Phil Bradley's weblog

More thoughts on Google Images

If you've not read my first blog post on the Michelle Obama image fiasco you might want to start there first. I should also point out that I'm going to be referring to some words that people might find offensive in this post - I've used asterisks and don't really think anyone will be upset, but it's your personal call.

Google's concept of images and SafeSearch is fundamentally flawed. Let's take a quick look at how this is supposed to work: "Use Google's SafeSearch filter if you don't want to see sites that contain pornography, explicit sexual content, profanity, and other types of hate content in your Google search results." Note please that we're talking about pornography AND hate content. One could assume from this that if you turn ON the SafeSearch filter you won't see hate content, yet clearly - as demonstrated in my previous post - that's exactly what you get. Yet ironically, if you turn OFF the SafeSearch filter you don't get to see the image. So things work exactly the opposite way around to that which any sane person would expect.

Let's look at how this works with other terms - I ran a couple of searches for sexual terms and another two for racial slurs, because I wanted to see how Google Images SafeSearch filter worked, and the results were very interesting. A search for f*ck gave 20 images with SafeSearch turned off, and 0 with it turned on. The same result was found for c*nt as well. This is exactly what I'd expect to happen. However, when I ran searches for n*gger and w*g, the results were very different, with an overlap of 14/20 and 12/20 terms. Clearly these racist terms are not regarded in the same way as the sexual terms; they appear to be less 'important' if you will. Rather than blanket coverage of 0 results, Google is quite happy to display images for racial slurs, but not for sexual content. Remember the 'hate content' element of SafeSearch? Please forgive my hollow laugh.

Google's search algorithm is also broken. If I want to do a search for Michelle Obama, I want to see images of her, yet the first or second image coming up is not of her at all, it's of an image that may once have been OF her, but isn't any longer. If I do an image search for 'bus' I'm not going to be impressed if I then get given a result for a rocket. I would expect Google to tweak the algorithm to fix that; I wouldn't expect to put up with it. Google should actually come clean and say that the algorithm hasn't worked, and pull the image for no other reason that search accuracy.

However, Google isn't doing that. In their own advert that they've taken out for this search they say "Google views the integrity of our search results as an extremely important priority." Clearly however they DON'T do this, because the integrity of their results here has been blown out of the water, with this wholly inappropriate and inaccurate result being returned.

They then go onto say "Accordingly, we do not remove a page from our search results simply because its content is unpopular or because we receive complaints concerning it." In other words, they are hiding behind their algorithms and saying that it's nothing to do with them. Well I'm very much afraid to say that it is because they wrote those algorithms in the first place! They chose how they were going to work, and they sat and tweaked them artifically until they got the overall result that they wanted. So the results are artifically affected by Google employees anyway! They can also make the point, which they do, that lots of people contact them asking for images to be removed, and they only do this if a law has been broken. It would be easy to deal with this - once 'x' number of people complain about an image, simply move it automatically into another category. The 'x' could be quite high to ensure that a small pressure group couldn't easily manage it, and they could always look manually at the image if necessary. You simply cannot tell me that Google hasn't got the resources available to do this.

We're left with one stark and rather nasty conclusion. Google is happy to block access to unpleasant sexual imagery, but they're not prepared to do so for unpleasant racial imagery. That should make everyone a little uncomfortable.

Google's ineptitude with Obama image, and it's not what you think

OK, there's a big fuss going on at the moment with Google images and a nasty little badly photoshopped racist image of Michelle Obama. If you run a search in Google images for her name (without quotes) you get this image popping up in first place. Google apparently removed it, then decided to put it back and take out an advert talking about it instead. Naturally enough, I was interested in seeing what all of the fuss was about. I did my search, and this is what I got:

No, I can't see the image either. 15 pages scrolled through later and I still can't see anything. Maybe it's because I'm in the UK? I tried both .com and .co.uk with the same result. I tried logging out, in case that was causing the problem. Same result. I tried IE instead of Firefox. Same result.

I'm starting to wonder at this point - I asked a friend to run the search, and she could see it. I then ran the search again in Chrome, and at last I saw the image. My first thought was 'WTF?' and then looked a little more closely - what was different between Chrome and IE and FF? Then I realised... in both of those browsers I have SafeSearch:Off. In Chrome it was on. So back I go to Google and try again, this time with SafeSearch on moderate. This time I get:

And once again, with SafeSearch:Strict I get this:

(I blurred the images myself by the way - if you're desperate enough to see it for yourself, you know what to do.)

Now there are a few things that interest me here. Firstly, I get to see racist images if I HAVE moderation on, and I don't get to see it if I DON'T have moderation on. Isn't that... y'know, the wrong way around? Secondly, it's the same image in both cases, but from two entirely different sites - Prisonplant in one case and Ohot-girls in the other. 

Google's coming under a lot of flak on this one - should they allow the image or not? Whatever they do, someone is going to be very upset, but for me, that's not really the question. My question is more along the lines of why are racist images acceptable in some moderated versions and not others - specifically, not in the 'adult/unmoderated' version, but in the strict moderated version? And wouldn't it make a bunch more sense to limit the image to the unmoderated version?

