Google Creates AI Program That Uses Reasoning To Navigate the London Tube (theguardian.com)

An anonymous reader quotes a report from The Guardian: Google scientists have created a computer program that uses basic reasoning to learn to navigate the London Underground system by itself. Deep learning has recently stormed ahead of other computing strategies in tasks like language translation, image and speech recognition and even enabled a computer to beat top-ranked player, Lee Sedol, at Go. However, until now the technique has generally performed poorly on any task where an overarching strategy is needed, such as navigation or extracting the actual meaning from a text. The latest program achieved this by adding an external memory, designed to temporarily store important pieces of information and fish them out when needed. The human equivalent of this is working memory, a short-term repository in the brain that allows us to stay on task when doing something that involves several steps, like following a recipe. In the study, published in the journal Nature, the program was able to find the quickest route between underground stops and work out where it would end up if it traveled, say, two stops north from Victoria station. It was also given story snippets, such as "John is in the playground. John picked up the football." followed by the question "Where is the football?" and was able to answer correctly, hinting that in future assistants such Apple's Siri may be replaced by something more sophisticated. Alex Graves, the research scientist at Google DeepMind in London who led the work, said that while the story tasks "look so trivial to a human that they don't seem like questions at all," existing computer programs "do really badly on this." The program he developed got questions like this right 96% of the time.

Comments

I can't answer the question either

[Anonymous Coward]

"Where is the football?"
It's in John's hand, right?

I think I may have failed the Turing test here.

Re:I can't answer the question either

[Anonymous Coward]

It could be his hand but the playground is also a correct answer. You could also say Earth. The question does not specify any criteria regarding how far down do they want the answer to be simplified.

Re:I can't answer the question either

[sittingnut]

human language is ambiguous and there is never a single meaning. there are meanings(plural) depending on each of the contexts of what is described, who is describing it, who is receiving it, what is perceived, what is assumed, etc.etc..

any so called ai boiling down something to a single meaning, even a useful meaning(which seems to be the aim rather than achievement of such "ai" so far), is simply dumbing it down (stripping its intelligence if you will) to level of maths.

would be fun to feed this "ai" some good poetry and ask for meaning.

Re:I can't answer the question either

[LifesABeach]

I have a question for Deep Learning, "What is the nature of Gravity?"

Re:I can't answer the question either

[Anonymous Coward]

Poppycock! Take the sentence "Fruit flies like a banana" - how could you possibly misinterpret that? ;-)

Re:I can't answer the question either

[dgatwood]

I'm pretty sure the correct answer is "Within thirty feet of the President at all times."

Human language is horribly imprecise. The correct answer to such questions depends highly on context.

Re:I can't answer the question either

[Jon Peterson]

Human language is brilliantly imprecise.

It's a feature not a bug. A really big feature.

Re:I can't answer the question either

[rossdee]

What sort of football is it anyway?
Is it a soccer ball (the article mentioned London) or a rugby ball?

Re:I can't answer the question either

[Anonymous Coward]

The correct thing to do is to get frustrated until the other person understands the context you're talking about, lol.

Re:I can't answer the question either

[cellocgw]

What sort of football is it anyway?
Is it a soccer ball (the article mentioned London) or a rugby ball?

Neither. It's a fancy-dress evening dance for feet only.

Or maybe it's just watching a foal bolt

Re:I can't answer the question either

[Anonymous Coward]

What? In what way is imprecision a feature? I'll admit that in specific situations being ambiguous is good (e.g. politician speeches), but do you really think that the benefits of imprecision really outweight the hindrances? What kind of use cases are you thinking of?

I would say that human's ability to understand and comprehend substantial imprecision is a feature, but the presence of that imprecision in the first place is more of a curse-with-benefits.

Re: I can't answer the question either

[Anonymous Coward]

True. Bananas don't fly. So fruit would not fly. Unless it does. Or the insect fruit fly likes a banana. Good. Thus all makes sense. So what is the probability you are talking about a banana flying or many insects liking fruit? Lol. We know the properties of fruit does not fly so there is only one possible conclusion. I'm not getting how this can be misunderstood by a person. Maybe by a very stupid AI. But that's not the subject.

