Apply HN: Grid – a Decentralized AI Assistant
12 comments
Building an AI Assistant is a very hard problem, in and of itself. Making your service decentralized seems like it will add a lot of complexity and present some limitations. Why can't you just follow good security practices? As evidenced by Google, most people are willing to let a company know a lot about them in order to use a valuable service.
It has been just 3 years since I got my internet connection. And thus I am aware of what it means to be disconnected. Our initial idea wasn't even remotely about AI. I just wanted to make something which could act as a backup network, something which people with no internet(we have quite a share of this type in my city) could also use.
You see, its not just about privacy. A decentralized network helps you connect to people around you. And about the AI part being difficult, well, good stuff is always difficult ;)
You see, its not just about privacy. A decentralized network helps you connect to people around you. And about the AI part being difficult, well, good stuff is always difficult ;)
Solving a hard problem can indeed be a good strategy. Trying to solve two hard problems at once can prevent a startup from solving either one. Since both of these have their own independent value, I would suggest that you focus on one first.
I agree. The AI part of the idea is relatively new with respect to the network part. To be specific, we have been working on AI since January and to put things in perspective, we started working in December 2014. All this time, we just worked on the decentralization aspect and the plugin framework.
And regarding the implementation of these two separate ideas, well, Grid is both of them at the same time and how we present it depends on what demographic are we dealing with.
So, yes, it's one thing at a time for us :)
And regarding the implementation of these two separate ideas, well, Grid is both of them at the same time and how we present it depends on what demographic are we dealing with.
So, yes, it's one thing at a time for us :)
Can you talk a bit more on "And if someone on the Grid makes a query that needs your data then it is properly anonymized first."
"Notify me when the crowd outside [this] stall is below 50"
This situation demands the location of your device. Now, the app would hide your identity by using a hash in place of your name or MAC. This is our initial solution and if something better comes up or if a more complex situation arises then we would implement something better.
This situation demands the location of your device. Now, the app would hide your identity by using a hash in place of your name or MAC. This is our initial solution and if something better comes up or if a more complex situation arises then we would implement something better.
Could you give a specific example of what this would do?
Sure! You could make a query like "Coffeehouses around me with with at least 2 charging points.", and it would give you a list which satisfies your criteria but would also filter based on what it learnt. In case it learnt that you hate crowded places then it would reflect that in the results.
It can also give you personalized notifications. In my particular case, I have to wait for my bus to go to college and it usually is quite packed when it arrives, so I go to the depot and take the bus directly from there. Now, this assistant of mine could tell me that the next bus is packed and I should go straight to the depot or I could just wait for 2 minutes and take a seat in the coming bus.
It can also give you personalized notifications. In my particular case, I have to wait for my bus to go to college and it usually is quite packed when it arrives, so I go to the depot and take the bus directly from there. Now, this assistant of mine could tell me that the next bus is packed and I should go straight to the depot or I could just wait for 2 minutes and take a seat in the coming bus.
How will the system acquire relevant data about coffee shop electrical outlets, bus time tables, and a person's personal space requirements?
What local markets will be targeted first for data acquisition?
What data sets will be targeted first?
What local markets will be targeted first for data acquisition?
What data sets will be targeted first?
The system is all about crowd sourcing information. Android can tell when a phone is charging or not and phones using our app could tell the grid. So, that gives us electrical outlets in the coffeehouse.
We won't be accessing some public bus time table. Rather, the grid users riding the vehicle could give us real time and reliable data about the same.
And, a person's personal space requirements, well, honestly, that's a tough one, and it needs more thought than what we have given to our current flow.
Since, Grid is all about optimizing daily lives, we would target grocery shops, local general stores and weekly markets, initially.
We are relying heavily on the sensors provided with the phones, along with the most important, location data. We believe we could have at least an alpha if we start with these streams of data.
Since, Grid is all about optimizing daily lives, we would target grocery shops, local general stores and weekly markets, initially.
We are relying heavily on the sensors provided with the phones, along with the most important, location data. We believe we could have at least an alpha if we start with these streams of data.
Without a bus schedule, how will the grid determine that one bus is the one the user wants and another bus is not?
With the data being crowd sourced, what is the strategy for addressing the "chicken and egg" problem inherent when depending on network effects?
With the data being crowd sourced, what is the strategy for addressing the "chicken and egg" problem inherent when depending on network effects?
A bus schedule would have been of great help if the system was pretty organized. But, in my city, it's a circus show. Multiple buses and jams on bus stops. To curb this issue, initially, users would have to mark the buses. This is something like the bitcoin community, where the users of the system help in operating it and in return get its services.
Chicken egg problem. My archenemy. Our solution for now, is to rely on our "Extra feature" ie Plugins. Existing startups and businesses can extend their own functionality by making their plugins. For example, there are tons of anti-rape apps on the market but almost zero which work offline. This would convert their users into Grid users. We could also get businesses to create exclusive content or offers in return for our Ad platform/Analytics.
Just in case you are thinking that marketers/businesses will get your personal data and profile, then let me say this again, your data stays on your phone. The app in turn will filter out all the ads based on your preferences.
Chicken egg problem. My archenemy. Our solution for now, is to rely on our "Extra feature" ie Plugins. Existing startups and businesses can extend their own functionality by making their plugins. For example, there are tons of anti-rape apps on the market but almost zero which work offline. This would convert their users into Grid users. We could also get businesses to create exclusive content or offers in return for our Ad platform/Analytics.
Just in case you are thinking that marketers/businesses will get your personal data and profile, then let me say this again, your data stays on your phone. The app in turn will filter out all the ads based on your preferences.
Solution: A mobile application which creates a decentralized network using Bluetooth and WiFi and uses this network to power the AI assistant. The AI learns from your activities, preferences and routine and gives you suitable notifications when you need it. All your personal information and everything the AI learns is kept on your phone only. And if someone on the Grid makes a query that needs your data then it is properly anonymized first.
Extra feature: Just like any network, Grid also supports running applications over it. Anyone can make simple plugins for the network!
Status: The app can create a network and has been tested(It sometimes crashes on Android 5+ devices :p but me and my team are close to fixing that). The plugin framework is also complete and just the documentation is left. We are now focused on the AI part, and we believe we can have an alpha by the mid of may.
Background: We are a team of 6 and are practically in our final year of college. We are classmates and have know each other for more than 2 years and we have been working on this idea for the past 11 months.
The idea could sound quirky or boring, and I really really would like to know your opinion, your remarks, your rant. I am all ears to what you have to say.
Looking forward to your comments :D