My view on the whole censoring thing - for what it's worth. Google has algorithms that they have put in place, so they're already arranging results in one way or another. They decide on what is offensive and what isn't - naked bodies and graphic sexual images are potentially offensive, which is why they're only available if you turn Safesearch off. However, once you start searching for other types of images, Google's approach seems slightly different - a search for 'the N word' brings up substantially the same images however you have Safesearch set. So images of naked people can be offensive, but racist images are not offensive, according to their algorithms is that it? This is a very slippery slope. Google is very keen to make sure that they don't do things by hand, and they let the algorithms take control. But, and it's a big but, who WRITES the algorithms in the first place? Yup, Google employees. It is disingenius in the extreme to blame your code, and use that as a get out clause when you write the code in the first instance. Google already fiddles around with their code - that's how they have got rid of certain well known Googlebombs - they've played around with code until it gets the result that they want. That's a very thin line from there to dealing with specific images.

Of course, they could use the defense that if an image isn't against the law, who are they to do anything with it? I'd certainly agree on that point - if an image is legal, then should they? Well, they're doing that anyway with hard core porn images, and using the Safesearch option to protect minors. Of course, we then get into the 'which laws are we referring to' which is a whole other can of worms.

In the final analysis, in my opinion, Google has got it badly wrong this time. That image should have been caught and slipped into the 'only view without safesearch' option. To do otherwise is just plain wrong - not just because of what the image is, but because they're not consistent. And if they're not consistent, how can they be trusted with the results they give us. I'm probably not the best person to ask, because my opinion has always been 'Don't trust Google', but there you go.

Google Chrome OS

Image by FreeCliDent via Flickr

You've probably seen mention of Google Chrome as a browser - it's now becoming an OS or operating system. The video below explains it in more detail, but if you want to get a quick handle on it - Google ChromeOS turns your computer into a dumb terminal. That's not necessarily a bad thing, but that's essentially what's happening.

Limiting Google results to seconds!

Let's face it - the 'Recent' option on Google results is great, but if you're looking for really recent material, you don't have much control over it. At least, until now. A nice post over at SearchEngineLand explains how you can limit your results to data that Google has added as recently as the last second or so. This is how you do it:

Run your search, and then add &tbs=qdr:X##&tbo=1 right at the end of the URL, with no spaces - let it run straight on. BUT before you try the search, you need to change the X## element, so if you want a search for material added in the last 5 seconds, change that to s5 or n5 for the number of minutes of h5 for 5 hours.

Google Image Swirl

Google is continuing to explore the possibilities of image search. They've recently done work on images and Creative Commons, Google Similar Images,  and Google image size search. They're now playing with something they call the Image Swirl, and you can play along by going to the Image Swirl lab page. Now, don't get too excited, because you can't just play around with any old term - but there are about 200,000 terms that you can use. I tried my usual 'Hubble Telescope' search, and this is what I was presented with:

Nice, but not exciting at this stage. However, you can see that behind each image is a stack of others. What Google is trying to do is to sort of types of images, so we have collections of images of the telescope, but also various different types of images that it's taken. Click on a stack, and Google Swirl brings it centre stage thus:

This should actually look a little bit familiar as it's akin to the Google Wonder Wheel that they've had available for some time now.  I can continue to click on images and Google will continue to present them to me, while keeping my history available, and this is where we end up with the 'swirl' concept:

I can of course simply click on an image at any point to go directly to it - a mouse cursor over gives me basic information about the image - size and web location.

Swirl is a great tool if you're interested in browsing through a lot of images quickly, and it's also very nice if you want to focus on a particular type of image - searching for buses led me immediately to the option of looking at different coloured buses as well as line drawings. Bing is going to have their work cut out to try and catch up.

Related articles by Zemanta

• Google Image Swirl Brings the Wonder Wheel to Photo Search (readwriteweb.com)
• Google adds a sexy new Swirl to Image Search (thenextweb.com)

Revisiting Google Squared

Suppose that you need to collect data on English castles, to include an image, description and details on a nearby town. Unless you've got a helpful book to hand (!) it might take a while to drag all of this information together. Then of course you remember that you also want phone numbers, which means you've got to go back and do the searching again. How about British poets, African countries or roller coasters? It's a time consuming and probably in some cases a fairly boring task.

This is where Google wants to help you out, with Google Squared.

The idea is simple - you tell it what you want to find, it creates a matrix filled with goodies based on what it discovers. You can then add in new factors according to taste, and Google will then go off and find the required data and slot it in for you. This turns a 30 minute job into a 3 minute breeze. Or... (pregnant pause) does it?

Let's go through an example - I've chosen English Castles, mainly because it's a subject that's close to my heart. This is the matrix that gets formed as a result of the two word search:

There's actually a total of 13 castles, which is a reasonable number to start me off; I can delete any of them, and add in others. The content is pulled from various resources - Wikipedia, specific castle home pages and so on - the searcher really doesn't have any control over this, so Google really does need to come up trumps here. Most of the time I think it does - I'm getting good initial data about Kenilworth for example, although the Hever Castle content is much more of a sales pitch, but it is pulled from the castle website.