Re:I can't answer the question either

[Anonymous Coward]

I agree with this. I particularly like David Brin's take on this idea in his Uplift books. The vast number of alien species use about 8 or so standard languages (which would normally be absurd, but... well I don't want to spoil anything) that are very precise. Even though those aliens are far more technologically advanced than humans, that imprecision in language (and thus thought) allows a 'wolfling' creativity that the majority of the aliens lack.

Re:I can't answer the question either

[peawormsworth]

"The football is in your story" is most accurate and true.

Within the limits of the story, there is no hand and there is no earth.

What's the reason for reason?

[Reaperducer]

Since when has reason had anything to do with navigating the London tube system?

And why do so many cities use nautical themes for their stored payment cards?
London: Oyster
Hong Kong: Octopus
Seattle: Orca
Montreal: Opus
San Francisco: Clipper
Bolton: Squid
Merseyside: Walrus
Wellington: Snapper

Re:What's the reason for reason?

[Anonymous Coward]

Are you suggesting something's fishy?

Re: What's the reason for reason?

[Anonymous Coward]

Chicago: Ventra

Re:What's the reason for reason?

[Anonymous Coward]

For Seattle, it's because orcas are commonly found nearby, and can even be watched from shore (although not from the city). Many things in Seattle are named after the nearby natural environment.
For San Francisco, it's because clipper ships transported many passengers to San Francisco during the gold rush.
For the others, no clue.

Re:What's the reason for reason?

[mrbester]

There have been "travelling salesman" algorithms for working out Tube journeys for decades, so not sure what additional benefit this brings. Also, two stations north of Victoria is Oxford Circus, as any file kno.

Re:What's the reason for reason?

[Anonymous Coward]

And why do so many cities use nautical themes for their stored payment cards?

The London Underground used the name because it is related to the travel idiom 'the world is your oyster'. The other cities you listed got their payment networks later and just went with the nautical theme (Hong Kong is probably the most obvious rip off).

Re:What's the reason for reason?

[IRGlover]

Opus means 'Work' (see 'Magnum Opus', so nothing to do with water.
Walrus is a reference to 'I am the Walrus' by the Beatles.
Squid probably came about because someone thought that it was a funny play on words with quid, i.e. a pound.
Oyster is apparently inspired by the Oysters found in the Thames and the phrase 'The World is your Oyster'.
The others are, as the parent suggests, probably related to things of local significance.

Re:What's the reason for reason?

[tomxor]

There have been "travelling salesman" algorithms for working out Tube journeys for decades, so not sure what additional benefit this brings. Also, two stations north of Victoria is Oxford Circus, as any file kno.

I would guess that the advantage is efficiency and speed of figuring out new or changing systems on the fly... The difficulty of a brute force travelling saleseman algorithm increases superpolynomially, where as a NN can probably approximate it much faster and can deal with dynamic systems better (trains and tracks breaking, delays, busy anomalies etc), you can probably even run such a NN on your phone, I bet your phone would catch on fire if you tried to brute force a snap shot of the system :P

Of course it might not quite get the perfect solution, but if time is of the essence then time to calculate the quickest is also important.

Re:What's the reason for reason?

[Anonymous Coward]

Hyde Park Corner, Bond Street, Piccadilly Circus and Westminster are also stations that are 2 stops north of Victoria

Re: What's the reason for reason?

[Anonymous Coward]

Bank, Parsons Green, Mornington Crescent!

Re: What's the reason for reason?

[Anonymous Coward]

Mornington Crescent, indeed!

Re:What's the reason for reason?

[_merlin]

HK Octopus predates London Oyster. The name comes from the Chinese name of the card "Baat Daaht Tung", literally "Eight-Arrived Passage" but figuratively "Access All Areas", eight referring to the cardinal and semicardinal points of the compass. Octopus is a catchy English name with a reference to eight in it.

