October 20, 2015

How bad is that artificial intelligence?

In the comments at last night's café post, Freeman Hunt wrote:
Another exhibit in the case that artificial intelligence is not so great:

Drinking in America: Our Secret History
$20.52

NEOPlex 3' x 5' Flag - "A Spliff A Day"
$6.95

Amazon is offering these as things it thinks I might want to buy.
I responded:
Freeman, "Drinking in America: Our Secret History" is a book I blogged about here very recently. "A Spliff A Day" isn't something that ever appeared in the text of this blog but it is in a photograph from Saturday, on a flag. That means that Amazon widget sees what's in writing in a photograph. Pretty impressive.
You may not want to fly a "Spliff A Day" flag. But you were reading a blog with a picture of one. Frankly, I'm amazed that the Amazon widget knows what's in the photographs.

37 comments:

rehajm said...

Yah, Freeman's right in that AI sometimes gets off track in quirky ways. Other times it's creepy accurate. We've been working wth Pinterest on a project and clearly it's algorithms identify subtle color diffrences and textures from photos, as well as content.

Ignorance is Bliss said...

None of the free web platforms ( Blogger, Flicker, Facebook, Youtube, etc ) are charities. If they aren't charging you, they are charging someone else for the opportunity to advertise to you.

The better they can target that advertising, the more money they make. So yes, they are pouring a lot of money into understanding your content, including the pictures.

( I can't wait to see what they make of Laslo's comments. )

tim in vermont said...

I once used the word "mensuration" in a comment on another platform, and got ads for all kinds of incredibly sophisticated micrometers, etc, for several days.

tim in vermont said...

I am sure that mensuration and menstruation as two functions the vagina equals a joke in some universe.

Henry said...

I always find it funny when I order some one-off and am immediately besieged with ads to buy more. You ordered a toaster oven? Here are more toaster ovens!

tim maguire said...

I bought a watch online from Walmart a couple weeks ago. Ever since I've been bombarded by ads for watches. Seems a little backwards to me.

Scott said...

Sometimes machine intelligence makes associations that are funny and odd, and to that extent they are useful in getting you to think about things differently than you might otherwise. But communication produced by machine intelligence is alienating. It reinforces one's feeling of being alone.

Have you ever looked at a book index? There are people who make a living by writing those things. Their task is to group concepts and keywords in ways that make sense to a human reader, so that they can find information quickly. And you can get a sense of the indexer's editorial persona by comparing the indexes of two different books on the same topic -- say, books on American history by Paul Johnson and Howard Zinn.

I've done some indexing on a few projects over the years, and it's a fascinating task. Machine intelligence has never been able to index a book really well. There is some element of humanity that needs to be there. Humans can tell when they are communicating with other humans, even if it's through a list of concepts and keywords associated with page numbers.

tim maguire said...

Henry, I wanted to say something about, "how many watches do they think I need?" but didn't like the wording. Now I see you already cracked that puzzle.

Known Unknown said...

Ever use Siri while you're driving?

She is famously bad at hearing me correctly.

Ignorance is Bliss said...

Scott said...

Humans can tell when they are communicating with other humans, even if it's through a list of concepts and keywords associated with page numbers.

Book indexing as Turing test.

rhhardin said...

AI can't do literary effects.

Its mistakes can appear as literary effects, however.

tim in vermont said...

I once clicked on an Instapundit link to sexy a Halloween costume for some reason, and was treated to pleasant Amazon ads for about ten days.

SeanF said...

A more likely explanation is that the recommendations you get when using the Althouse Amazon portal are partially based on other searches/purchases that have been made through the Althouse Amazon portal.

In other words, Amazon's Artificial Intelligence didn't read the text in that photograph - one of your other readers did (with their Natural Intelligence), and typed it into Amazon manually.

Ann Althouse said...

If SeanF is right, that any reader could do weird searches through my search box and cause weird things to pop up in the sidebar for all readers. I haven't been seeing weird things, so I wonder if that could be the way Amazon is programmed.

