Hacker News Reader

Comment by neya | Hacker News Reader

I don't think Apple has advancements in machine learning tech as much as its competitors. For example, take a look at how primitive Siri is. It struggles a lot in comparison to its competitors even for some basic commands.

Or, take a look at Apple Music's recommendation system in comparison to Spotify's discover weekly.

Now, take a look at Amazon and Google. They both have their own cloud infrastructure (which ironically Apple uses) with machine learning built right into it. Both have state of the art image recognition systems (proprietary) for various applications (Eg. Google Image search, Amazon shopping, etc.)

I am not saying Apple can't do it, I just think they're lagging behind and this move totally makes sense.

View on HN

From what I read Apple are taking a different approach? Google/Amazon are harvesting massive amounts of data and training against it and Apple are taking a privacy first approach saying you shouldn’t have to harvest lots of data for intelligent AI.

Currently the harvesting data for models look to be working better for Google / Amazon with current technology. The question is how will Apple progress long term and given time can they achieve AI with some level of privacy. I don’t know enough to have any comments but like Apple’s public stance on this.

Thank you for this comment, long term is rarely even considered in consumer technology related to smartphones or "devices".

Are you suggesting Apple plans to do AI without harvesting data, or with transparent privacy controls on data? There's literally nothing for machine learning to learn besides harvested data. It's like suggesting that a power company is trying the strategy of using no wires for delivering electricity to houses..

I'd be excited if Apple found a way to trustlessly aggregate data and perform homomorphic training of their ML models, but they seem to be laying off that flavor of talent..

It's actually not that Apple isn't collecting data sets, but they're doing so at a more measured pace by thoroughly anonymising it using differential privacy. If the AI needs to use raw personal data in it's calculations, that is done on the user's device to keep that information private.

This appears to be slowing them down compared to their competitors, but it's not clear how much as against them being a bit behind for other reasons.

Others also anonimyzes data using differential privacy, k anonymity etc. It is a common practice and enforced when working with private / PII data.

> There's literally nothing for machine learning to learn besides harvested data.

You can obtain training data from sources other than your customers. Apple have employees driving vans around with cameras and LIDAR.

Is this what their differential privacy stuff is about? It's not trustless (not least because 100% of the data comes from a closed source OS they control) but they do seem to be trying to figure out a privacy-friendly way to harvest data to train ML models.

You seem to be equating AI and statistics-based ML. Statistics-based ML is certainly producing the most results of any AI technique right now, but it is not necessarily the only one that ever will.

To play devil's advocate: Perhaps Apple adopted this public stance as a strategy to hurt their competitors after failing to achieve parity.

You can harvest data and anonymize it. My friend at Google supposedly is doing this via research at Google. I'm not going to say Google care about our privacy as much as Apple but they are researching this probably for medical and for when opportunity arises. To be perfectly honest I think Apple's stance on privacy is all marketing, there are cases where they refused to unlock phones for the FBI but there are other cases where they ship recordings of your talk to SIRI to a third party company.

> You can harvest data and anonymize it.

Presumably the anonymization needs to happen prior to collection for it to be effective?

Yeh most company anonymize it because "data snooping" makes your algorithm worse. But try telling a layman that.

Worse performance is definitely not from lacking data. They have more than enough data to make the kind of dense samples needed for ML; remember that ML is about sampling and fitting a generalization to some population, not collecting all the data from every person and overfitting to it.

It's more likely lacking infrastructure, which OP alluded to. ML requires infrastructure that companies like FB/Amazon/Google naturally possess. Apple is playing catch-up.

Other comment also mentioned that Google et al anonymize data, and have options to opt-out of data collection. We don't have evidence that they are necessarily worse at privacy than Apple (they could be, we just can't know for sure unless you work there and happen to know exactly how it all goes down).

Apple has data on you too. They're just worse at making use of it (and you can argue less incentivized but I don't know -- getting people to use apple-only products is a pretty strong incentive to make things work well, and Siri/Apple Maps don't work that well).

