Why a Blacksmith can never be an Algorithm
An associate asked what I thought about Mahalo. In addition to pointing him to our previous post, Mahalo is a Ho’opa’ani, I gave him the following response which I’ve decided to share with you all.
In the real world the human touch is coveted; held in the highest regards for quality craftsmanship and skill. In the land of the search engines the human touch is of low utility, at best. The same reason you wouldn’t let an algorithm make your wrought iron gate, is the same reason you shouldn’t let a human collect the endless sites and information on the World Wide Web.
That said, there’s a place for listings of reviewed sites. Heck, there’s a company that you might have heard of that was founded on just this service. It’s called Yahoo.
The belief is that the major search engines all have some hand manipulation of results. As long as this is constrained to issues where their algo simply fails (such as time sensitive issues), I have no problem with it. It’s a best of both worlds situation.
I have some doubt that sites such as Mahalo that use purely hand built results can provide quality results, however. There are some major issues:
- Scaling - You can only scale the service by throwing people at it. If you’ve not built results for a particular query, it’s either going to be blank, they’re going to have to syndicate results from a true search engine, or they’re going to have to build their own true search engine. There are nearly infinite query possibilities; Mahalo can’t possibly cover them all. Today’s major search engines are probably going to provide better results on average than Mahalo. This is unlikely to change in the near future, if it can even change at all.
- Momentum – Here’s the dirty little secret: Providing better query results than Google is not sufficient. All Google has to do is to be good enough to convince people not to change. Mahalo might have the best three results as the top three, and Google might just have them on the first page. That’s probably enough to maintain loyalty. Yahoo and MSN face this same problem, by the way.
- Bias – People are biased. With a small number of reviewers, you can possibly maintain quality control, but as it scales, bias (and outright corruption) will creep into the system. This is part of what happened with Open Directory. If you could find them, there were editors that would add your site to a category for an under-the-table fee.
- New Sites – How are new sites supposed to be found? It’s bad enough with the search engines, but at least with them you can optimize your pages and build links to your site to crawl up the rankings. Other than a major PR campaign to get the editors’ attention, it’s not clear how you do this with Mahalo.
Some might say, “What about Wikipedia? It’s results are both man-made, and more often than not - credible.” Wikipedia, however, is not the exception because Wikipedia is an entirely different frame of reference. People are drawn to Wikipedia for “known-unknown” information, whereas people are drawn to Google and Yahoo for the absolute unknown: maps, email, job listings, shopping, chatting and more. All of which, no human-driven search engine has yet to fully optimize.
In the end it is a well-known fact Mahalo’s creator Jason Calacanis does not like the idea of search engine optimization, or its providers. And while I agree with him that there are some in the SEO industry that prey on the naivety of new sites and businesses, this sentiment cannot be carried to the industry as a whole, simply because the industry works. New sites are created everyday, and no amount of human strength can match the combined aptitude of Google and Yahoo. Calacanis’ entrepreneurial efforts should be applauded, however the idea of human powered search engines is merely a revolution started for revolution’s sake.
August 3rd, 2007 at 11:22 am
Great post, but I have a few comments.
This would seem to be true:
If you’ve not built results for a particular query, it’s either going to be blank, they’re going to have to syndicate results from a true search engine, or they’re going to have to build their own true search engine.
However, Mahalo staff are very creative and have dome up with an even sillier alternative: show apparently random results. You can see a screen shot of the search results I got. Evidently, a particular type of increase in knitting is related to “Gay Marriage” and “Asteroids”. You can see the screen shot of Mahalo search results here.
August 15th, 2007 at 6:50 am
[…] You should really read his full post, entitled “Why a Blacksmith can never be an Algorithm” […]
August 15th, 2007 at 12:40 pm
Great post… you’ve nailed four of the most important issues we face–and we face many more than just four issues!
Let’s me work backwards….
4. New Sites: We find new sites two main ways:
a) Proactive: We have 50 filltime folks and 800 part-time Guides right now. Each of them specializes in a particular area or two. So, in sports we have 30-40 folks, and in food or science we have 20-30 for example. These folks are reading all the top blogs, news sources, delicious, reddit, digg, Netscape, etc. for their particular topics–DAILY! When new sites come up they find out about them. These folks also talk to each other constantly.
b) Reactive: It is very clear how to get our attention: submit you site for any search term! Thousands of folks are doing this every month already. Users can submit sites to us all day long…. in fact, we’ll see one site owner submit their pages for 20 different related search engines. Getting our attention is, in fact, EASY!
In fact, it is much easier to get our attention than Google or Yahoo’s. Those folks don’t talk to you about your position on the SERP–we will!
3. Bias: This is a big issue, but like Wikipedia or a newspaper we have an NPOV rule and we enforce it and police it religiously. Take a look at a page like George Bush or abortion and you’ll see what I mean. In fact, I would challenge you to find one piece of bias on the site…. if you do I’ll buy you a beer for each one!
Second, our site is 100% transparent. You can see the history of each page and every edit is done a by a real person with a real name getting paid real money. If someone does something biased we will find out and they will be fired. It’s that simple. We have an easier time with this than say Wikipedia because we have a paid relationship with our Guides/editors–and they don’t want to lose the paid relationship.
Third, every page has a message board where we will publicly debate any bias.
So, again, find me one example of bias in the 9,200 pages we have right now and the beer(s) are on me!
2. Momentum: This is actually probably wrong. In the end the better product usually wins on Internet. In fact, Google is the prefect example of this. They were 10-30% better than other search engines and they slowly took marketshare away from folks. So, in fact, if you are better you should get people to move. Remember there is zero lockin on the Internet. People can switch from search engine to search engine easily with no cost (switching social networks? ok, there is a cost there… you have to rebuild your network).
1. We actually do syndicate machine search already… so, we really don’t have to have everything…. we need to have enough so that you consider using us in addition to Google or Yahoo. In fact, if we can get just 1% of searches that is, as folks have pointed out, a very significant business. Additionally, we have 9,200 pages right now that service 20+ searches each… we did 700 pages last week. We are in our third month…. do the math, this could get big quick!
Anyway, I totally think you nailed the key issues and in fact your key issues are EXACTLY what the VCs said to me when we were raising money for the company… so, at the very least a) you’ve got a career ahead of you in venture capital and b) I’ve got a lot of work to do!
Mahalo for thinking of us and come back to the site every 60 days I suggest… it seems to change radically at that pace.
Also, I might suggest trying out Mahalo Follow… our toolbar/sidebar for Firefox that only opens when we have a human crafted result: http://greenhouse.mahalo.com/Special:FollowDownload?id=9