Friday, February 1, 2013

Ensemble Classifiers And The Wisdom Of The Crowds

Here is the start of my long-term, self-funded project looking at the relationship between ensemble classifier research in computer science and in the social sciences.

So far, I found some serious bugs in the introductory material of the widely-praised best-seller: The Wisdom of Crowds, by James Surowiecki, published 2005 by Doubleday; Anchor.

I documented the probable bugs in this web page: Cattle Weight.

The open-source Wisdom of Crowds text that contains the bugs is available as an Amazon preview.

Sunday, December 23, 2012

A Trivial GMail Bug

Previously, I posted details about a GMail bug: A Trivial GMail Bug. I tried now to see if the GMail archiving bug has been fixed but it still exists.

If you create an incoming GMail filter (e.g. delete all incoming emails containing the word cat), then send an email from the same account using the word cat, the following occurs:

GMail sends your email successfully, but moves a copy of your sent email to your trash folder instead of to your sent email folder. 

Sunday, December 16, 2012

Google's Audio reCAPTCHA: A Large Bug

CAPTCHAS enable computer programs to determine if a person or a robot is requesting access. Usually, it is desirable to block robots from accessing a program.

Some people are sight-impaired and must use audio CAPTCHAs.

If you want, you can hear Google's audio reCAPTCHAs by clicking any Blogger Comments icon, and then Audio.

Here is the large bug: 

If you listen to the Google Blogger audio reCAPTCHA instructions, they say: "Please type every digit (e.g. 1, 2, 3, 4) you hear". The problem is the visual instructions for the same audio reCAPTCHA are: "Type the words (e.g. one, two, three, four) you hear".

Should sight-impaired users type digits or words?

Wednesday, December 12, 2012

Google Advanced News Search: A Small Bug

If you go to and click Advanced News Search, a form appears with fields you can use to filter news results.

Two of these fields are:
 - Source (e.g. CNN, New York Times)
 - Location (e.g. California, India)

Google defines these fields here:

"In the Source field, you can identify a specific publication whose articles you'd like to see exclusively in your search results. Likewise, in the Location field you can specify from what geographic location you'd like to see articles."

This is the small search bug: The "Location" label is ambiguous. "Location" can mean the news stories occur in a particular location, or the news sources are located in that location.

Google's intent is that the publication sources are in a particular location.

To fix the ambiguous field labels I suggest these three fields instead:

 - This publication source:
 - Sources located here:
 - Stories occurring here:

Wednesday, November 28, 2012

On The Utility Of Google Search Results Metrics

Perform a Google Web search for [a]. Google displays something like this:

About 25,270,000,000 results (0.22 seconds)

What is the utility of finding 25,270,000,000 results?

If you display the default 10 results per Web page, you would have to click Next about 2.5 billion times to get to the last entries. At a second a page, that's approximately 80 years.

The number of results can indicate the popularity of a set of search keywords. However, I'm going to call the multi-billion results metric a bug because most search results are not accessible by users. 

Would an ice-cream shop list 25 flavors if it can really serve only the 5 most popular flavors?

Trivial Bug: Google Timeline Of News Articles

If you scroll down the Google News Web page, you can find this text:

"The time or date displayed (including in the Timeline of Articles feature) reflects when an article was added to or updated in Google News."

The trivial bug is that the "Timeline of Articles" feature no longer exists.

Sunday, December 11, 2011

Google Verbatim Feature: Trivial Bugs

Google Verbatim is described here.

Bug #1: If you run a Verbatim search, you cannot simultaneously use any of the other search features such as time range or location. You can use one of those features if you surround each of your keywords with quotes instead of using Verbatim.

Bug #2: Verbatim means using an exact word-for-word phrase. Google Verbatim does consider search keyword order, but it often also returns non-verbatim search results.

Example: A Verbatim search for [Stephen King Book List] can also return searches for [List Stephen Book King].