Machine learning is literally a numbers game. The bigger and higher quality your data set, the better your models. So, it's contradictory that a company like Apple assuming they've taken a "privacy first" stance, can also simultaneously claim to be a competent player in this machine learning arena.

The other thing is, it's really tiring to see "Apple is a privacy first company" narrative on HN. Because, nobody knows exactly what Apple is doing behind the scenes with your data. There's literally no way to know what Apple is upto other than trusting their marketing material that claims to be privacy centric.

For all you know, they could still be doing stuff what other companies are doing secretly and nobody can prove nor disprove that. Next, iOS is closed source. Without access to source code, this privacy narrative is nothing more than just a marketing gimmick. We're a tech crowd and we should definitely take it a step ahead, investigate and then conclude, rather than blindly believing marketing posters that Apple puts out.

Once your data moves to a private cloud like iCloud, there's no way to guarantee anyone's privacy or safety for that matter other than blindly trusting the vendor. Don't believe me? Ask Jennifer Lawrence.

> Don't believe me? Ask Jennifer Lawrence.

You are aware that the attack was a spear phishing attempt, rather than an inherent weakness in iCloud?

> you shouldn’t have to harvest lots of data for intelligent AI.

Er.... that's not how it works. As a Valve employee put it (a talk about ML recognising cheaters), machine learning is just an equation:

y = f(x)

where you can actually derive the FUNCTION f if you have enough x/y pairs. Without those pairs then most of the successful techniques in machine learning fail to function.

Yes you can generate your own data but generating realistic stuff is a huge can of worms.

Not true. You can code your prior knowledge into the model, and make do with way less x/y pairs.

Data is a substitute for doing engineering by hand. If you do the latter, you require less data.

I don't see how it makes my statement untrue. The most effective ML techniques are more effective the more x/y pairs you have. Sure, you can get a leg up if you partially define F yourself but then you're no longer being as hands off as before. In practice the WHOLE of F() will likely be a blend of machine learning as well as heuristics as a product of various layers that are one or the other.

Not the poster you're replying to, but it's just that you said "that's not how it works", and then suggested that the only way to get F is through data, even if it means generating synthetic data.

So that's wrong and untrue, as you recognize in this latest comment by saying you can partially define F. To make a concrete example, the convolution operator, the number of layers, their connectivity--all of that partially defines F. Finding a structure is what allows you to learn from the x/y pairs instead of just storing them in a look-up table.

It makes sense, then, that finding a better structure may allow you to learn from fewer x/y pairs. This is Apple's purported approach.

It also means that if you have a huge number of x/y pairs, you might benefit from (or get away with) using a simpler structure. This is Google's approach in the domains where they have a lot of data.

It is true that an important lesson has been "when designers build structure into machine learning systems, they often build in the wrong things, and they end up making commitments that the data cannot undo". But there's still a lot of structure that we put into machine learning system, and we're still learning about what good structures are for different applications, and different regimes of data.

Traffic is a chaos system, you can't engineer the edge cases. At best you have heuristics.

I don't think "traffic is chaotic" is the issue with driverless cars. We'd be satisfied with a car that solved everything except for global (i.e. non-local) traffic flow issues.

That said, a dynamical system may exhibit chaos for one setting of the parameters, and may not exhibit chaos for other settings of the parameters. It may be true that the equations governing traffic flow have the potential to exhibit chaos, and do exhibit chaos when 1. humans drive, or 2. the drivers have only local sensing. But that doesn't mean it would still be true for driverless cars.

But again, that's not the issue at play. We wouldn't fault an driverless car for getting stuck in traffic. It's more like it has to work when it's snowing, when you can't see the lane markers, when a kid in a trashbag costume jumps into the street, etc. Those are the edge cases that we care about right now.

Plus, I don't quite get what is the privacy being violated when taking a lot of pictures of red traffic lights...

Mine the exif data for the time and geolocation of the people taking the pictures and what camera/phone they used. You might even be able to tell which direction they were travelling in.

There are different ways of training things than relying other people's data. For instance alpha zero is still the most remarkable achievement in ML or 'AI' in my opinion, and it used 0 human games in learning to play chess. It training was entirely from self-play. I think models that rely on learning from these sort of weighted monte carlo type systems are likely to end up far more robust than ones that try to learn from humans. In the AlphaZero case this was provably true.

