To fully realize the promise of IoT and the wealth of data it can provide, it’s essential to connect it to artificial intelligence, aka. “a brain.” There are, however, several challenges associated with both running AI at the edge of networks and sharing databases securely.
Combining AI and IoT enables “making a series of valuable predictions about devices that are embedded within the real world,” said Christian Catalini, assistant professor of technological innovation, entrepreneurship and strategic management, and founder of the Cryptoeconomics Lab at Massachusetts Institute of Technology. “This allows you to change how you respond to changes in the environment and to new information coming in.”
But you can only make valuable predictions if you have access to good data. “Blockchain can help create marketplaces for data and their exchange, and also improve data privacy,” Catalini added. Once data privacy improves, more databases are likely to be shared.
Challenges running AI at the ‘edge’ of networks
Running AI at the edge faces “multiple challenges,” said Martin Casado, a general partner at Andreessen Horowitz. “First, you need a lot of data, and this is something edge models aren’t very good for. Normally with AI, a whole bunch of data gets compiled within a data center, you train models and then push them to the edge. While this makes sense for some use cases, it won’t work in environments where you need to operate while learning because there’s a storage requirement.”
Another challenge, Casado added, is that “AI is computationally intensive and the current sets of processors just aren’t built for AI workloads. Graphics processors are being modified and hybrids are coming up to make it better, but we need an architectural shift to move central control to the edge.
“That said, the most compelling use cases for AI are at the edge because it’s primarily good for image recognition and geospatial processing — which is what things at the ‘edge’ need to do. So, the need for AI is at the edge, but there are a number of technical challenges that still need to be overcome to make it work.”
Moving compute to the edge for machine learning analytics and AI requires a lot of power, and “without the right compute support, it won’t work,” concurred Andreas Kind, manager of industry platforms and blockchain for IBM Research in Zurich, Switzerland. “There are different possibilities for compute support, such as neuromorphic computing and in-memory computing, as IBM Research is doing, which don’t follow the standard ‘von Neumann’ architecture you’d normally use to build a computer. They function more like how our brains work — with a very low-power chip that can process unstructured data like images or video.”
But you can also opt for a more “distributed” approach to compute systems at the edge. To do this, “you can use a group of nodes to work on a particular task,” Kind said. “This, of course, becomes a security issue because you need to connect different computers or mobile devices that are owned by different people — while ensuring no new security holes or attacks would be possible.”
Blockchain-based platforms for the edge
Andreas Kindmanager of industry platforms and blockchain, IBM Research
Why use blockchain-based platforms to connect AI and IoT at the edge of networks? For starters, “blockchain systems bring together data in a way that wasn’t previously possible,” Kind said. “AI analytics and cognitive computing are all about having access to data — without it, you can’t apply algorithms and do prediction as well.”
Business networks tend to be made up of multiple players who are often reluctant to share their data with each other “because they don’t fully trust each other, so we end up with ‘silos’ of companies that do analytics within their organization, but not necessarily in this combined set of data,” Kind noted.
As a technology, blockchain opens up a shared distributed ledger, which is a very specific database that features strict confidentiality, access and consistency requirements. Since these requirements are enforced, “there’s more openness,” Kind said. “Parties see an advantage of opening up toward a shared distributed ledger — providing one truth between multiple companies — because there’s more protection in terms of cryptography and protocols. It allows those of us on these networks to see databases of data that have never been brought together before, which creates opportunities for new forms of analytics and AI.”
Unlike blockchain’s Bitcoin or Ethereum context, which tends to be about bypassing established finance players to anonymously pay for services, IBM Research is working on an approach for businesses that uses a permission blockchain. Permission in this case means that you need to go through a registration process before being allowed entry.
“All participants need to prove their true identity as a company, employee or individual before receiving a certificate from which they can transact anonymously,” Kind said. “But there’s always the possibility to ‘turn the cards’ to look behind the anonymous transactions. There are requirements for privacy and confidentiality, but also for auditability. In industries such as finance, food, and airlines, regulatory authorities may need to know who was behind a certain transaction.”
And it’s crucial that all of this must be able to scale so that you can conduct high-throughput transactions and know that it went through fast — within a second.
Connecting AI and IoT
Connecting AI and IoT involves trust issues because the internet of things relies on sensors and infrastructure that exist far beyond the realm of data centers — reaching into homes and vehicles, as well as public and remote spaces.
