How Agents + Decentralized Interfaces Help The De-Siloazation of IoT

From IIW
Jump to: navigation, search

Agents and Decentralized Interfaces Help the De-Siloization of IoT


Thursday 3F

Convener: Sam Curren, Evernym

Notes-taker(s): Scott Mace


Tags for the session - technology discussed/ideas considered:


Discussion notes, key understandings, outstanding questions, observations, and, if appropriate to this discussion: action items, next steps:


Sam: Silos for light switches and Alexa is the problem that I have. IoT is assigned to silos. There’s not a good alternate way to deploy a bunch of IoT devices managed by users that makes this happen. If the dance is complicated, they’ll take it back to Costco.


Q: It’s a sad rainy day slide.


Sam: This is the problem that sucks. DIDs resolve to service pointers. Which points to some thing that I control. I’ve been talking about extensible APIs. Vlad calls it decentralized interfaces. One thing these silos have is an API. The device knows how to talk to the API. That’s one of the reasons for silos actually existing. Has to connect to a predictable API. Let’s say I have a big thing capable of running APIs. I could install an API that happens to be a light API. If I have a thing, maybe cloud, maybe in my house, and can install an API interface according to some standard, now we’re removing one of the barriers to connecting this API to something I have control of. The API has some behavior. Kim Lane had a blog post and said where’s the WordPress of APIs. So many have similar functionality.


George Fletcher: Are you thinking standardization of these APIs across verticals?


It’s going to be messy. I’m unlikely to have one API for a variety of light switch devices. The people that created them did not agree for some reason. Maybe a competitor who didn’t want to share.


Q: Or one manufacturer thought only white light, the second had colors.


Sam: A bunch of the shared logic is going to be the same. The same WP-like plug-in could support two interfaces with the same basic logic. In doing so, we attack the problem with a couple of options. We want to make our stuff work. So various devices can plug in. Now it’s messy. Sometimes there’s movement from an older API to newer APIs. That’s mostly OK.


Q: Amazon wants to know the stuff you’re doing.


Sam: The larger question is the value proposition for device manufacturers. Let’s pin that for a second. The voice stuff is still going to go to the server of the richest man in the world. But instead of going to another silo to issue that command, Alexa messages my general API which then responds instead. I can avoid Alexa having to be tied in the cloud directly to all these things. I can have logic on how to turn on and off various lights.


Q: I spent some time on a contract doing smart lighting. There is no margin for devices at all. The only thing you can do is attempt to monetize data on the devices. Practically everybody says I’m not paying for the devices. How can you be profitable?


Sam: Alibaba will get there. For example, can buy non-UL certified WiFi switch. They’re cheap, $8-10, less than 1/3 of U.S.-packaged tech. Not as easy as setting up Alexa, but that will come along. A lot of money in selling silos, no money in selling hardware. A super good question. The trends i’m seeing, we’ve got GDPR going on. I can see government regulation in Europe mandating certain behaviors for IoT devices. Can be more extensive laws. The value they have in running a silo is making use of the data. You still need services to make use of data. I.E. power monitor. Produces power use signature over time. Services apply machine learning and identify individual devices. Still value in the silo. We can with consent provide a data feed, an opportunity to perform same service but not run afoul of the regulation. We get to pay for more valuable services. $7 device from Alibaba.


Q: Carl Ewett has talked about the idea of personal islets.


Sam: I can pay more for a silo, might get one for free, sell your data.


Q: We got really hot last year in the summer. Bought AC. Takes a lot of power. PG&E would like to charge us $12K to upgrade. Bought open-source tech for energy monitoring, 14 sensors, in main breaker box. As soon as energy consumption goes over a certain threshold, turns off certain devices, like the pumps for my pool. I cannot buy a commercial system that does this for me. Can’t depend on the cloud, because as soon as my ISP goes down, circuits would go out. Raspberry Pi runs my pool pumps. WiFi is unreliable. I would like to have exactly this. Raspberry Pi registers with central controller, does some key exchanges. Based on these measurements of power consumption, what are the decisions.


Sam: You’ve got logic in here. One of the things we don’t have yet because we’re arguing about credentials exchange, the intelligence can happen in two ways. Companies like Evernym will support agents like this, power management. We can install behavior that actually does that. Now you can do power management or whatever feature. But Evernym isn’t the only model. Open source projects can evolve it like WordPress. If you happen to be running one of those, you will have a way that write an extension that fits nicely into this framework, publish it for review. Code can be put on a GitHub repository. That’s rougher than I’d like for my mom’s sake. I can trivially update my own fork of it. Not advocating an exact follow of that. End up with a WordPress-like agent that allows us to pick. That may be forked or improved by someone else.


Q: Why the device manufacturers would buy into the model.


Sam: We talked about regulation, that might happen.


Q: If I don’t have a silo at all I don’t have to build it. But if there’s no margin in the thing I’m building, I don’t care about security or other stuff that should be in that thing.


Sam: The people have realized the money is in running the silos. You can buy ZigBee compatible wireless devices that already exist. WinkHub, SmartThings hub, HomeSeer hub. I can buy a $35 light switch from a company that does not produce a hub, I can install it and it works. The pattern is already starting to happen. Consent to monitor or evaluate data will flow into other things that make that happen.


Alan: Carl has talked about the company still owns the data, they have incentive, gives you better control.


Wendell: If you don’t actually hold the data, that’s a way out of GDPR. If you leave it on the consumer’s device.


