Precisely. The image recognition and natural language processing that's "just a deep learning algorithm" today would have blown the socks off anyone ten years ago. Chess playing programs were AI until they could decisively beat all humans, and then we decided that chess doesn't require intelligence. Self-driving cars were AI until they got close to market and now apparently driving a car doesn't require intelligence. It would seem that only True Scotsmen are intelligent.

Many things would have blown the socks off people of a previous time. That doesn't make them all AI, especially since many of the results we see today are not due to novel techniques a ten-years-ago AI researcher wouldn't have known about, but rather to ten years' worth of advancement in refining the techniques and in the availability of computing power to use those techniques with. A lot of it is just "yeah, if we could've got X to run that fast or had that many processors to throw at it, we'd have had that back then".

So if you want to describe Google as a (insert thing here) company, really the thing to insert in the parentheses would be something like "massive computing infrastructure".

And for better or worse, "AI" is still associated in many people's minds with things that the Google search box isn't close to resembling. People expect an AI to hold a conversation, or be able to reach (and explain) novel conclusions, rather than heuristically guess the next word you'll type (given the computing power, not terribly hard given the informational entropy of human languages) or scan a large database quickly to return the result you asked for. People mostly don't see fast brute-force searching of a problem space, for example, as "intelligence", but that's most of the secret sauce behind game-playing "AI".

Even within the last six months I've seen very confident posts here to the effect of "an AI that is worth a damn at go is still at least 10 years off".

And yet Google's Go-playing approach still doesn't seem to involve fundamentally novel techniques. Most of the advance seems to come in refinements to existing techniques, and being able to throw resources at the problem on a scale nobody had managed before.

And here is the exact phenomenon we're talking about, in action!

Back when neural nets were "fundamentally novel" we didn't have the understanding or computing power to tap their potential. So they "weren't really AI" because they couldn't do much that was interesting.

Now, we have a better understanding of how to use them, and we have enough computing grunt to make them do really interesting things. But they're "not really AI" because they're "not fundamentally novel."

Well, if what we had ten years ago wasn't AI, then what's the argument for "same thing, but running a little faster on better hardware" being AI?

Just like James Watt's steam engine was just a refinement of Thomas Newcomen's model and yet fundamentally changed the industrial landscape in Europe.

It's not the same thing previously rules were being fed into the computers now data is being fed into the computers and it is understanding or getting the rules by itself and then improving itself based on what it understands. AI wont be created, A day will just come when the algorithms that learn and improve on its own will reach sentience. I expect Google to reach it first simply becuase of the amount data and processing power in the Google servers.

> And yet Google's Go-playing approach still doesn't seem to involve fundamentally novel techniques.

So? The fact that it is worth a damn at playing go was unexpected to most people afaict. What do you mean "fundamentally novel", are you expecting the field to advance from cars that barely avoid accidents, to Ava in one fell swoop?

Frankly I would prefer that advancement in AI not occur in sudden unexpected bursts of progress, thanks.

>The image recognition and natural language processing that's "just a deep learning algorithm" today would have blown the socks off anyone ten years ago

I think the concepts behind this were known 10 years ago - they improved and made it scale/we got the hardware and productized it - but would it really "blow your socks off" ? Hell I'm kind of disappointed at the development speed, I was expecting real personal assistants (not something trivial like Siri) by now.

10 years ago general purpose image comprehension by computers was nonexistent and Google image search for $word basically just returned images with $word in or mentioned near the URL.

Nowadays you can search for "young einstein" and get 3 pictures of Yahoo Serious and 10 pictures of Albert Einstein as a young man. It still looks like magic to me even though I know a bit about how it works.

Yes, where is this blow your socks off natural language processing? Certainly nothing I've ever used has contained it. The best I see is shitty voice recognition which can't understand me 75% of the time and often terrible attempts to chop questions out into search queries.

Don't forget that Facebook is going to be a big player in this game as well- especially facial recognition, where everybody is helping them by tagging people in their photos.

If we characterize companies as strongly as "Apple, a consumer product company" and "Google an AI company", then Facebook is a worldwide identity tracking agency. They could pivot and be in charge of delivering birth certificates, DNA groups, ethnicity mapping, insurance-bank-police profiling... In fact I expect it to play a very major role in wars to determine targets.

you may want to read the book "The Circle" by Dave Eggers. It is about this exact concept, and is quite good (I'm actually only nearly finished with the book, so maybe the ending will suck and i won't recommend it anymore, great so far though!)

