantly reduce time and cost – an estimated 50 percent savings on AI spending – for communications service providers (CSPs). The solution simplifies AI data modeling by performing data modelling without accessing raw data and uses blockchain technology to secure it. This means no data is stored and instead stays onsite at the customer’s premises. It also addresses common AI/ML development, propriety data and data storage challenges CSPs and enterprises face. Discover how it addresses these challenges, below.
The challenges for data driven operations
Building AI models is difficult, costly, and complex.
- Talent is expensive
- Enterprises need AI talent today to be data driven
Customers are afraid to share data.
- Lack of security
- Lack of transparency
- Lack of auditability
- Privacy regulations
If data is shared, we are faced with storage issues:
- Dependency on cloud providers
- Licensing costs
- Security and protection of data
What challenges are you trying to solve?
There are many challenges for data-driven operations from both the telecom and customer perspectives. The three main ones are related to AI/ML development, proprietary data, and data storage:
- AI/ML development: While data scientists are often responsible for the protection of data, it is obvious that not all companies can hire such specialized personnel. We want to make AI accessible for the whole telecom industry, which is why we were looking to develop a solution with a low code aspect that does not require the sharing of personal data.
- Proprietary data: We wanted to design a solution to give customers confidence that their personal data is not being mined – it is not leaving the servers. This also ties into regulations, enterprises, especially telecoms, must adhere to regulations on customer data use. These regulations focus on biometric data, geo-location data, call identification data, cell site data, and other data which can identify the user equipment (UE). Due to these concerns, spectrum providers are hesitant to share data with hardware telecom manufacturers or their internal teams. As a result, even if an AI/ML model is written in a programming language, there is no assurance that it will provide any results, due to a lack of data needed to train this model. Without this data, models often lie dormant, become stale or show inaccurate results.
- Data storage: Even if a customer does not mind sharing certain data with a trusted partner, it is expensive and creates a dependency on hyperscalers.This dependency is from the Hyperscale's increased market share of cloud hosting data, and the need for data backup/redundancy.
How will your solution solve these challenges?
Our solution has three strong principles that have shaped its design:
- Minimizing the footprint of any solution we have deployed on the customer’s premise: Our solution has a very small footprint agent in charge of data inventorying and job execution running on our customers' infrastructure. Our long-term goal is that eventually our agent should not be an Ericsson intellectual property (IP), but instead be open implementation.
- Reusability: This applies to the low-code aspect of the solution and not having access to raw data. The first thing that data scientists tell you is that they cannot do their job with raw data. We are creating a model that we can develop and train to enable the upscaling of our AI portfolio to multiple customers, including those who do not wish to share data. We are prioritizing enabling reuse across all our opportunities.
- Tracing and logging: Our solution enables fine-grained logging. We log everything. For example, every time a data scientist has access to the information, or there is a manipulation of potentially sensitive data, we also log the purpose. If anything happens to the data, there is a very detailed audit trail.
Snapcode.AI achieves these principles through the following features: a trusted data layer and a low-code/no-code (LCNC). These are important because:
- A low-code solution: Regardless of their current AI/ML cloud goals, businesses need to understand and plan for a distributed architecture. They must focus on designing future-proof solutions to keep up with the fast-paced customer needs. Enterprise-class Business Support Systems (BSS) based on LCNC (Low-Code/No-Code) protects organizations from AI/ML platform architecture issues. Previously, there were only two options for converting core systems: buy the latest commercial off-the-shelf (COTS) software and modify it, or upgrade and rebuild legacy systems in-house using traditional coding. Both methods are expensive, time-consuming, and require skilled people, and they are usually built-in technology silos, raising the risk of inconsistency with business needs. A modern method for core system change is known as low-code/no-code. LCNC refers to any activity or stage in system that eliminates the need to manually write or rewrite a piece of code, reducing software development time. An LCNC platform allows us to construct business processes and logic in enterprise beans, generate reusable code, and design user interfaces. Enterprises can use low-code platforms to further upskill their workforce, beat traditional development times, and get results faster. Low-code development platforms allow enterprises to outsmart their competition by quickly modifying how they provide services to customers via their preferred channels, such as mobile apps and self-service portals. As a result, they provide their employees with the tools they need to do their jobs more efficiently and effectively, resulting in a solution where everyone benefits.
- CSPs can rest assured the data is protected: One way to protect data is to establish a trusted data layer where data providers have a secure audit trail, with transparency and adherence to privacy regulations. After all, CSPs need to trust the AI/ML model developers with their proprietary data or federated and secure data repositories. Our solution authenticates each user's data usage by looking at a record of all data logs from the past, where each transaction is run by a large, distributed, and uninterested network of computers (nodes), which means that users cannot renege on agreements or hide past data-logs for fraudulent purposes, and they can always expect the network to be up and running. So, blockchain's decentralized system has two big advantages:
- Establishing trust through immutability: The private, permission-based shared ledger makes it impossible for anyone to change historical records. Any change to a record from the past will affect how all future data logs are logged in the blockchain (that is, the hashes of the AI/ML model inference). This makes the change very noticeable. It is also impossible to change all copies of the blockchain because the system is decentralized, so it cannot be done all at once.