Alan: Can be encrypted with device owner’s key. The company doesn’t have liability.


Sam: With that as a layer, makes it a little easier to do. Personal data has changed because we have some cool technology answers.


Wendell: Have you uttered the words Pico or Digital Twin?


Sam: These are features that can provide these types of things. Phil knows that picos can function agents in this sense. If we can get the manufacturers to trust users installing this, we will see siloless tech being offered.


Q: What is the user buying that has an agent on it?


Sam: Raspberry Pi, or in the cloud. One opportunity is to have an address for your agent. Can be made available on your home network. Could be a home automation hub or WiFi router.


Alan: I just bought your smart house, now what?


Sam: Yeah, the transfer of smart home devices. And have assurance it’s not all going to the other guy.


Phil: The connected car product called Fuse had this feature you could transfer the feature associated with your pico to another owner. Delete some personal data like your trips but not other data.


Sam: Air BnB data. One thing we haven’t talked deeply about is we have this agents, but they can represent units of collection – a house independent of the owner, or an HOA. Various levels of things you can stick those for various effect.


Wendell: Houses are 100 years old. How are you going to develop standards, specs and protocols without getting bogged down in doing backward compatibility.


Sam: The interesting piece is the definition of the interfaces. I don’t yet have perfect answers for. You define a chunk of interaction in a way you can actually support. So if you define a little chunk of API, it’s not going to be a lot of it, and that allows you to then insert other parallel pieces of API next to it in ways that don’t conflict. Competing APIs? May support three. Then there’s a discovery process. Do I happen to speak anything you speak? Then pick the preferred.


Q: A lot of IoT information management happens to be time series data.


Wendell: This framing assumes common networking, TCP/IP, REST, microservices. But back net and analog signaling is cheaper and just doesn’t break. How will this scale across a wide range of technologies? We might not be using CAT 5 in 25 years but we might still be using analog phone wires.


Sam: If I have a temperature sensor running off a tiny wireless module that isn’t IP anything, the signal gets upgraded along the way.


Wendell: There’s local nets with particular interesting properties. At some point you upgrade those into a protocol you do understand for a central agent. Is that the point?


Sam: Yes.


Wendell: So you need a gateway.


Sam: Yes.


Q: Which protocol?


Sam: Six months ago, RESTful, RESTish. Moving toward messaging.


Q: Uber has complicated schemes by which they are passing messages from the device into Kafka.


Sam: Things like MQTT, I see being wrapped, passed into the ecosystem. This is fairly forward thinking. The interesting part to me is figuring out how the interfaces will actually run. You can define messages in terms of Swagger arguments.


Wendell: Flexible because it’s text-based. Making function calls. You don’t tend to get dynamic updates. Swagger systems don’t tend to call you back.


Q: What are the actuators in this?


Sam: I’m assuming it can call Web hooks. The more interesting case is signaling. The gateways will perform this functionality. I’m waiting for the IoT device hack where they can reconfigure which lights are connected to which light switches. Useful things Alexa could do. Ask questions of your agent. Tell me what your temperature is without having to resort to a weather service.


Q: Greengrass can talk to Alexa.


Sam: If you’re a nerd. Not generalizable enough for my liking. It’s $15 less for a siloless light switch. I’m sure they can go low, they just realized they can get $35 for the siloed. Prices from Amazon.


Alan: That argues against your point they are selling the hardware to get your data.


Q: IoT meetup around here, getting people to spill the beans how IoT companies are monetizing. There are different levels of maturity. Trying to insert themselves into your life on an ongoing basis, just like Twitter or FB, with an app on your mobile phone. That’s the ultimate model.


Alan: Why isn’t the device that isn’t collecting data less expensive.


Phil: Amazon is famous for loss leaders for other reasons.


Sam: But the light switch is a good example. ZigBee has become the well-defined interface. They undercut the other guys. All home automation folks, they hate selling light switches. They can’t charge a lot. Want to automate audio and video. I suspect that’s why it’s happened with light switches. I put a flow meter on cold water intake to my hot water heater. I can approximate the reheat rate, I can ask Alexa how long can I shower. Frustrated I can’t make more use of this. Right now innovation is being stymied because it’s hard to establish these patterns. This will help lower the cost of innovation.


Q: Pretty much impossible for a silo of any kind to solve all the use cases for the house. The house industry solves this problem with the Home Depot. The IoT, if ever real in the house, will look more like the Home Depot than like the iPhone.


Sam: A lot of home automation stuff is easy to script.


Q: There’s a base infrastructure that needs to occur: Discovery, security, a place to store time series data over the long term.


Sam: I’m working on how to manage the security of the communication between the device and the agent.


George Fletcher: You need introduction. I want to provision it with credentials and tell it where it can interact with the agent as a whole.


Sam: Interface discovery as part of the introduction. Yo agent, how can I talk to you.


Wendell: One thing you will need is a system to go analyze all those rules over the years, figure out if your house can do something unsafe. You can get pathological, unstable, risk point behaviors.


Sam: This is the experience of every nerd’s wife. My wife wants the lamp to fade up when alarm goes off.


Phil: We walk into bedroom and want lights to come on, unless someone’s in the bed. Then it’s the dog.


Sam: What happens when it fails? If it fails badly, it’s not a feature as well. I want a cookie jar that doesn’t open itself until I weigh myself that day.

IIW26 Th 3F HowAgentsHelp.jpg