Apple Goes on Hiring Spree To Improve Siri's Smarts (engadget.com)

Thinknum has noticed that Apple has been on a Siri-related hiring spree in recent months, with 161 positions listed as of the end of March. There's been an uptick ever since summer 2016, but there was a distinct surge in February of this year. From a report: Nearly all of the new positions are for engineers, and it's notable that some of those spots are meant to tackle Siri's relative weakness in general questions. There are entries for Siri software engineers tackling "General Knowledge" and the "Knowledge Graph," for instance. Other notable recruiting efforts include a machine learning scientist as well as multiple engineers to address areas like Proactive Intelligence (Siri's ability to anticipate info you might need), the SiriKit developer framework and speech generation.

Comments

Amazingly

[110010001000]

Throwing people at the problem won't work. Siri can't even understand follow up questions like: "Who is the President of the US?" followed by "How old is he?". So much for "AI". It isn't even a very good Eliza.

Re:Amazingly

[Dutch Gun]

I just tried that with Alexa, and it recognized the follow-up question. So, good for Amazon on that one. But Alexa still falls woefully short in other areas, though. For the shopping list functionality, you can't even say something simple like: "Alexa, add milk, eggs, and cheese to my shopping list."

Alexa: "I added "milk eggs cheese to your shopping list."

Me: "Alexa, remove "milk eggs cheese" from shopping list.

Alexa "You can remove items from the shopping list in the Alexa app."

*facepalm* This is why I have a good chuckle whenever someone calls these digital helpers "AI". Well, it's still early days, so these things will improve over time, presumably.

Re:Amazingly

[PolygamousRanchKid]

Alexa "You can remove items from the shopping list in the Alexa app."

It could have been worse. Alexa might have answered:

Too late! The articles on your shopping are already shipping from Amazon!"

Re:Amazingly

[jellomizer]

Throwing people at the problem does work. Only if the company listens to the ideas of those who bring up suggestions and those who brought up suggestions are willing to change their mind if a better one is given.
This is a tough haul for any organization.

But also it may be that Siri development is under staffed. A lot of good ideas and designs but just not enough man hours to implement.

Re: Amazingly

[Anonymous Coward]

Siri is supposed to be stateful.

Go back and watch the keynote where it was introduced. The demo was something like:

"What's the weather?"
(gets answer)
"How about in Chicago?"
(gets answer for Chicago)

The implication has always been the Siri is stateful and will understand followup questions. Except that, as anyone who's tried it is aware, Siri is not in the least stateful and doesn't understand followup questions. Although there are also exceptions to that, as certain interactions always have followups. For example, if you ask for directions to a location, Siri will then ask you if you want to call the number for that location or get directions. (Needless to say, this is somewhat infuriating, especially while driving, which is when you'd use Siri rather than just search directly.)

Re:Amazingly

[swb]

You laugh, but people will probably figure out how to do "voice ai squatting" on Amazon, selling bogus "combination" products that match this (ie, some product that is called "milk eggs cheese" but isn't any of them).

Re:Amazingly

[arth1]

But also it may be that Siri development is under staffed. A lot of good ideas and designs but just not enough man hours to implement.

I doubt that adding man hours to implement ideas and designs will make Siri any better. It will make it bigger and some functionality will change, but it will not necessarily become noticeably better.

I think systems like Siri need to be able to learn on their own, without any well-meant tweaking or shortcuts that hinder context learning. Hire thousands of people to grade the answers it gives without well-meaning engineers attempting to add right answers or well-meaning management censoring some answers.

One day, it may be smart enough to understand that when you ask "Where does Jane work?" followed by "What's her phone number?", you likely want her work phone number if it's daytime, but if followed by "What's her address?", you're less likely to want her work address. But simply tossing more manpower at programming won't get us there - it will likely get a more bloated expert system with more hardcoded shortcuts that prevents actual learning.

Re:Amazingly

[Dutch Gun]

If they ship me an omelette, I'll be genuinely impressed.

Re:Where's my pony?

[DontBeAMoran]

https://mylittlepony.hasbro.co...

They need to fix their speech-to-text first

[JoeyRox]

You can't improve your response to queries when you can't even get the query itself right. Siri still has trouble with basic dictation. Garbage In, Garbage Out.

15 Years Behind

[bobbutts]

About 15 years ago Google was aggressively recruiting expects in natural language processing.

Context dependents era and the new turing test

[goombah99]

The new turing test is something we get to play every day-- ask your phone a quetion then refer to the previous question or an obvious inference from it. Can the phone follow you like a human would?

Were still in the context dependent era of AI. We have not yet produced an AI that can handle the non-sequitur. Indeed siri does less than even that, it doesn't follow a train of thought. You can't refer to something you just said in a new instruction.

The problem Siri has is that it's trying to do too much with too little context. Google and the self driving cars do their jobs seemingly more proficiently because they are context aware. Google by spying on you so it can guess what you are doing when you ask it something. Cars because, cars are found on roads and roads have rules.

I think Siri would improve fastest if apple could bridge the gap between restraint in collecting information and selling information. Google has no such qualms.

Re:Where's my pony?

[antdude]

In a factory to make glues. :P

Ob

[Hognoxious]

Siri, fuck off!

There is no fuck here..

Buy up Cycorp

[Tablizer]

They should buy up Cyc (Cycorp). If somebody finds a way to combine neural nets with knowledge-bases like Cyc, they could have something much closer to everyday common-sense. And, more trace-able.

Re:Where's my pony?

[Shemmie]

Well that was a naff April Fools Day.