Wednesday, November 23, 2011

The Curfew Tolls The "Knol" Of Parting Day

Google announced that it is paring back on some of its free offerings such as bookmarks, and friend connect, and knol. 

Perhaps, Google is reverting to its founding tenet: Do one thing and do it really well.

Tuesday, November 22, 2011

Google Advanced News Search: Large Bug

The bug is that, as far as I can tell, users can no longer access the Google News advanced search page from a keyboard.

Previously, you could display the advanced news search page using a link that was
accessible from a keyboard or by using a mouse.

Now, to display the advanced search page you must use a mouse to click an arrow within the search field. 

This raises an issue of accessibility for people who cannot use a mouse.

Monday, November 14, 2011

Google's Recipes Feature More Large Bug Examples

Previously, I posted examples of Google recipe feature bugs.



Here are a few more bug examples:

Example #1:

Using the recipes feature
[fruit salad with pineapples apples oranges and bananas]
About 7 results

Regular Google search
[recipes fruit salad with pineapples apples oranges and bananas]
About 826,000 results

Example #2:

Using the recipes feature
[grilled sandwich with eggplant tomato cucumber and mozzarella]
About 9 results

Regular Google search
[recipes grilled sandwich with eggplant tomato cucumber and mozzarella]
About 190,000 results

Example #3:

Using the recipes feature
[iced tea with lemon mint and passionfruit]
About 7 results

Regular Google search
[recipes iced tea with lemon mint and passionfruit]
About 123,000 results

Google's recipes feature is described here:

Sunday, November 6, 2011

Chrome Browser: Another Trivial Bug

Here's the trivial bug: Chrome Browser -> Options -> Personal Stuff -> Passwords and Autofill features should not be available per browser. 

Those personal stuff features put users at risk in multiuser computer situations such as public cafes, classrooms, and libraries.

Tuesday, November 1, 2011

Google's Advanced Search: Trivial Bug

When you run a Google Web search, Google's algorithms try to look for your search keywords in several places including Web page titles, text, and URLs.

If you want, you can use Google's Advanced Search filters to narrow down your search to specific areas.

This trivial to small-sized bug occurs when you try to search only in Web page body text: Advanced Search Page -> Where your keywords show up -> In the text of the page.

The problem is the Web body text searches incorrectly use the search operator as a literal search keyword instead of as an operator. You can easily see the bug if you try any of these searches:

[allinbody: house]

[allinbody: car]

[allinbody: tea]

[allinbody: football]
[allinbody: algorithm]
[allinbody: science]

Sunday, October 16, 2011

Choosing A Google User Name: Another Trivial Bug

Try to create a Google account with, for example, a user name "alicesmith" and also a second Google account with a user name "alice.smith".

After you create the
"alicesmith" account, if you try to create an "alice.smith" account Google displays a message: "alice.smith is not available."

The Google error message is a trivial bug because it can lead you to incorrectly conclude a third-party already has created the "alice.smith" account.

In fact, the "alicesmith" account owns the "alice.smith" account. Google ignores nonconsecutive dots in a user name, and should probably specify that detail on its "Create An Account" form.

Thursday, September 1, 2011

Google's Recipes Feature: Some Trivial Bugs

Google has a "recipes" search feature described here:

"Search for a recipe or ingredient (ex: chili, banana bread, lentils). After searching for a recipe or ingredient on Google, select Recipes in the left-hand panel on the search results page. You can filter your results by ingredients, cook time, or calories."

The Google "recipes" feature has several trivial bugs:

Bug #1:

Ordinary searches are better than searches using the "recipes" feature because they A) return a greater number of recipe results, and B) allow you to add contextual keywords to make your searches more precise.

Finding recipes using a standard Google query might require an additional mouse click or two to drill down to the recipes, but the tradeoff of gaining more relevant recipes makes it worth while. 