But in the more general case, the optimal solution for humans and the optimal solution for machines is not necessarily the same. So in driving, for instance, it may be the case that operating a vehicle in one way is safest for a human - yet not necessarily so for a machine. It's not hard to come up with examples. We are awesome at intuition and awful at microsecond reactions. Computers are awful at intuition and awesome at microsecond reactions. The optimal style of 'play' is likely going to differ. Yet if the machine is trained based on human data it would end up trying to shoehorn itself into replicating human behavior which may be suboptimal for it.

All that said, this sounds more like marketing talk. My guess is that Apple simply decided that self driving is extremely difficult and is going to come with some major growing pains which could negatively affected their crafted brand image, making it generally a poor investment of time and resources.

>you shouldn’t have to harvest lots of data for intelligent AI

That shows they don't understand how AI work. That literally like code for traditional programming. It's literally impossible based on how modern ML works.

You are conflating AI and ML. You can have AI with no ML.

Apple is far behind its competitors in pretty much every area except hardware. They are still terrible at running stable web services. There is a new critical bug or security hole in their OS every month. And all of their non-hardware offerings - Apple Maps, Apple Music, iCloud, iTunes, Siri etc. - range from very bad to average.

IMO they need to be investing a LOT more on the software side of things if they intend to stay competitive, especially since iPhone sales are already drying up.

The major differentiator of Apple hardware is the software that runs on it. iOS and MacOS are basically unmatched by other offerings.

In what meaningful way are they "unmatched"?

Speaking for myself: a combination of ease of use, features, quality and a mature ecosystem. They certainly don't have a perfect track record in any of these, QA being a well-documented example in the last few years. And the alternatives have gotten better, but they're still not good enough to convince me to switch.

iPhones have better hardware/software integration, which is easy for Apple to do because they control everything. They get better performance out of their hardware because of this.

Sales levelling off != drying up.

Apple has always had problems with their online services, but iOS security is a problem? Compared to what, the dumpster fire over in Android?

I was referring to macOS in that line.

Phones are moving towards commodities even apple found this when the X got push back at $1000 per unit.

I think the price is a mistake, but I don't think it's "failure" reflects a weakening Apple's of brand competitiveness as much as it is a sign of the relative maturity of smartphone hardware.

In other words, people are buying phones less often, because phones for 3-4 years ago are still quite capable of meeting the needs of most users, so people are naturally going to start buying phones less often.

I suspect Apple was very aware of this trend, and the price increase was an effort to increase margins to make up for an expected decrease in unit sales. After all, Samsung also indicated lower guidance on their Smartphone numbers for this year, so Apple is not the only company affected by this trend.

Maybe they missed the mark a bit, and a significant number of potential customers chose not to upgrade this year, but I seriously doubt those customers turned around and bought an Android instead. I suspect they will just get plan to get another year out of their 6S and wait for what comes out this year.

> I suspect they will just get plan to get another year out of their 6S and wait for what comes out this year

This is the situation I'm in. Have a 6s that I'm not in a rush to upgrade until something appealing appears. It's plenty fast enough for everything I do. I actually started looking at Android alternatives thinking I could get a high-end phone with a removable battery but learned that's not a real thing anymore. Plus I don't like the idea of using too many Google services.

Still not sure what my upgrade path with be yet. Maybe an iPhone 8 but it feels like I'm just getting the same phone as I've got but without a headphone jack.

They can still charge a premium over any other phone manufacturer, and they sell at the about the same volume. They have a lock in that seems likely to protect that premium for the foreseeable future. As long as they can charge a premium they get a nice profit margin.

iOS is arguably the most secure operating system in widespread use. Not sure why the OP thinks otherwise. macOS while not as secure is still in pretty good shape. But largely irrelevant to Apple's financial future.

Their power cord keep on dying for my macbook 2013.

I had a temp iphone and it took over a lot of my texts by redirecting it toward apple's server first. When I got my android back I receive no text msgs from my friends who have iphone.