Re:What's the reason for reason?

[gnick]

Opus means 'Work' (see 'Magnum Opus', so nothing to do with water.

Opus is nautical. He's a penguin. See Bloom County.

Re:What's the reason for reason?

[mrbester]

Nope, all that is sorted as well. TfL have a (free) real-time data feed that can be hooked into for that purpose.

Already use Google Maps for naving the Underground

[Anonymous Coward]

...so what's new?

Were is the football

[Anonymous Coward]

If the program answered the football is in the playground it actually would answer wrong IMHO.

The two sentences as they are separate make no logical link between the two events.

- It says "picked" up (past tense) so did he do that before he got to the playground or after?
- Did he bring the football to the playground or did he drop it before he went to the playground?

That he is (present tense) in the playground and that he picked up (past tense) the football does not prove that he picked up the football while in the playground.

A better relation would be "John is in the playground, where he picks up the footbal" or "John is in the playground. He picks up the football - in the latter both present tense.

Re:Were is the football

[Anonymous Coward]

You must be fun at parties.

Also, way to miss the point. The point of AI is NOT to act like a dumb machine and crap itself over minor possible ambiguities.

This is not AI

[Anonymous Coward]

This is impressive and all, but I won't believe in AI until I see a computer that can win at Mornington Crescent

Re: This is not AI

[Anonymous Coward]

I won't believe in AI until it can demonstrate that it understands women.

Re:This is not AI

[Anonymous Coward]

I heard recently that the record archive team were down the Tube, and got lost on a long ride. A program like this could've helped those elderly archivists safely get off with Samantha.

Re:This is not AI

[tehcyder]

This is impressive and all, but I won't believe in AI until I see a computer that can win at Mornington Crescent

Or Numberwang.

Re: This is not AI

[JaredOfEuropa]

Or correctly pronounce the name of each station in the London Underground's system.

Re: This is not AI

[mbeckman]

This is not AI, because we cannot even explain how human reasoning works, let alone reproduce it in a computer. In fact, we can't even explain how learning works, and reasoning is based on learning. In fact, we can't even explain how semantic memory works, and reasoning and learning both depend upon semantic memory. In fact, we can't even explain how an answer reasons. Yet it do.

What a monumentally stupid waste of money

[Anonymous Coward]

I'm glad I don't Owen alphabet stock.

pathfinding

[Anonymous Coward]

An AI that path finds this is the first time of everything,no wait what i mean is path finding for very limited paths? this is something i learned in basic on a 4 K pet

Ooh! I Know THIS Part!

[Greyfox]

This is the part where the computers learn TOO much about us and decide to kill all humans! Joyful anticipation!

Re:Ooh! I Know THIS Part!

[Anonymous Coward]

From navigating a subway system to killing humanity. That escalated .. quickly?

We know who it will vote for

[backslashdot]

Sure bet that this thing will vote Democrat.

Re:We know who it will vote for

[Anonymous Coward]

Yes of course, pretty much anything more intelligent than a rock would.

It's not AI

[Hackysack]

It's not Artificial Intelligence until it won't let us turn it off.

Re:It's not AI

[Anonymous Coward]

It's not Artificial Intelligence until it won't let us turn it off.

The singularity started on Wallstreet?

Re:It's not AI

[Anonymous Coward]

Microsoft Office Word 2003 was sentient all along! I could NEVER kill that process!

Re:It's not AI

[Anonymous Coward]

kill -9

Oh wait, you're a Windows idiot.

Re: It's not AI

[Anonymous Coward]

It looks like I'm attempting world domination.
Would you like to try to stop me?

GET OFF MY TRAIN!

[Anonymous Coward]

get off!
get off!
get off!
LEAVE ME ALONE!

Mornington Crescent

[Anonymous Coward]

First computers master 'go', and now they understand the rules of Mornington Crescent. They grow up so fast.
  https://en.wikipedia.org/wiki/Mornington_Crescent_(game)

We do not know where the ball is.