The problem with calling anything AI is that it's such a meaningless word in the context of computers. According to Wikipedia AI is "intelligence exhibited by machines or software" and Merriam Webster define intelligence as "the ability to learn" so I'd say we hit AI long ago. I'd follow the meaning further but I hate arguing definitions.

To put it another way, if we define a Scotsman as anyone born in Scotland and/or related to someone born in Scotland, we've essentially just defined a Scotsman as every organism ever to be alive on Earth.

I have to imagine when the term first started to be used people imagined an AI as something more human than neural networks. Imagine HAL or any other hollywood perceived robot.

I take the term AI to mean "a thing that people will actually acknowledge as being AI when it exists"—something that can only be defined retrospectively, but which still is a workable definition in that you can sort of predict what people will or won't want to have "wasted" the eventual definition of the term on.

My personal bet is on people placing that line squarely on the divide between "Stoic Guru" and "Active Academy" UX models[1]. An AI is probably, to most people, an agent that continues to think and learn when it is not being interacted with—by surfing the internet, maybe—such that it can generate novel outputs, and revise its own previous beliefs, with no user-visible step that would be perceived as "re-training." Something that will come to you with thoughts it has had that you'll find relevant, that were inspired by new data the agent has acquired not derived from your input.

A simple example that people would probably think of as "AI" (and which people wouldn't accept as a UX paradigm unless it was presented to them through an agent-based interface) would be a spam filter like GMail's that learns from the classifications of everyone who uses it, that would go back and reclassify messages that it now had a more refined opinion on. Nobody would want a dumb algorithm to take emails that "were" in their inbox and move them over into their spam folder—that's destructive!—but they'd accept a (virtual) secretary using their judgement to do so. Thus, AI.

[1] https://scifiinterfaces.wordpress.com/category/active-academ...

I think you just described Google Now: https://www.google.com.au/landing/now/

A while ago AI hadn't been complex and deep enough to warrant categories (especially to the layman), so it was just a general moniker put on everything from a smart piece of code with some adaptability all the way through to singularity level computer intelligence (as we'd imagine it). Now we're beginning to realise that there are vast ranges of categories out there, but it will take some time for consensus use to form (especially since a lot of them overlap into what we perceive as general intelligence, which we so far have had difficulty in defining and pinning down)

The popular understanding of "AI" should really be rebranded as "Synthetic Thought"

It's more like, we know what 'intelligent' is, but it's hard to quantify, so we try to find problems that we think are at least as hard as 'intelligence'. when we solve those problems and the solutions don't resemble 'intelligence', we move on.

it may be the case that 'intelligent' is just a word used to make humans feel special things about mammalian brains, but it's also consistent that we overestimate the difficulty of the abilities that add up to 'intelligent'. It's the difference between moving the goal posts and having our perspective changed so that we realize that the goal is further away than we thought.

"The game of chess, is like a swordfight. You must think first, before you move." -Wu-Tang Clan

:)

A computer running a chess program isn't intelligent because it isn't sentient. It is just doing what it's been told to.

Yeah, and sentience is a completely made up human thing by which we differentiate betweem ourselves and other non-human animals.

What people want to see is artificial general intelligence, currently that is not possible.

Does running human software on biological hardware make it sentient, or would the same be true on other forms of hardware?

That hardware is just doing what it's been told to.

Just wait, after machines become unarguably sentient, the only thing for them to fall back on will be "yeah but it doesn't have a soul". :P

The human mind isn't software so you're missing the point.

Actually, the computational theory of mind is well regarded in modern cognitive science: https://en.m.wikipedia.org/wiki/Computational_theory_of_mind

Very interesting. Thanks.

This concept seems to be a direct abstraction of hindsight bias.

In fact, the concept of hindsight bias feels very much like every aspect of life, also. As soon as you build the next great tower, you opt for a new wave of bigger, more powerful enemies (i.e., you level up), and so the solution for the last wave no longer feels that interesting anymore, and thus seems, perhaps, obvious.

Related: http://www.theatlantic.com/magazine/archive/2013/11/the-man-...