Example #1:

Using the recipes feature
[apple pie]
About 4,630,000 results

Regular Google search
[recipes apple pie]
About 9,760,000 results

Example #2:

Using the recipes feature
About 35,900,000 results

Regular Google search
[recipes pizza]
About 106,000,000 results

Example #3:

Using the recipes feature
[tapas madrid hot spicy inexpensive]
About 1 result

Regular Google search
[recipes tapas madrid hot spicy inexpensive]
About 259,000 results

Example #4:

Using the recipes feature
[smirnoff cranberry twist]
About 1 result

Regular Google search
[recipes smirnoff cranberry twist]
About 147,000 results

Example #5:

Using the recipes feature
[tomato sauce mozzarella olive oil garlic basil oregano fennel pepper]
About 2 results

Regular Google search
[recipes tomato sauce mozzarella olive oil garlic basil oregano fennel pepper]
About 394,000 results

Example #6:

Using the recipes feature
[spanish olive crusted salmon]
About 4 results

Regular Google search
[recipes spanish olive crusted salmon]
About 219,000 results

Bug #2:

When you use the "recipes" feature, a list of ingredients displays with yes/no option buttons. The first part of the bug is that it is not clear what result you get if you do not select either yes or no for an ingredient. Maybe the ingredient will be included in the recipe and perhaps it won't?

However, the main part of bug #2 is that the set of ingredients you can include/exclude is limited and in many instances arbitrary.

When using the "recipes" feature, if you want a list of salad recipes to contain cucumbers, broccoli, and gorgonzola cheese and exclude jalapeno peppers you are out of luck. However, you could run a query for those recipes using a traditional Google search.

Here are the severely limited forced choice inventories of ingredients you can include/exclude from a recipes feature search for [salad] and for [omelette] :

Example #1:
Using the recipes feature search for [salad]:
spinach yes/no
chicken yes/no
poppy seeds yes/no
dill yes/no
cranberries yes/no
sesame seeds yes/no
mustard yes/no

Example #2:
Using the recipes feature search for [omelette].
haddock yes/no
watercress yes/no
crème fraîche yes/no
tarragon yes/no
feta yes/no
peppercorns yes/no
cheddar cheese yes/no
chives yes/no

Sunday, August 21, 2011

A Small Google Search Bug

When you use the + prefix with Google search keywords, Google matches exact literal strings and does not return synonyms. If you search for spaghetti you can get a result for pasta, but if you search for +spaghetti you will only get spaghetti.

Should a Google search using the + prefix return more, fewer, or the same number of results as a search without the + prefix?

Here are a couple of possibilities:

1. The + prefix should return fewer results than the same query without the + operator: The + operator does not return synonyms, so if you search for keywords and their synonyms you should find a greater number of results pages.

2. The + prefix should return more results than the same query without the + operator: If searching for keywords and their synonyms requires more computational power than matching exact literal strings and if Google searches have time-out durations the +prefix keyword searches will return more results for any given amount of computational time.

The small bug is the discrepancy between these results:

1. Single keyword queries: + prefix returns fewer results.

2. Two keyword queries: + prefix returns more results.

Either the number of results feature is incompletely documented, making the metrics somewhat useless, or there's an algorithmic bug.

The examples below use arbitrary search keywords. In almost all cases, the + prefix returns fewer results for single keyword searches, but more results for two keyword searches.

Example #1:

About 556,000,000 results

About 534,000,000 results

About 212,000,000 results

About 204,000,000 results

pizza bicycle
About 17,200,000 results

+pizza +bicycle
About 53,100,000 results

pizza OR bicycle
About 727,000,000 results

+pizza OR +bicycle
About 1,440,000,000 results

Example #2:

About 939,000,000 results

About 903,000,000 results

About 756,000,000 results

About 729,000,000 results

coffee tea
About 75,900,000 results

+coffee +tea
About 120,000,000 results

coffee OR tea
About 1,630,000,000 results

+coffee OR +tea
About 3,160,000,000 results

Example #3:

