MIT Lecture at Accenture:
“Developing Collective Intelligence: Can we get there by 2010?”
by Prof. Srini Devadas, MIT
On the one hand, there are technical requirements; on the other hand, there are the socio-political networks required:
Put these together and you’ve got “CLOUD Computing”. Now, what is “CLOUD Computing”?
= on-demand computation + data access anytime from anywhere.
…which means, billions of computers providing machine power.
BUT, there are many components of the solution that cannot be done by computers. So you need HUMAN POWER! Let’s call that “CROWD Computing”.
= billions of humans providing brain power…
What are emerging (online) applications of this?
- Amazon EC2
- Google ApplicationEngine
- Microsoft Azure
Applications and Security Models:
à Individual user backs up public data
à User backs up private data
à User wants to back up and share private data (e.g., photos, Flickr)
à User wants to run a private application for private data
Integrity of Application
RECALL our 3 Scenarios: Issues:
Ongoing Research (especially at MIT):
à Trusted Computing Modules (TPMs) = small crypto-processors
à Fully homomorphic encryption (private computation) (cf. IBM researcher, Gentry, 2008) – data encrypted by client to produce encrypted result that can be decrypted only by the client
à Designing secure systems
* Issues: Information flow tracking, Security testing of systems and applications software, etc.
Computational CHALLENGES in our 3 Scenarios:
Let’s look at the HUMAN INTELLIGENCE angle in every Scenario:
The GOOD news and BAD news about computational power:
Let’s consider again: ‘CROWD Computing’…
Economic motive: You wanna get the solution cheap !
= you want to harness the power of a group of people which computers cannot provide … affordably
= You want to contract the services of a group of humans, make them compete, for a fee, to get to a goal.
RESULT (via mturk.com): Better, quicker results! … more cheaply 🙂
- “ESP”, an Online Game developed by Lois von Ahn (CMU) : Label randomly chosen images ― The more accurately you label the image, the more points you gain…the more $ you earn…
Back to our 3 Scenarios: CLOUD + CROWD Computing :
q Match image (of rash) with images from a database; Best match? Let Doctors ‘compete’, then identify a doctor as the ‘chief’ (one who has the last word)
q and so on and so forth…
q and so on and so forth…
‘Collaborative Coding’ (CC):
* Throw in Human competition, then ITERATE
* you can use mturk.com
* You can MICRO-OUTSOURCE 🙂
* You can use the CLOUD to automatically generate code, e.g., Convert “housekeeping” rules to high-level policies 🙂
Can we get there by 2020? APPLICATIONS:
* A million software ‘developers’ together write a distributed online video game.
* Seen on a blog !
“Global Math community”
* Neighborhood Crime Prevention
* There are TRAFFIC problems everywhere! How about a traffic-solving community? 🙂