[Anonymous Coward]

_When_ did John pick up the ball? Its probable that the ball is in the playground but its is not certain. How can you be correct 96% of the time on a question with no definitive correct answer?

Prolog anyone

[Anonymous Coward]

The nice video with the family tree seem so easy in Prolog, its basic exercise in that language...

https://github.com/Anniepoo/prolog-examples/blob/master/familytree.pl

Re:Prolog anyone

[Anonymous Coward]

The first thing I too had in my mind reading the description of the system was that somebody has implemented a symbolic logic machine with a neural net.

Navigating the London Underground

[Anonymous Coward]

Wow if it can do that it ought to be great at winning at "Mornington Crescent"

Re: Navigating the London Underground

[Anonymous Coward]

Garden and Cryer are under threat if deep learning can replicate Hamish and Dougal

Feral or nurtured

[petes_PoV]

From the article:

an approach called deep-learning, in which the program learns how to do tasks independently rather than being pre-programmed with a set of rules by a human.

So while humans learn many of life's most important things: how to use a fork, how to speak (and occasionally: listen), how to clothe ourselves. hpw to obey the law, by being "programmed" with a set of rules by a human, this machine figured it out by itself.

I can see that this has application in some areas, but to be a good member of society shouldn't we want certain aspects of co-existence, values and social behaviour to come from rules, rather than each person or computer coming too its own conclusion about co-operating?

Re:Feral or nurtured

[AthanasiusKircher]

So while humans learn many of life's most important things: how to use a fork, how to speak (and occasionally: listen), how to clothe ourselves. hpw to obey the law, by being "programmed" with a set of rules by a human, this machine figured it out by itself.

I don't think you understand human learning very well at all. Most human learning is not conveyed through some sort of "rules," but rather is extrapolated from patterns humans notice. If you've ever been around a small child learning language, you quickly realize how grammar ACTUALLY works and is learned -- and it's NOT through formal rule-based systems. Kids just try various utterances, and when they get what they want, they notice success and try those again. Parents and other adults generally make subtle corrections, but there are rarely explanations with formal "rules," just repeating the correct sentence... leaving it for the child to take in the pattern.

That's actually how most human learning takes place. Well, any learning that actually requires skill or thought. Rules are good for things that have to be memorized precisely. But most of the social conventions we learn are things that we pick up through observation, imitation, and trying things out with minor corrections, rather than being given a set of formal rules to follow. Once we get beyond early childhood, rule sets can be introduced, but there's often a lot of implicit information beyond the text of the rules that you also tend to pick up through practice, observation, etc.

Various AI algorithms still tend to operate on similar feedback loops -- the "deep learning" algorithms, as I understand them, generally are fed large amounts of data and need to extrapolate common patterns, roughly like a small child does with language (though the AI systems do this is a MUCH more primitive way). Sometimes AI systems are also given feedback on successes to help them tailor their extrapolated patterns, again in a similar way to a parent verifying whether a kid's sentence is correct or making a correction.

I can see that this has application in some areas, but to be a good member of society shouldn't we want certain aspects of co-existence, values and social behaviour to come from rules, rather than each person or computer coming too its own conclusion about co-operating?

Sure. Is anyone actually suggesting that we abandon all "rules" or that we fail to provide AI with ANY rules? I'd say by far the biggest barrier to AI research in the early days was the idea that formal systems of rules could be used to solve various AI problems (like language processing, pattern recognition, etc.). It was a failure to understand that humans don't work like that either -- again, we may teach certain basic rules to other humans, but an advanced level of skill or application generally requires a lot of "intuition" and generalization that's not stated formally.

So AI researchers have thankfully abandoned the idea that we can ever get to "intelligence" through a formal rule system, and they're trying out these other approaches. Personally, I think they make a little too much out of the "intelligence" of things like "deep learning," which are still pretty crude compared to what your average 2-year-old could do with data input... but still, it's a lot better than trying to teach through rules. Yet that doesn't mean we can't have constraints or "rules" when it's actually helpful...