- Security through distribution: The blockchain is durable because it does not rely on one computer, but on a network of computers that will be monitored and managed by an alliance. Part of the revolution of the blockchain is how it combines distributed ledgers with other technologies that make them more secure and private.
- Using a hash value: When a lot of data is put together, it can be compressed into a much smaller unique numerical code called a hash value. This is a type of technology called a hash function. If any part of the data inside a hash code changes, so will the hash code itself. These codes can be made with asymmetric cryptography, which means only certain private keys can break them. In other words, hash codes can help you find any changes in data and keep the information, like the details of a transaction over a blockchain, private. New blocks in a blockchain have a hash value that is linked to all the other blocks in the chain. This value is called a "hash."
Who is the solution for?
Snapcode.AI is particularly well-suited to business-to-business (B2B) vendors, like Ericsson. This is because business-to-business-to-consumer vendors (B2B2C), such as streaming giants, own their own data and can use (in accordance with GDPR and other legal frameworks) and monetize it. In contrast, companies who are selling software and products in a B2B context have been, in a sense, taken by surprise. This is because telecommunication companies such as Ericsson used to sell software, but now that data and business success is more correlated, we have to assure our customers that, even if we sell the software to them, the data is theirs. We will now be able to sell data services or ML services to our customers without them having to ship data to us or even provide access to the data.
What are the main benefits of developing the project within Ericsson ONE?
Ericsson ONE has given us the high-level attention we need: support and funding to make our idea tangible. After all, in all organizations, managers are busy working on the company’s defined roadmap. Even if they find your idea disruptive and valuable – like they did with our solution – it is still difficult to get buy-in due to time constraints and budget restrictions. This is one of the reasons why startups are so disruptive because, unlike enterprises, they do not look at a roadmap but instead look at the market.
Everyone in Ericsson ONE is working extremely hard to create a process for success – this is a formula that has not been cracked by anyone in Silicon Valley. So, whether you want to be an intrapreneur or an entrepreneur, I would say that from a funding and access perspective, Ericsson ONE provides the best opportunity to realize this dream.
What is the current status of the project and what are the next steps?
From a technology point of view, we have been running this like a startup. We are very lean and delivered what we need for the next round of Ericsson ONE funding gate—Pioneer’s Nest.
We already have a number of high-profile customers, including NEC America who will be imbedding SnapCode into their product line, with a commercial launch expected for Q3/Q4, 2023. Some other leading CSPs are already reaching out to us and want to work out how we can collaborate – the future looks bright!
Reflections on the project
The best way to create an innovative solution is first to understand what the problem is. That is when you can come up with, as they say in Silicon Valley, a ‘billion-dollar solution’! Because innovation really is a solution to a challenge – after all, if there isn't a problem, why fix it?
Rani is presently at Ericsson’s Global AI Accelerator in Systems Management. Among Rani’s awards is being read into the United States House of Representatives Congressional Record in 2022 and listed as one of the 10 Most Influential Women in Technology by Analytics Insights in 2021.
Rani has over 20 years of executive management experience in both public and private firms. Rani is a strategic thinker with deep technical expertise in AI analytics, cybersecurity, entrepreneurship, business transformation, and enterprise software. She has filed over 15 key industry patents, is a published author and a frequent guest lecturer at University of California Berkeley, Merced, and most recently Adjunct Professor at Southwest Law School.
Julien Forgeat is a principal artificial intelligence researcher at Ericsson Research. He joined Ericsson in 2010 after spending several years working on network analysis and optimization. He holds an M.Eng. in computer science from the National Institute of Applied Sciences in Lyon, France.
At Ericsson, Julien worked on mobile learning, Internet of Things and big data analytics before specializing in machine learning and AI infrastructure. His current research focuses on the software components required to run AI and machine learning workloads on distributed infrastructures as well as the algorithmic approaches that are best suited for complex distributed and decentralized use-cases.
Shohreh is currently a director in program management focusing on AT&T OSS projects at Ericsson. Shohreh has 24 years’ experience in various roles and across all Ericsson regions. Author of multiple patents, Shohreh went from sales in the Middle East and Eastern Europe, to research and development as product manager in core networks, before spearheading the IMS program in Market Area North Africa. She then moved on to hold account management responsibilities establishing strong credibility with C-levels leading to important break-in deals. From R&D to sales and with a demonstrated ability to build trusted relationships Shohreh has accumulated a wealth of business development experience across different markets.