By the way, I make money through that search box if you use it and buy something. I think it's great that if I blog about a book, it gets featured in the sidebar. It's been the most effective way for me to monetize the blog. (And I really appreciate your using it.)

tim in vermont said...

@SeanF,

Maybe, but there are lots of volume OCR solutions out there, for example programs that analyze screen shots an employee might have made to determine if it contains SSNs, or other sensitive data, and the like.

pm317 said...

Frankly, I'm amazed that the Amazon widget knows what's in the photographs.

Text recognition in images is perhaps the easiest of content analysis. I had given it as a masters project topic to a student who now works at Google. The commerce of it is catching up with decades of research -- it is big business now for data mining and that awful word personalization which is what you are seeing.

Ann Althouse said...

Obviously, the blog can be read for free, so this blog relates to what Ignorance is saying: "If they aren't charging you, they are charging someone else for the opportunity to advertise to you."

That said, I only write what I'm interested in writing and I never think about what could collect more readers or readers who'd be more attractive to advertisers.

Ignorance is Bliss said...

SeanF-

I'm currently taking a grad school class in artificial intelligence. I assume that artificial intelligence is being used to recognize text in pictures. In this case, more likely by Google than by Amazon.

Ann Althouse said...

Now, Meade could take his dog blog and work it for monetization. It's totally focused on a subject that collects customers for specific advertisers.

SeanF said...

Tim, yes, that's true, but you're talking about Amazon somehow getting a hold of a screenshot from Freeman's web browsing.

If I'm right, it's perfectly legitimate behavior by Amazon - grouping the Althouse Portal searches/purchases together.

If Ann's right, Amazon's doing something incredibly creepy and privacy-invading, and something that - by all rights - the web browser software shouldn't even allow.

I mean, if Freeman accesses her bank account and then goes to Amazon, does Amazon know her account number and balance?

Ignorance is Bliss said...

SeanF-

I believe we are talking about Blogger ( and therefore Google ) analyzing images that the Professor posts, extracting information, and passing it on to Amazon's widget.

Laslo Spatula said...

There is no "Freeman Hunt" tag on this post.

Sometimes I just like to click on the 'Freeman Hunt' tag and wander down Memory Lane.

I am Laslo.

pm317 said...

does Amazon know her account number and balance?

No. Maybe the bank name or website she accessed.

JSD said...

Amazon is not directly tracking you and placing the ads. Each web page has billboard space reserved for advertising. Amazon (and others) are bidding on that space as the page loads on your screen. Companies, like Google Ad Exchange and AppNexus, run instant auctions for the ad placements. These are the companies that identify you when you go on the web and then immediately auction off the ad placements to Amazon, LL Bean & Cabelas as you browse. This only the beginning.

Original Mike said...

"If SeanF is right, that any reader could do weird searches through my search box and cause weird things to pop up in the sidebar for all readers. "

Busted!

Quaestor said...

Not too long ago some AI experts were of the opinion that intelligence is a matter emergent complexity, just get enough neurons talking to each other and intelligence just happens -- the critical fudge being enough. And enough of what? Human brains have 100 billion neurons, give or take, but elephants have almost three times as many. There are no pachyderms punching the clock at FermiLabs to my knowledge. So maybe it's synapses, the interfaces between neurons. We have about a quadrillion of those, whereas the common house cat has 10 trillion. Human brains are 100 times more synaptic than cat brains, whatever that means. Are we 100 times as smart as Garfield? Depends. If my survival hinged on being able to catch mice I might be in trouble compared to many (but by no means all) cats, but if a cat's life depended on solving a quadratic equation or deriving a third-degree partial, then it's bye-bye Pussy. Something tells me synapses are as much the result as the cause of intelligence.

The problem with AI is the A and the I. What's the difference between artificial and natural intelligence? Is one more genuine than the other? And what is intelligence? As of yet nobody has come up with a succinct definition of intelligence useful enough for a processor designer to know when he's getting close. The hot CPUs from Intel have about 1.6 billion transistors on the die, but transistors aren't neurons. They probably are not even analogous. If they were a Xeon chip would be about as smart as a honeybee. The machine I'm using now has two Xeon SMPs. It doesn't make honey, or even syrup. As for synapses -- forget about it. The only part of your CPU that's even vaguely analogous are the GP registers, and there are eight of those per core. If intelligence is a matter of emergent complexity, Intel has a long why to go.