It also seem like they're making cheaper stuff so they can increase their margin via tech support and peripherals.

I agree they should invest more in software so they don't have to resort to these crappy tactic.

Also they should make a cheaper iphone region lock it to china and india only if they want to increase their margin. Their current strategy for these markets is stupid.

There are announced security holes in every operating system. But unlike Android, iOS users can actually get security patches for years after they buy their device.

Android still has a horrible permission system after all of these years and privacy is laughable.

And as far as “iPhones sells are drying up”. They are still predicted to report the second highest revenues this quarter of any company in history - only surpassed by their previous record.

"They are still predicted to report the second highest revenues this quarter of any company in history - only surpassed by their previous record."

Profits, perhaps, but not revenues.

Yes because we all know that revenue is more important than profit.....

I didn't say that it was.

This is totally unfounded. Their market-leading bokeh camera effect is powered by a “state of the art image recognition system” including custom silicon (the “neural engine”).

Siri has gotten a lot better especially at natural language recognition.[1] The HomePod for example is exceptionally good at detecting Hey Siri commands while loud music is playing. That’s all local machine learning processing coupled with a great mic array. It’s more the breadth of services that are the issue, but that’s not machine learning.

The music recs.. well you don’t cite anything. I can say the weekly For You new music mix works good for me.

[1] https://www.cnet.com/news/siri-vs-alexa-vs-google-assistant/

I keep hearing that people noticed Siri getting better, but my experience is totaly different.

For example, how can it screw up something as simple as enabling alarms: https://ibb.co/vkYpGcq (this has happened multiple times to me).

Why most of the questions are answered with a google search link? Why it keeps randomly forgetting context (example: I tell her to wake me up at 1pm, then repeat just 1:01pm, and so on to add more alarms. Most of the times, it works, but sometimes it stops understanding what I want, even though the speech recognition is right and I’m saying the request in the same format as my previous requests, just with an incremented time).

I’m not saying other Apple products can’t use AI, just that Siri seems to be very low on their priority list (and I don’t blame engineering, this is something that comes from management).

I complained a few months ago that Siri had a tough time recognizing my accent heavy speech but these days it is better.

I'd add that Apple News has had a much better hit rate than Google News. It consistently has added new suggested categories based on my observed preferences.

What you're describing on the HomePod is "ASR," automatic speech recognition, which is a separate function from natural language recognition. ASR is converting spoken language to (the correct) words, basically dictation; NLR is "I have these words, now what do I do with them." This is a distinction I've learned from working at Viv Labs, which is producing its own voice assistant. (Samsung's Bixby 2.0 is basically our Viv, and it's an entirely different product than the 1.0 version, which was basically a rebranded S-Voice.)

Having said that, I think Siri is better at NLR than it's often given credit for, and Alexa is much worse than it's often given credit for; Alexa's NLR seems less akin to its competitors than that of a 1980s text adventure, i.e., "use exactly this precise phrasing or you won't be able to [get the treasure | turn on your living room lights]". Amazon's real strength is in making it so damn easy to write Alexa skills and to forge deals with hardware/appliance vendors, which goes back to your "breadth of services" observation.

They reason is because they don’t snoop on their users and so don’t have good training sets available.

Training set for driving is more readily available.

Apple originally kept Siri audio for up to 2 years "for testing and product improvement purposes". I am not sure if that is still the case.

https://www.wired.com/2013/04/siri-two-years/

The formal grammar structure behind smart assistants is basically the same. Siri is as primitive as Cortana, Google Now and Alexa, the difference comes in the amount of data the interfaces have access too. Outside of the text to speech and speech to text algorithms, there is barely any machine learning when it comes to understanding what the user says. As of it right now comparing smart assistants is equivalent to comparing dictionaries, Siri having less entries and people writing them than the others.

There’s a lot of unknowns with liability and regulation. I’d rather let someone else take up that legal risk than spend all this money on basic research just to have my shit shut down by ordinance or a manslaughter charge.

Apple bought Shazam, and their recommendation system is quite good.

Reply on news.ycombinator.com