On Apple and artificial intelligence
The best place to follow my thinking on AI and other technologies to sign-up to Exponential View.
Does Apple understand artificial intelligence?
Well … it depends who you ask. There were announcements at Apple’s developer event this week which suggest Apple is going to continue its investments that improve user experience through technologies machine intelligence technologies.
Three things stood out. These suggest that Apple is taking the opportunities of machine intelligence quite seriously.
- 1Apple announced support for basic neural networks in iOS. In the short-term this means that Apple can offer accelerated inferencing (predicting) on iOS devices, including accessing the phone powerful GPU process. Today, predictions (such as image recognition) happen on servers on the internet and an app on an iPhone needs to send an image back to the Net and wait for a response. If there is not internet access, the recognition system won't work. With Apple's basic neural network support, simple machine intelligence applications can run locally (and quickly) on the device saving a round-trip to the cloud. This will improve user experience. This isn't perfect, of course. There will be limitations, due to the memory and GPU footprint on the device, to the kind of inferencing that can be done - don’t expect a full automated speech recognition library to fit on a phone just yet. Of course, Google is pursuing a similar route to move TensorFlow onto Android devices and that team has at least one Apple alumnus who previously woked on image processing applications using early GPUs.
- AI systems need scads of user data to learn about the world. And this may compromise end user privacy. Apple introduced ‘differential privacy’, a cryptographically secure method for training over user data without compromising individual privacy. It “looks like Apple is honestly trying to do something to improve user privacy” says Matthew Green, a world expert in the subject. This is an important move - and let’s see whether consumers can understand it enough for it to be a product differentiator vs Google, Facebook or Amazon, or pressure higher standards of personal privacy elsewhere on the Internet. (MIT Tech Review looks at Apple & differential privacy .)
- Apple opened Siri to third-party developers, allowing them to plug into the voice interface. This is similar to the approach taken by Viv and Amazon’s Alexa. It seems like this will be the way we will get seamless voice interfaces that give us access to a broad, general range of underlying resources.
These seem like practical steps in using AI to improve products. None of this really encouraged the meme that “Apple doesn’t get AI” to die.
But Apple is a firm that cares about the UX almost to distraction. And the realms of artificial general intelligence and artificial super intelligence are not at the point where they do improve the UX, so it’s unsurprising that Apple isn’t announcing systems that beat Go or model the human brain.
For evidence of this commitment to UX, take a look at their 1987 vision of the future the Knowledge Navigator. Even today we don't have the technologies to bring this to life, but one day we might - and it will be chockful of AI.
More practically, Apple has been buying companies that deliver the AI stack for the past few years. These include Polar Rose (machine vision), Chomp (search), Novauris (speech recognition), Cue (virtual assistant), Topsy (stream processing), Acunu (big data analytics), FoundationDB (databases), Metaio (augmented reality), Perceptio (machine vision) & Emotient (machine vision). Expect more AI in Apple’s products.
But I would be surprised to see large-scale open source efforts, of the kind we have seen from Google or Facebook. Open source has rarely been Apple’s bag.
It was encouraging to see Apple continue to take steps to use machine intelligence to improve its products. We could hardly have expected them to do anything else.
Azeem
KEEP UP TO DATE WITH ME
For more on artificial intelligence, sign up to my newsletter.
--
8 年Let's wait what Apple is doing new in India.
Semi-Retired Engineer
8 年This is a huge step. There should be no need to run everything in the cloud as advanced programmers judiciously implement smart normalization and/or dimensionality reduction--as well as the inevitable sharing of processing power available through HomeKit/CarKit/etc.--to optimize algorithm efficiency.
Manufacturing IT/OT | Digital | Data | Automation | Integration
8 年Artificial intelligence is taking away the intelligence that humans have.