When Google Fails


cc licensed flickr photo shared by westpark

“Just Google it.” (No one ever says, “Just Yahoo it,” much to Yahoo’s dismay.)

I’m not convinced that Google’s artificial intelligence has made it out of grade school yet. To prove this, all you need to do is imagine a destructive idea—making Google have the wrong answer.

You can already see this idea being carried out on a small scale, with people learning how Google ranks pages, and subsequently getting their pages ranked higher. These pages are only intended to serve impression ads, and often do not have the information the user was looking for when they first searched. While this phenomenon is still a growing one, it has so far been contained by the narrow interest of the perpetrator.

But, what if a group of people decided that they had no revenue interest in any one particular manipulated Google result, but instead wanted to make as many common inquiries to Google return results that are completely inaccurate and irrelevant—at least within the first 10-20 listings.

This concentrated effort would show that the algorithms that Google uses could not effectively address the problem before the troublemakers would be able to adapt and evolve themselves. A band of misfits numbering less than a thousand could probably stay ahead of Google’s efforts to stop their trouble making. This is largely because Google is doing everything in house.

To illustrate, imagine a factually correct Wikipedia result that someone one day decides to alter. The mistake will be corrected in a very short period of time, relatively speaking, because each visitor is like a tiny quality control department. A page of irrelevant Google results, on the other hand, will go much longer before being corrected, given what is required for Google’s personnel and algorithms to notice the problem.

I believe a vulnerability that undermines Google Search is an Achilles heel to the entire brand. There is a threshold of error that users will collectively allow, and if Google continues to move toward this instead of away from this, users will begin to search for an alternative (pun most definitely intended.)

I’m trying to describe a way for a new service to effectively compete without giving away too much of the delicious secret sauce.

7

The Decider

What if there was a decision making program you trusted more than your own thought process (or at lest equally), that not only told you what you want and need, but why you want it or need it? It’s the logical extension of preference profiling, social networking, data mining, heuristic analysis, and dynamic “learning” algorithms.

It wouldn’t just be making recommendations based on previous behavior, e.g., we recommend this book because you bought this other book, but rather based on anticipated behavior. It wouldn’t just serve you content, it would have a conversation with you. It’s taking a search engine, a social network, artificial (and perhaps personal) intelligence, an advertising platform, and a retailer—and throwing them in a blender.

Oddly enough, this idea is from my Spring 2009 notebook, and a website called Hunch launched later that same year, which is quite similar to what I had in mind. Hunch makes recommendations based on a profile you create. This could easily go off on a tangent about data portability and how your actual profile is much better than anything you can create from scratch in a few questions, but enough of that.

7


cc licensed flickr photo shared by ramyo

ME: What’s up?

DE: You should use Turbotax to do your taxes.

ME: Not today. What else?

DE: You should buy NIke red tennis shoes from Zappos for $129.

ME: Um …

DE: You need a new pair of shoes.

ME: Well, yeah.

DE: And, you only have $150 to spend.

ME: True.

DE: Shoes will give you the greatest value for the money.

ME: I see. But, why tennis shoes?

DE: Because you already have relatively new dress shoes and casual footwear.

ME: Got it. Red?

DE: It’s your favorite color.

ME: I love you.

DE: I know.

A Ghost In The Machine

If you made a computer program that sent [on your behalf] random emails, texts, and comments, how long could you get away with it?

That’s the line from my frist Moleskine cahier notebook, but we could use some elaboration. Essentially, the idea is to construct a computer program that learns your writing style from your emails, text messages, comments around the web, social networking updates, &c., and is somehow intelligent enough to compose what it sees as likely new messages. As soon as you flip the switch, you have an artificial self posing as you in the online world, that posts new things at random times. The question, then, was how long could this charade last before it was detected?

I like the idea because it seems like an interesting way to make a substantial piece of commentary about how real you can appear to others as long as you are active online.


cc licensed ( BY NC ) flickr photo shared by thelunch_box

A friend of mine, Brian Brushwood, recently thought of something along the same lines—he calls it Afterlyfe. What a great name! According to his website, the process is activated upon Brian’s death! From the site:

Once I stop checking in, Afterlyfe will assume I’ve kicked the bucket, and go into action, taking control of my facebook and twitter pages.

From that point on, Afterlyfe will use all my previous tweets and facebook updates to recreate a digital simulacrum of my life. The goal is to make me the world’s first virtual ghost.

For starters, we’re going to keep it simple:  the default settings will be that the moment I die, Afterlyfe will make an exact copy of my last year in tweets, and release them at the exact times they were originally posted, year after year.  Every year, followers will see me complain about taxes come mid April.  You’ll get my same Christmas tweets. Happy birthday wishes I made in the past will continue to arrive, year after year, right on schedule.

Now, Brian and I never spoke about this before he revealed his project, and I give him 100 percent of the credit for coming up with his idea; this isn’t that kind of post. I do think, however, that merging both ideas could create something even more compelling than both on their own. That is, use the concept of activating the program upon death with the idea of intelligent software that generates new material to post.

With a sufficiently advanced program, you could have some amazing things happen. I think you could continue to learn new things about someone after their death—from them!

7

PS: I bet Facebook dies before Brian does.