The model evolving around AI and IoT is “akin to a human being in that the IoT devices are the sensors like eyeballs, hands and ears located away from the brain, which does the processing in a data center somewhere far away,” Casado said.
Why keep the processing within a data center? Because “it has the storage and computational power you require,” Casado explained. “But then you need to connect the ‘eyes, hands and ears’ to the ‘brain’ within a data center in a secure way. There are many ways to securely connect these devices to the brain. Blockchain is just one of many — its primary value is providing a particularly paranoid security model for connecting devices to AI. Many IoT use cases don’t require that level of trust, but blockchain allows you to prevent a device from becoming a bad actor.”
If, for example, your IoT system involves a pole-mounted camera within an agricultural context somewhere out in a remote field, “it won’t necessarily have a protective boundary,” Kind pointed out, whereas “data centers have physical security with cybersecurity — firewalls and intrusion detection systems — so they’re somewhat protected. The distributed nature of IoT makes it much more difficult to protect.”
Some future applications will demand more security. “Cars are becoming more instrumented, outfitted with hundreds or thousands of electronic control units (ECUs), and will increasingly become ‘in charge’ of the entire safety of the car,” Kind said. “This, in turn, relates to software updates on ECUs, because the firmware must be updated for its communication and transactions.”
As cars become much more autonomous, and particularly for consumers in car-sharing situations, “you won’t expect to need to pay for tolls or parking because it will become a mobility service,” Kind added. “Cars will need to handle payments and transactions.”
If a car is idling at a traffic light, for example, it may need to negotiate with nearby companies that have solar panels on rooftops to figure out which of these companies can provide the cheapest energy within the three seconds of idling. “So, there’s an increasing need for trusted transactions at the edge,” Kind said. “Blockchain provides exactly the sort of fabric required for these trusted transactions.”
Blockchain-based platforms available now
Blockchain-based computing platforms do exist already — most notably from iExec and Golem. But there’s also an open source Linux Foundation platform called Hyperledger, which IBM initiated with the open source community, as well as the Hyperledger Fabric platform beneath its umbrella.
“Our work in blockchain uses the Hyperledger platform, which is an open source stack hosted in IBM’s data center and cloud infrastructure,” Kind explained. “This platform allows you to very easily connect the data you get from IoT via specific adapters to integrate a variety of existing sensors and protocols. Then, you can integrate and connect the transactions that are related to these sensors with blockchain systems that might belong to different organizations — ones that maybe aren’t willing to make available all of the data coming from these sensors by default. A third step is using the cognitive AI components to infer new insights from this combined data.”
Hyperledger is available now; “in terms of basic building blocks, it’s already quite advanced,” Kind said. “But we still have research and engineering to do for particular industry domains to determine how to best combine blockchain, cognitive AI and IoT.”
A role for software-defined networking?
For those of you wondering about the “edge” and software-defined networking (SDN) in this context, it can indeed be used to connect IoT devices to a central AI brain. The concepts of SDN are already being applied to decouple control from the data plane within the blockchain realm.
“We had a very similar issue with SDN’s ‘brain’ or, in this case, AI as the controller that was remote and controlled the network elements,” explained Casado, one of the visionaries behind SDN. “Imagine that the network elements are part of a blockchain and you can just update the blockchain to change the forwarding elements.”
They chose, instead, to believe that the end devices are trusted and created secure connections between the devices and the “brain” or the central control. “What you can say is, ‘I trust this device; the device has a secure connection to the AI using standard VPN technology, and that’s sufficient for many environments,'” he said. “Maybe not the most paranoid environments, but it’s OK just to have an encrypted point-to-point link between the IoT device and the AI/brain.”
Because SDN is the networking layer, to connect IoT devices to a data center for AI, you’ll need to create a network to do it. “Say you’ve got a bunch of drones that go out and do things like roof assessments for commercial insurance. This information gets streamed back to a data center where the AI is processed,” Casado said. “The network between the data center and the drone can be provided by SDN. It can be more flexible, and to make it even more secure, you can use blockchain to implement that SDN system. I don’t know if anyone’s doing this today, but you could actually build a more sophisticated way of connecting IoT devices with the AI back end using SDN and blockchain.”
Still early days for blockchain-based platforms
It’s important to keep in mind that the technology surrounding blockchain is still in its early stages, Catalini pointed out. “There are many experiments underway, from IoT to AI, to new types of digital platforms,” he added. “It will take time for these solutions to develop and to reach consumers and businesses.”