THE INTEL(LIGENT) CHIP ARRIVES

Intel's new brain-like chip puts machine learning in the heart of your device

Dateline: 29 September 2018
AI is dead. Long live accelerated AI.
Now deep learning by computers has been internalised in the CPU, the very heart of the machine. It's called self-learning.
Cognitive computing required deep neural networks that had to be trained on extensive datasets, but Intel's neuromorphic chip combines training and inference for on-chip learning on the fly. That sounds like a mouthful, and it is.
Simply put, AI systems required supercomputers and advanced software to learn how to deal with different situations; hefty computing horsepower. Intel's Loihi2 chip learns from new data, without extra lessons from the cloud; like a human would.
That means your phone, car, camera and watch or drone can now make intelligent decisions much more quickly, without a data link, and using very little energy. Your robotic assistant is going to become a whole lot smarter, and use less battery power to boot.
It's called pushing intelligence to the edge of the network, and it means all those connected devices, on the internet of things, are going to be able to think for themselves; and learn on the job. A thousand times quicker than Whatsisname.
Soon your phone will know your heart rate is abnormal, and not just the result of jogging or chatting to an attractive person, and call the paramedic for you, even if you're unconscious. Which could be heaven, or it could be hell, depending on your point of view.
Either way, accelerated AI is coming to switch your future into the fast lane. Are you ready for it?
Published: 28 September 2017
Links to related stories
1. Intel’s New Self-Learning Chip Promises to Accelerate Artificial Intelligence - Intel, 25 September 2017
2. Microprocessors configured more like brains than traditional chips could soon make computers far more astute about what’s going on around them - MIT Technology Review, June 2014
3. MindBullet: EMERGENT INTELLIGENCE IS A WEB OF DEMOCRATIC DEVICES (Dateline 13 May 2024, Published 14 May 2015)
4. MindBullet: COGNITIVE SYSTEMS RULE, OK (Dateline 22 February 2022, Published 11 July 2013)
Warning: Hazardous Thinking at Work
Despite appearances to the contrary, FutureWorld cannot and does not predict the future. Our MindBullets scenarios are fictitious and designed purely to explore possible futures, challenge and stimulate strategic thinking. Use these at your own risk. Any reference to actual people, entities or events is entirely allegorical. Copyright FutureWorld International Limited. Reproduction or distribution permitted only with recognition of Copyright and the inclusion of this disclaimer. © Image courtesy of Intel.

Comments

Doug Vining

Doug Vining Truly intelligent machines will be able to plan

Harri Valpola, founder of The Curious AI Company, says all of the AI currently in use is second rate. He wants to change that. AI systems that are trained on lots of data can be better than humans at solving problems involving new data, like predictions, but they are incapable of the kind of abstract reasoning that humans and animals employ when they want to achieve something - planning. This kind of thinking is called 'relational reasoning'.

Harri Valpola wants to build an internet of artificial brains that are capable of planning, and through their interconnection, also capable of processing continuous data signals for relational reasoning. That would certainly make AI revolutionary. Perhaps the new neuromorphic chips from Intel will help to achieve this dream!
A FuturesForum post (titled: "Truly intelligent machines will be able to plan") refers to this MindBullet. The full FuturesForum post can be read here: http://www.futureworld.org/PublicZone/FuturesForum/BlogDetails.aspx?PostID=ebac4be5-2b55-4418-af66-3097e6cce312
Posted: 4 October 2017 at 09:04
Delete
Doug Vining

Doug Vining Machines teaching themselves to learn

Than sounds a little convoluted, but it's exactly what the latest version of DeepMind's AI, AlphaGo Zero, has been built to do. Learn from simulations, rather than real world data, and figure it out all by itself. And it works, exponentially better than its predecessor!
A FuturesForum post (titled: "Machines teaching themselves to learn") refers to this MindBullet. The full FuturesForum post can be read here: http://www.futureworld.org/PublicZone/FuturesForum/BlogDetails.aspx?PostID=456a6897-d4e2-49e9-b26d-e97d5c3f2614
Posted: 23 October 2017 at 17:28
Delete
Doug Vining

Doug Vining Embedded intelligence

Key to the mass adoption of machine learning and intelligent things, is the ability to drive intelligence to the edge, in everyday devices and sensors, rather than relying on supercomputers in the cloud to do all the heavy lifting. A startup called GraphCore is developing an IPU to replace GPU and CPU in machine learning, promising dramatic improvements in performance. So it's not just Intel building chips for the future of 'intelligence everywhere.'
A FuturesForum post (titled: "Embedded intelligence") refers to this MindBullet. The full FuturesForum post can be read here: http://www.futureworld.org/PublicZone/FuturesForum/BlogDetails.aspx?PostID=d4dda372-c334-42ce-b223-1d519edd3043
Posted: 26 June 2018 at 14:01
Delete
Doug Vining

Doug Vining Another chip for intelligence at the edge



A FuturesForum post (titled: "Another chip for intelligence at the edge") refers to this MindBullet. The full FuturesForum post can be read here: http://www.futureworld.org/PublicZone/FuturesForum/BlogDetails.aspx?PostID=e1afabb6-6464-45b3-9834-2b112b44ee41
Posted: 26 July 2018 at 15:33
Delete
Comments by users of MindBullets are those of the authors and are not necessarily shared, endorsed and/or warranted by FutureWorld. All MindBullet content is Copyright FutureWorld International © 2018. All Rights Reserved.