Siri’s Inventors Are Building a New AI | Hacker News Reader

Siri’s Inventors Are Building a New AI

wired.com

86 pointscyphersanctus44 comments

From the article: "Google Now has a huge knowledge graph—you can ask questions like ‘Where was Abraham Lincoln born?’ And it can name the city. You can also say, ‘What is the population?’ of a city and it’ll bring up a chart and answer. But you cannot say, ‘What is the population of the city where Abraham Lincoln was born?’”

Out of interest I tried this question on WolframAlpha and it happily returned the answer[0].

[0] http://www.wolframalpha.com/input/?i=What%20is%20the%20popul...

You can also ask this query directly into Google's knowledge graph[1] - but the point still stands that Google doesn't have a way to formulate that query directly from a natural language sentence.

[1] http://www.freebase.com/query?lang=%2Flang%2Fen&q=[{%22id%22...

But, the answers are useless to me if they're returned as images. (And yeah, I'm aware of OCR). I'm not saying the results are returned as images right now, maybe they are, maybe they aren't, but Wolfram has pulled this stuff in the past so my trust in the service is zero.

I've said it before, but articles like this do more harm than good for the field of AI research. Overhype has killed interest in the field numerous times before, and if we're not careful it can do it again. Advice to reporters who have an AI story: please don't hyperbolize both what the AI can do and how it works. If AI companies over promise and under deliver enough times, interest in an exciting field may die down again.

How is this overhype? If anything I feel it's a pretty mundane system they're describing - some sentence parsing along with various database lookups. They seem to stear pretty clear of AGI hardness.

There's a delicate balance to strike between hype and optimism, but I agree that a lot of writing on the subject lands squarely in the realm of unfulfillable hype.

This is a lovely idea, but it seems less "AI" and more a kind of natural-language SQL for diverse data sources.

Not that that's not pretty cool too. It's a surprise to me that even today, for many of the best shopping websites, I can't search for furniture that fits certain dimensions, chairs and settees that could be delivered through the size of my door, etc. There's no app that can show me the closest retail store to me which sells Widget X and has it in stock NOW.

Even the smartest property search websites can't show me houses for sale within a minute's walk of a bus stop and three of a train station, in a certain geographic area, where cable TV or fibre broadband is available and is within walking distance of a nice fish and chip shop. All these data points exist, but nobody yet seems to have identified the market opportunity in linking them all together.

It's not as if the general public don't see TV shows like 24 and Criminal Minds doing the whole "Chloe, show me all convicted felons within a 5 mile radius released from jail in the last six weeks with a history of ordering Chinese food on Tuesdays", after all. They must realise what happens when you put data together, and how useful a technique it could be for answering even everyday questions.

AI is a funny field. As soon as we understand something (read: build something) that we think is AI, it's no longer AI, it's just the result of some computation.

As humans, and more importantly in this context, things with "real" intelligence, we don't like the thought of explaining away our consciousness and experiences. I'm inclined to think that natural-language SQL for diverse data sources wouldn't just be AI, it would be real intelligence, on the order of what humans have, complete with qualia. See this excellent Radiolab episode[0] where they discuss an experiment about the relationship between thought and language.

On a side note, you should check out www.walkscore.com if you haven't already. It's a great website for looking for houses (to buy and rent) that does some of the things you mentioned.

[0] http://www.radiolab.org/story/91725-words/

Doing anything with everyday natural-language pretty much requires AI.

Offshoot comment: I'm not sure what that style of writing is called, but I hate filtering through all the backstory, and human interest stuff to find out what the technological advance is.

A tool taught in Journalism 101 is the "inverted pyramid": the idea is to structure your article so that the biggest details (5 W's) are at the top, and move down the information gradient with the smaller details ("human interest stuff") at the bottom. The writer may have skipped class that day.

The writer may have skipped class that day.

The writer took a class that seems to be very common in contemporary journalism, and what it teaches is very annoying to me, and apparently to the parent commenter as well.

This is longform journalism, which doesn't follow that convention, and focuses more on the "human" story.

This is likely more PR puff than actual news reporting. This company was mentioned on NPR this morning as well. I'm guessing they're just making the rounds and building interest.

I don't know if I'd qualify this as "longform journalism", the article is pretty small. I can see where the writer may have been going for that style, but that requires a commitment to a certain minimum amount of content (and, for readers, a certain amount of insight gleaned to make it worth the time investment). Definitely a fluff piece, as you said.

It's called 'burying the lede' and while it would get you fired from daily-newspaper type reporting (where you should articulate the important facts as clearly as possible up front) it's a staple of American long-form journalism. I don't know enough about the history of American literature to know who pioneered this sort of rambling slow-reveal narrative style - maybe Mark Twain? It can be very pleasurable to read for relaxation, or if the work aims to slowly introduce some philosophical idea. Indeed, it can be essential in book-length works, in order to make complex ideas digestible by leavening them with experiential information that provides the reader with valuable context.