About 276,000,000 results

About 269,000,000 results

About 102,000,000 results

About 99,300,000 results

ant caterpillar
About 6,350,000 results

+ant +caterpillar
About 16,700,000 results

ant OR caterpillar
About 365,000,000 results

+ant OR +caterpillar
About 731,000,000 results

Example #4:

About 621,000,000 results

About 575,000,000 results

About 1,400,000,000 results

About 1,290,000,000 results

baseball football
About 321,000,000 results

+baseball +football
About 325,000,000 results

baseball OR football
About 1,860,000,000 results

+baseball OR +football
About 3,610,000,000 results

Example #5:

About 38,000,000 results

About 37,200,000 results

About 167,000,000 results

bout 163,000,000 results

crayon pencil
About 8,260,000 results

+crayon +pencil
About 8,480,000 results

crayon OR pencil
About 399,000,000 results

+crayon OR +pencil
About 398,000,000 results

Sunday, July 31, 2011

A Trivial GMail Bug

The bug is that GMail partially uses your customized incoming text filters when you send GMail.

Try this experiment: Create a GMail filter that deletes all incoming emails that contain the word "cat". Now send an email that contains the word "cat".

GMail sends your email, but it also moves a copy of it to your trash folder instead of to your sent email folder.

Friday, October 1, 2010

Improving Google's Two-Factor Authentication

Google is rolling out two-factor authentication for Google Apps. See these posts:
Here's the Google authentication flow:

1. User: Enters a valid user name and password (something known by the user).
2. Google: Sends an SMS verification code to the user's mobile phone (something the user has).
3. User: Enters the SMS verification code (something the user has) and logs into the Google App.

In my opinion, the Google flow can be improved. I would remove step #2 because it causes a potential latency issue. The Google-to-user phone link is a third-party application that may be unreliable. Latency presents a serious problem if a user needs to access time-dependent information such as a set of medical records.

The Google authentication flow requires three steps. Each step, in my opinion, should correspond to a single authentication factor. Currently, the "what the user has" factor requires steps #2 and #3.

The solution I suggest is to send the next SMS immediately after the user logs out of the application. Under this scenario, the user actually has the SMS instead of waiting until Google sends it in real time during a login attempt.

As a backup, and in case the user misenters the verification code, the verification code screen should also contain a "Send new verification code" button that displays if the user enters a correct user name and password.

Thursday, September 9, 2010

Improving The Google Instant Feature

Google Instant is a new search feature that automatically returns pages of search results as a user types partial (e.g. "c") and complete (e.g. "cat") search keywords. See the Google blog post, Search: now faster than the speed of type.
Currently, Google Instant displays search results only for the topmost entry on the Google Suggest drop-down list. If you search for "cat" or "car" or "computer" and type "c", Google displays a results page consisting only of "craigslist" entries.

"Craigslist" is the Google search prediction for users who begin Web searches with the letter "c".

In my opinion, Google Instant could be improved by displaying the search results page in sections.

If the search results page displays 12 entries, the first 3 search results could be for "craigslist", the next 3 for "CNN", the next 3 for "chase", and the last 3 for "costco".
These sections correspond to the first 4 Google Suggest entries, instead of only to the highest Google Suggest entry.

When a user wants to view search results only
for the topmost Google Suggest entry, the user can click Search or hit Enter.

Wednesday, September 1, 2010

Several Trivial GMail Bugs

You can see these trivial bugs here: GMail Inbox -> Show search options -> Date within field. As an arbitrary example, in the "Date within" field specify "Date within 3 days" of "9/1/10". Click "Search Mail".

Note: The current date is 9/1/2010.

Google runs the query and displays this spurious message: "after:2010/8/29 before:2010/9/5".

Here are the bugs:

Bug #1: "before:2010/9/5" is a bug because Google's Gmail cannot run a search into the future side of the stated time range.

Bug #2: "after:2010/8/29" is also bug because "Date within 3 days" of "9/1/10" should include the current day and the 3 preceding days. After 2010/8/29 suggests 1) 2010/8/30 and 2) 2010/8/31 which is a 2-day, not a 3-day range.