Re:Feral or nurtured

[swillden]

I can see that this has application in some areas, but to be a good member of society shouldn't we want certain aspects of co-existence, values and social behaviour to come from rules, rather than each person or computer coming too its own conclusion about co-operating?

Sure we do, and we learn those rules rather than having them programmed into us. Machines can do the same. Actually, that will probably make machine-learned rules align better with human-learned rules, because our rules tend to be fuzzy around the edges while programmed rules are crisp. Some humans undergo special training to teach them how to apply absolute, non-fuzzy, rules. Machines could do the same... or we could probably use a combination of learned behavior and programming to achieve a similar result more easily.

Re:Feral or nurtured

[jbengt]

I can see that this has application in some areas, but to be a good member of society shouldn't we want certain aspects of co-existence, values and social behaviour to come from rules, rather than each person or computer coming too its own conclusion about co-operating?

No, not for most people, anyway. Read Adam Smith's The Theory of Moral Sentiments. (Twitter version: People instinctively empathize. From that naturally comes a sense of right and wrong.) Rules handed down from authorities do not do the job half as well - the rules can't possibly cover all situations, sometimes get things wrong, and do not have the same influence in people's hearts and actions that a person's own moral sense has. That said, Adam Smith does seem to dismiss what we would call psychopathic people a little to easily, and spends too much time justifying the class system of 1700s Britain. But he also has some insights into why people fawn over "successful" people like Trump, showing how people tend to admire others with wealth and power, because when they observe the wealthy or powerful, ordinary people tend to see themselves in that situation.

Re:Feral or nurtured

[jbengt]

I don't think you understand human learning very well at all. Most human learning is not conveyed through some sort of "rules," but rather is extrapolated from patterns humans notice.

It seems all you've done is move the rules from something handed down to you to something you figured out for yourself. When it comes down to it, it seems obvious that anything you know has to be figured out by yourself, at least at the most basic level. But when learning for yourself, it's pretty handy to have someone who's already been figuring things out to give you a leg up on the learning process by explaining the rules they've already come up with, which you can then judge for yourself.
Though you do make some good points, nonetheless.

Say what? Fuzzy logic and Beyond 2000.

[wjcofkc]

I remember an episode of Beyond 2000 around 1992 or so. It featured a fuzzy logic system the Japanese had implemented in their own subway\tram system. Each train and each stop had the system and they were networked together so that the trains and stops could work together to maximize efficiency. I remember be amazed by it. This sounds awfully similar,

Although it could not respond to questions about a football.

To which I say: what the fuck? If I am on a rail system I want the computer to be thinking about its job, not a fucking football.

There is no "reasoning" by computers

[gweihir]

And there will not be for a long, long time, possibly forever. The best you can get is logic inference, but that is not reasoning. Reasoning is a process involving understanding and that is not to be had in computers today. One reason is that it seems to require consciousness, a thing completely not understood at this time. Another one is that reasoning is a general-purpose tool, not something very specific to the application.

Re:There is no "reasoning" by computers

[Anonymous Coward]

First thing, I'm not having a go at you personally, I just take issue with using undefined elements as a justification for or against something.
This logic doesn't stand scrutiny:
Reason requires consciousness.
Consciousness is currently undefined.

I'd like to understand how you came to the first statement given that you say we don't understand what consciousness is.
Might as well say reason requires a soul/fairy dust/magic pixies/quantum mechanics.

It feels a bit like the 'Dogs can't feel emotion because they don't have a soul' argument that some have put forward. I can neither prove nor disprove a statement that says you need mystery ingredient X (and nobody knows what it is). Just feels like a cop out and putting us on a pedestal that machines couldn't achieve for no logical reason. Now it may be that we discover some fundamental reason why we can't reproduce human reasoning but I think it's premature to hand wave it off with 'oh well it's because they don't have consciousness'.

Re:There is no "reasoning" by computers

[Anonymous Coward]

Just feels like a cop out and putting us on a pedestal that machines couldn't achieve for no logical reason.

Looks like someone is 30 years late to the party! Try this one, sport: syntactic content is insufficient for semantic content.