But a lot of the time it can just be annoying or a stylistic cliche. As soon as I realize I'm reading an article like this, I usually skip towards the end to see whether there's any sign of a significant payoff. Sometimes there is and I'll go back to the start settle in for the ride, more often I pick up the few facts that I wanted to know and save myself a lot of eye-rolling.

To summarize: Viv is a smarter Siri, that could be implemented in various devices and applications, in different contexts.

However, “Viv is potentially very big, but it’s all still potential”.

The article also, is nothing but hype.

Solving natural language query is at least as hard as asking the system to write a program for an arbitrary goal. So if you can ever build a program that can take an arbitrary goal and output the program source code then you can also solve arbitrary natural language queries efficiently.

So most "supposedly" AI systems do brute force or probabilistic templating - i.e. mapping the query to a known structure of sentences. Then you get translation of these known structures of sentences to known machine level structured queries. This is why systems like Siri would fail on its face as soon as you ask even mildly "hard" question. For example, queries that starts with "How many..." such as "How many teeth humans have?" are very easy to solve and in fact it can be your weekend project for Wikipidea corpus with impressive accuracy. But if you change this query to "What is the factorial of the number of teeth that human have after subtracting number of legs octopus has?" then you would get no where.

Interesting, if we consider building AI as a startup, what would be the answer to standard investors question - what problem does it solve?

It allows complete confidentiality and 100% presence for your personal assistants. It also brings down personal assistants to the price that everyman can afford... and that's where marketers show why you could never live without one.

Ok, it's me maybe, but why would you need personal assistant, really?

Wait till you have kids, high-powered jobs, and information-dense hobbies. I would love to have a personal assistant.

Jaba, I see you're drunk. Laser Sintering 3d printers are 50% off today. Shall I go ahead and order you one?

EDIT> And a property. Fucking houses.

That's why I made the comment about marketers. A personal assistant is something you never knew you couldn't live without.

There have been a couple of companies that have tried this, mainly Numenta (Jeff Hawkins) and now Vicarious (Dileep George). The latter it's had a successful funding round, getting millions from big-name investors and entrepreneurs.

The value proposition for both of them was creating biologically inspired AI, and Numenta eventually came out with an anomaly detection program, Grok. I think it was largely bankrolled by its founder (he was also the inventor of the palm pilot), though I'm not entirely sure. The value proposition for Vicarious is that it has a secretive, general-purpose AI (not to be confused with Artificial General Intelligence, aka Strong AI), but save a few videos of their program solving CAPTCHA's, nobody knows too much about their offerings yet.

Numenta is very much alive. Grok is available in the form of an AMI on the AWS EC2 marketplace, and if you're curious about how it all works under the hood, NuPIC, the core implementation is free software, available at https://github.com/numenta/nupic

When they say Siri is “partnered” with different companies, does that mean that it can draw from and direct to the companies' services/menus/store catalogs? Does that exclude smaller companies that are not yet able to afford a partnership, thereby bypassing them when making recommendations to the user?

If anything could actually read your mind we'd all be in big trouble. If such a technology is ever developed in secret life as we know it would be terrible. Thankfully these types of headlines are pretty dumb.

That already exists: http://newscenter.berkeley.edu/2011/09/22/brain-movies/

But that has nothing to do with this, at all.

Offtopic: I've never considered Siri to be intelligent, but more as a bot. Google Now is better but still not what I'd consider remotely intelligent.

I can't wait for the virtual assistant that realizes something will be a waste of my time and flags it as such.

Turning that infographic into reality would mean Google dominance on search would wane.

Or Google could buy it. Or Google researchers could replicate it. Or Google could actually make it work because they already have the infrastructure to scrape, curate and query a humongous knowledge graph.

This title could have been written by buzzfeed. I am interested in information and not hype (unless of course it really does read your mind!).

Welcome to the internet of 2014. Seems like every second article I open fits that description lately.

There appears to be no allusion to mind reading in the actual article title, or in the article itself. The closest is in the photo byline about predicting desire, so the word choice appears to be down to cyphersanctus, presumably for the wooooo factor, unless wired have changed their title.

Half the articles I now see on Facebook have headlines like "Somebody Did Something... and You'll Be AMAZED at What Happened Next"

I now have a policy that prevents me from clicking on any such link, even if it sounds appealing.

Are you familiar with a Chrome extension called rather? It's a great way to block out posts you don't want, like those containing buzzfeed links. The replacement content is so-so in my experience. http://www.getrather.com/

I thought at one point it would be fun to gather all the titles of this nature I could to create a hype/buzz title generator maybe using a markov model or something. No time though.

There's this: http://www.headlinesmasher.com/

The submitted title was "Forget Siri: This Radical New AI Teaches Itself and Reads Your Mind". Did Wired actually post that and then change it? Or was this just a particularly bad violation of the HN guidelines?

Either way, we've reverted the title and tried to take out the remaining linkbait.

Given that ridiculous headline, I refuse to read the article on general principle.

You didn't miss much.

Time flies like an arrow; fruit flies like a banana; AI flies like a human brain.

Reply on news.ycombinator.com