Bug #3: Although the query message says "after:2010/8/29", the results actually include emails received on 2010/8/29. The fix would be to change the word "after" to "from". The "before" part of the message should be deleted when it involves future days.

Bug #4: The Gmail query message label: "after:2010/8/29 before:2010/9/5" uses a yyyy/m/dd date format. Though not incorrect, mm/dd/yyyy or dd/mm/yyyy formats are more typically used.

Saturday, August 14, 2010

Some Chrome Browser Bugs

These bugs are caused by Chrome setting up a dependency on Microsoft third-party software. Chrome shares the IE browser's "Internet Properties" dialog box.

In Chrome, click the wrench icon then Options -> Under the hood -> Change Proxy Settings. The IE Internet Options dialog box displays. [Hard to believe.]

Note: The Mozilla FireFox browser avoids this dependency.

  • Up to date Chrome 5.0.375.126
  • Up to date IE: 8.0.7600
  • Up to date FireFox: 3.6.8
  • Up to date Windows Pro 7
Bug #1 (security):
A malicious exploit specifically targeted at IE browsers that changes internet option parameters will also adversely affect Chrome's shared internet option parameters. For example: 1) Reducing browsing security levels, 2) Allowing active content to run, 3) Changing how the browser connects to the internet.

Countering these exploits against IE and Chrome may require an IE patch or update.

Bug #2 (security):
IE users in multiuser computer situations (cafes, classrooms, libraries) can reduce the browsing security level of Chrome users and vice-versa.

Bug #3 (security):
Chrome browsers running on Windows computers do not support SSL3.0/TLS1.x encryption.

Bug #4 (security and usability):
A Chrome user can try to change features that Chrome does not support in the internet parameters dialog box. For example: IE InPrivate browsing, SSL3.0, TLS1.x.

A user connecting with Chrome to a site requiring SSL3.0 will get an error message even though SSL3.0 is selected in Chrome.

Bug #5 (security and usability):
Under the internet properties "Advanced" tab there are several features marked with asterisks (*). A label notes that "* Takes effect after you restart Internet Explorer".

Question: Does a Chrome user who changes one or more * features need to open and close an IE browser session to activate the options in Chrome? From the Chrome GUI the required user behavior is completely unclear.

Bug #6 (security and usability)
: Chrome displays some duplicate parameters that are separately and inconsistently configurable. These duplicate parameters include configuring: 1) a start page, 2) auto-complete forms and passwords features, 3) a default browser.
  • Start page: You can set duplicate home page parameters that are different. Set the first value in Internet Properties -> General Tab -> Home page. Then set a different home page here: Options -> Basics -> Home page. Chrome will use the value you set here: Options -> Basics -> Home page, ignoring the home page value you set under internet properties. However, the value you set from Chrome under internet properties -> Home page will change the startup page for any IE users on that computer.
  • Auto-complete forms and passwords: You can set this feature inconsistently in two places. Set the first value under Internet Properties -> Content -> AutoComplete -> Forms and User Names and Passwords on Forms. Then set inconsistent values in the duplicate fields here: Options -> Personal Stuff -> Passwords and Form Autofill. Chrome will use the values you set here: Options -> Personal Stuff -> Passwords and Form Autofill, ignoring the values you set under internet properties. However, the values you set from Chrome under internet properties -> Content -> AutoComplete will take effect for IE users on the same computer.
  • Default browser: I have FireFox as my default browser. In Chrome, if I go to Basics -> Default browser it says: "Google Chrome is not currently your default browser. Make Google Chrome my default browser." In Chrome, if I go to Options -> Under the hood -> Change Proxy Settings -> Internet Properties -> Programs -> Default web browser it says: "Internet Explorer is not currently the default browser. Make default."
In my opinion, the above set of problems should be assigned the following bug severity levels:
  • Security: between medium and high
  • Usability: critical