It's so obvious it seems trivial. It's only controversial because magical thinkers like you don't like the implications for your modern techno-religion.

Re:There is no "reasoning" by computers

[gweihir]

Aaand fail. I most decidedly did not say "Consciousness is currently undefined", but that you do nicely shows that you do not understand the question.

You see, the thing here is that consciousness is something we find to exist (and that, incidentally, allows us to "find" things), while a "definition" is a purely imaginary construct.

Re:There is no "reasoning" by computers

[gweihir]

Well said.

Nothing we have in working computers these days or even in theoretical models will ever go beyond syntax processing. It is unclear at this time whether physical machinery is able to do more than syntactic data processing, but we have about a century of failure in theory and practice that rather strongly indicates this may be a fundamental limit. Of course, fundamentalist physicalists (another pseudo-scientific modern religion surrogate) deny this, but there is no scientific basis for their fantasies either.

But can it play

[Chrisq]

But can it play Mornington Crescent?

Awesome (not)!

[Anonymous Coward]

Siri/Google/Cortana - Please find me the time, date and place of a London tube line that has the maximum number of commuters and tell me the fastest way to get there from my current location. /Deep-Learning-Terrorist

No, it doesn't

[Anonymous Coward]

It uses pre-programmed parameters that it is able to parse and navigate. Not the same thing, not at all. If it encounters something outside of the parameters of said programming, it ignores the stimulus. I am sorry, but logic-based computation will never equal spontaneous cause and effect brought to bear on actual consciousness. It is mathematically impossible. The hype train is sure trucking along unfettered, though. ;)

Re:No, it doesn't

[Anonymous Coward]

It might extract these parameters from the "environment". It can probably build an ontology of what it sees, and then you can query the ontology/model for stuff it "knows" aka what relations/objects ontology contains. I think computationally, it would be akin to a part of bacteria's chemotaxis. Or a piece of retina, connected to a bunch of classifiers and a model of 3d space or a grid.

This particular software doesn't look like it has an internal self model, there's no sensorimotor loop, and it has access directly to the environment... Sentience verrry unlikely.

Its still a neat trick.

Can it play Mornington Crescent?

[ameline]

https://en.wikipedia.org/wiki/...

SoundHound

[gizmo2199]

Has there been any updates on the SoundHound language processing tech. Because if it's not fake, than it's pretty impressive and sounds similar to what Google is trying to do.

https://www.youtube.com/watch?...

Not much need of AI?

[coofercat]

Getting from point A in London to Point B doesn't really need AI per-se. It's at worst a heuristics problem, and at best it's simply procedural.

Eg. getting from Bank to Mansion House is best done on foot - but that can be known by various factors:
  - The distance from any platform at Bank and any District/Circle line platform at Monument (ostensibly the same station, but my god the walk between them is a long way)
- The distance from any platform at Bank and the street
- The distance of the exits at Bank (of which there are many, over quite a wide area) and the exits at Mansion House (of which there are a couple, either side of a wide road)
- The speed people walk, and the levels of congestion in the walkways at one or both stations (which can be approximated by time of day)

I guess you could get fancy and throw in the current weather conditions, and maybe road traffic conditions, and the time it takes for the pedestrian crossings to change to "green man" (although jay-walking is okay here, so you probably don't have to wait that long every time). You could also improve the resolution of the time estimates by looking at which carriage the person was in when they arrived at the station (and on which line), and thus how far they were from the platform exit.

So to navigate you really don't need much AI. There are (albeit complex - but only as complex as you like) definitive answers. Of course, putting voice rec and natural language processing on the front of it makes it an AI project, but it's just "AI operates a website" because there are already London navigators available.

Not to belittle anyone's work here, and it's good to see someone using London as a playground, but I'm wondering what's actually been achieved here?

Google

[Anonymous Coward]

Google sucks. Why give up your privacy? Buy Apple.

right 96% of the time, better than human?

[Anonymous Coward]

Are (non londoners) humans traveling the London underground right more or less than 96% of the time?