Hacker News Reader

Comment by tomnipotent | Hacker News Reader

Why are you surprised that Apple isn't involved in a field that has no material impact on their business? Outside of Siri and related projects, I don't see where AI/machine learning has any immediate benefits.

Twitter and Intel missing seems like something to take note of, but I didn't even think twice about Apple. I'm more surprised GE is missing than Apple.

View on HN

I use Android, but Apple are doing some good personal assistant stuff like tagging friends in photos, -without- uploading them to the cloud. A true personal assistant that can use machine learning locally and with discretion sounds pretty great

I'm really starting to appreciate this kind of thing about Apple. They seem to be concerned about keeping data in the hands of their users as much as possible, when the rest of the industry defaults to a cloud-hosted solution that requires the user to completely give up control.

That's a nifty feature, but I am sceptical that Apple can offer the same kind of experience when compared to cloud-based solutions that leverage massive amounts of aggregated data and backend processing.

Apple has made a pretty big deal about privacy, but they do collect user data (see the interview a few weeks ago with Apple's mapping lead -saying they are collecting user data to improve maps).

Ultimately if users want ultra smart AI assistants (Siri, Google Now, Cortana, etc.) they will need to give up some of their data.

That may be very unpopular with the HN crowd, but I suspect the vast majority of consumers will be OK with that tradeoff.

Apple also works on technology that is meant to help users give up less of their privacy. They talked about this at WWDC 2016 under the headline 'Differential Privacy'. They're trying to use statistics to make it harder to attribute data points to individuals while still providing a workable data set on the collective of users.

Here's an article about that: https://www.wired.com/2016/06/apples-differential-privacy-co...

I use Google translate offline on Android when in China and it is great. Seems uploading the trained NN to a phone when online and then using it privately/offline is a great way to go.

At WWDC they talked about how they put AI all throughout the system. It's not just when talking to Siri. It's things like suggested apps, photo tagging, deciding when to refresh apps in the background, text prediction, and probably lots of other stuff that I can't even think of because Apple's approach to AI is to make AI largely an implementation detail of the experience rather than a user-visible component.

>> Apple's approach to AI is to make AI largely an implementation detail of the experience rather than a user-visible component.

That sums it up nicely, I'm a bit embarrassed those features didn't cross my mind when thinking about Apple/AI. Reinforces your point - its use in invisible, much like their presence in this consortium.

Apple makes great stuff (typing this on a Mac), but I really don't get the adoration / mystic that surrounds them.

Those features have been on Android forever, but people don't call it "AI".

Right, which begs the question "What is AI?"

In this context, "AI" means "using techniques that come from the field of AI research", such as convolutional neural networks and whatnot.

The big area for AI is to replace a human being with an AI program and not have the customer notice it.

Then those webchat and help desk jobs will be gone and replaced with AI doing robocalls and acting like a chatbot.

I'm getting Robocalls that ask for a Chris or something and as soon as I say there is nobody here by that name (so they can take me off their list) the bot cuts in saying "Maybe you can help?" and then tries to sell me something. Most of the time the Caller ID is spoofed and it calls back with a different Caller ID and number, etc. I'm on the Dot not call list, but Robocallers just war dial numbers until it reaches someone and then tries to hook them in.

Not only that but AI can take the jobs of writers, programmers, managers, legal assistants, and in robot form burger flippers and checkout kiosks.

As we reach 2020-2030 the AI is trying to get as smart as a human being. As long as it has electricity it can work 24/7 without getting tired or hungry like a human being would.

Because machine learning is one of the most important tricks they're going to be able to use to make their devices even more intuitively usable. There should be entire teams at Apple salivating at new ways to extend powerful functionality to users _without_ making the interface to do so gross and complicated. And they probably are -- they've acquired several ML startups lately. (Hi, Turi friends!)

I'm surprised at this "Apple not AI" narrative. I've had Google Maps & Waze on my iPhone for years and neither of them seamlessly turned on "where did I park my car" feature. I just turned my iPhone on one morning and it just told me. Now, mind you, this may not be a fancy deep learning model. It could be some simple linear model, maybe a tree model, hell, maybe it's hardcoded rules? But despite these things Apple delivered a great feature which is in the realm of what I would consider "AI enabled feature".

Mind you, I don't have an Android device, so maybe Google Maps does this automatically for you on Andriod. It doesn't do it on the iPhone, at least not automatically (aka. I don't bother to look it up, which is kind of the point).

It's not Google Maps, but rather Google Now that has the "where did I park my car" for quite awhile (2.5 years) now:

https://support.google.com/websearch/answer/6015842?hl=en

Google Now generally seems to be where all that automated stuff goes, so if you don't use it that would explain why you don't see any of it.

Google Now does this automaticly for a very long time. They don't have enough permissions on iOS, I suppose.

Cross referenced it, because Cortana can't do it in iOS for that reason

Well, not surprised of course! Anyway, perhaps I'm just a simple man, ... besides searching the web, I think that's the best AI type feature for the consumer yet!

It uses a combination of Bluetooth (if you're near/connected to your car's stereo), motion sensor and GPS.

If you started moving at 70KM/h nowhere near a train or bus station, and then suddenly stopped moving at that speed and shortly after started moving at 5KM/h, you've clearly just got out of your car. Use bluetooth for extra verification if possible. When certain, get location (or load location) from GPS and mark spot. Tadaa, no fancy learning model needed.

Those sound like features that would be perfect for a learning model, otherwise you're stuck hardcoding the conditions and thresholds, no?

edit: anyway, besides the point.. I just love that feature haha

Yeah didn't mean to sound that abrasive.. damn booze.

It is a nice feature indeed

Acutally not neccessarily. You can as well end up in a traffic jam. In fact that's how they display traffic info on maps: many phone users doing what you just described.

The new version of the Photos app does image recognition. It seems to work fairly well, and unlike Google Photos, happens locally on the device and doesn't require cloud uploads.

Reply on news.ycombinator.com