Computer scientists have mostly put AI on the shelf along with other programming tricks. So far AI is merely an amusing but occasionally useful way to index a database. The neat thing is the way the software can parse spoken English into a SQL query, but that's all it is, really -- database queries with an occasional bit of randomness thrown in to keep teenage girls amused and using up data plan airtime. Instead of AI the serious computer geeks talk about MI -- machine intelligence -- genuine intelligence that's suppose to inevitably replace wet intelligence (i.e. brains). MI is basically what AI was before AI became a technology. It's ten years away according to the consensus, and has been for about 65 years, ever since Turing and Von Neumann argued about the Chinese Room.

pm317 said...

So far AI is merely an amusing but occasionally useful way to index a database.

Not to the government folks at DARPA, NSF, and other places. They think there is a magic bullet and fund all kinds of stupid AI research in universities. The AI guy/gal is rolling in money for their worthless projects.

tim in vermont said...

@SeanF

Tim, yes, that's true, but you're talking about Amazon somehow getting a hold of a screenshot from Freeman's web browsing.

No, they only have to analyze this blog once. I will leave it at that as it appears I already got too technical. I was just giving another example of and unrelated commercial product that scan images for text content.

Freeman Hunt said...

That's neat if it reads pictures. I think it is probably drawing the flag link out of aggregate searches performed through the search box though.

The point of that ad-serving program is to serve ads that will lead to my buying something. Amazon has plenty of data on me. I'm constantly served ads for things I have no interest in buying.

It seems that the ad-serving program is serving by population. "People on this site may be more likely to buy these things." That's great, but it requires no intelligence. Match links that have been posted on the site, match searches that have taken place through the site, etc.

tim in vermont said...

I am thinking that MI will show up about the same time practical nuclear fusion does. Ten years from whenever you ask when it is coming. I doubt it can be discovered heuristically.

My bet is that it has to do with group theory, that there is some group of states that can be navigated in a stable fashion without losing its "identity." I don't know why I think this, it is just conjecture, but I think it may be right.

A strangely evocative article, for an article on math on the subject of group theory.

For those of you shy of clicking on links:

https://www.quantamagazine.org/20150312-mathematicians-chase-moonshines-shadow/



Freeman Hunt is one of my favorites too.

Scott said...

If you want to monetize this blog, see if you can get the Washington Post to host it.

David said...

It's not machine intelligence, its MTurk (or some similar service). Amazon and other online companies pay human beings to sit and their computer and look at photos posted by users to (a) make sure its appropriate (i.e., not child porn) and (b) tell the company what it is a picture of. There are lots of different MTurk jobs and Amazon has opened it up as a mediator for companies and workers to find each other. Experienced, efficient MTurk workers can earn about $8/hour.

MTurk is named after the so-called mechanical Turk, a supposed chess playing computer from the 18th century that was actually operated by a little person in a hidden compartment.

n.n said...

It doesn't. Not directly. Its insight arises through correlation. The keystone primitive underlying the digital revolution.

BN said...

"How bad is that artificial intelligence?"

How bad is "natural intelligence?"

a.) Pretty bad
b.) Not too bad
c.) Very bad
d.) Who gives a shit, we're not reproducing anyway.



BN said...

"For those of you shy of clicking on links:"

...those links have already been clicked by about 1x10 to the millionth AIs...

BN said...

"That's great, but it requires no intelligence."

Nothing artificial about that.

BN said...

"The AI guy/gal is rolling in money for their worthless projects."

"Worthless..."

That's cute. You think the purpose is to redistribute stolen (er...) confiscated (er...) taxed money... on worthwhile projects.

It's not. The purpose is to redistribute money. Period.

Objective attained!

Next...