The fourth column is 'country' which seems rather redundant give the nature of search, but it is concerning to see that the Tower of London has been relocated into the United States. To be fair however, a mouseover of that cell does show that the information is regarded as being low confidence, and I'm given three other options:

However, I'm still not impressed with the options available - 'London' would be helpful and to be honest, the clue rather IS in the name.

That's all that Google Squared turns up for me, but it's possible to add in other data options, and this is what Google is offering me:

I think these are reasonable options, and of course I can add my own. Choosing 'telephone' gets Google to fill in 9 cells for me, but these are all listed as 'low confidence' - understandable in the case of the Tower of London, since the number given is that of a named doctor working at Guy's Tower, Guy's Hospital, London. Really not acceptable at all. This is where Google Squared breaks down, catastrophically. I have to go through each cell in turn, checking and double checking, and even though the source of the data is given - I still have to check it all again anyway. Google isn't saving me time at this point - it's actually costing me time, since I reckon that I could find good quality authoritative data quicker myself than going through each option and checking myself.

I tried to add in other data options, such as 'purpose', 'disabled access', 'architectural style' and 'public transport'. Some of these were my own ideas, others were suggestions by Google. All met with total failure - the one which made me laugh the most was Public transport for Kenilworth castle is apparently:

Oh dear.

In case I was just being difficult (always a possibility with me!) I tried some of Google's own suggestions. To say 'laughably inept' is to be on the kind side. Let's take a brief look at their idea of British poets:

Jeremy Reed? The description itself says '...an American professional baseball player'. I have no idea where Google's got that from - a search in their own database doesn't turn this guy up. All that I can think is that in the Wikipedia there's a reference to Jeremy Reed, described thus: "Jersey-born writer Jeremy Reed is a British poet and prose stylist." Perhaps Google pulled that from one place, re-ran the search elsewhere and came up with that. Really, this is about as stupid as pig dribble.

They're not all this bad, and in fact the 'Romantic movies' one is actually quite good. However, none of them are good enough - not even close to it. Suggested columns are hit and miss at best, information is missing in most cases, all the information needs to be doublechecked and in some cases it's so bad a blindfolded senile monkey could do better.

Google - take this away and bring it back when it works please. I love the concept, but the execution is dreadful; you should be ashamed.

Google Search adds more search filters

Google has added in some new search filters which are quite interesting. When you've run a search you'll see a 'Show options' link top left below the search box. Opening that up shows options to limit results to video/forum/review/blog sites, and recent results (including the past hour in an attempt to fight back against what Twitter is doing perhaps). A nice new option is 'All results' to limit your search to visited pages or not yet visited pages (obviously you need to be logged into your Google account for this to work). Another fun one is 'fewer/more shopping sites' which reminds me of the old filter the Microsoft used to offer for more/less commercial sites, so I'm pleased to see this return in another form on another search engine. Finally you can limit to images from a page, or 'more text'

Really nice ideas, and excellent innovation. Who knows, this search engine might well catch on! More details on the options on the official Google blog.

Google - Internet Stats

Google - Internet Stats. Do NOT get excited. I did, and now I'm disappointed. The strap line is 'Welcome to our collection of the latest internet stats'. Sounds excellent, doesn't it. So I bundled in and did a search for internet growth uk. Reasonable request I thought, but apparently not, as there were no results at all. What - nothing? Nope, nothing. Nil. Zip. Nada. On yer bike pal.

How about 'internet growth' then? The little symbol started to whir. It continued to whir. It carried on whirring in the background while I went off to Twitter, and you know how long you can spend there. Finally I remembered what I was supposed to be doing. Still whirring. Eventually it stopped. Hurrah! "24% of mobile phone users in the UK have watched mobile TV and/or video." What use is that to me?

The sorry tale continued. This is basically a collection of statistics that Google has found from various resources which it doesn't link to, so it's not easy to check the original source. The most recent statistic that I found was dated April 2009. No, not a specific date, just a month.

Genius!

Locking Google SafeSearch

Locking Google SafeSearch.This is not an option that's available yet; it's something Google is considering, so don't get excited or hunt for the option yet. "To lock SafeSearch settings, you'll need to log in using a Google account and select the locking option from the preferences page. "Use this page to lock the SafeSearch settings selected. The lock takes effect when you sign out. when locked, SafeSearch displays a distinctive image as the background of the search results pages," explains Google."

Google Fast Flip

When you're looking at a news stand or collections of magazines in WH Smiths I imagine that you do what I and most everyone else does - we look at the pretty pictures to see what attracts us. Google folk obviously do the same thing as well, since they've now created Google Fast Flip which allows users to skim through news stories by viewing an enlarged thumbnail. Looks something like this:

 

You can obviously choose different sections to look at, and can flip back and forth between different screens. I found the images slightly too small to see easily (certainly on my laptop) but certainly there were a few items that caught my eye that I probably wouldn't have looked at.

It's an interesting idea, but in all honesty I don't tend to visit Google news all that much - isn't that what RSS and news alerts are for?