Google's Amit Singhal Interview: Developing The Knowledge Graph

[engine]

Google's Amit Singhal Interview: Developing The Knowledge Graph [mashable.com]

The future of Google Search, though, could be a very different story. In an extensive conversation, Singhal, who has been in the search field for 20 years, outlined a developing vision for search that takes it beyond mere words and into the world of entities, attributes and the relationship between those entities. In other words, Google’s future search engine will not only understand your lake question but know a lake is a body of water and tell you the depth, surface areas, temperatures and even salinities for each lake.

Google now wants to transform words that appear on a page into entities that mean something and have related attributes. It’s what the human brain does naturally, but for computers, it’s known as Artificial Intelligence.

It’s a challenging task, but the work has already begun. Google is “building a huge, in-house understanding of what an entity is and a repository of what entities are in the world and what should you know about those entities,” said Singhal.

[tedster]

I'd say this is the long-term view of where Google is headed - or really, where search is headed altogether. We left the world of pure text match years ago, but not by a lot so far.

It's definitely something to keep in mind as we try to make sense of the way SERPs are evolving. Right now, experienced searchers might not even formulate a search such as "the ten deepest lakes in the US" but would probably come at this idea from related searches that stand a better chance of finding a text match - and then go on from that initial set of findings.

The article states "while 200 million entities is a lot, the world of knowledge is vast." Indeed. Given all the possibilities that a true AI would need to address, 200 million entities is a mere sneeze.

[onebuyone]

What's wrong with pure text match btw? Why change something that is working?

[jinnguyen]

Yeah right, the Japanese started the research on Artificial Intelligence in 1972 - and failed. Nowadays, when you search anything on Google, the results always returned the opposite of what you looking for. If you search "how to" then Google will placed the #1 position "how not to" ... search for something Yes, they gave it a NO. So much of a headache for me. Nevertheless, their ads on that page are laser targeted. That's what they called "Artificial Intelligence."

[Zivush]

If you search "how to" then Google will placed the #1 position "how not to"

You meant "how not to"! Don't mess it up.

[loner]

"Sometimes a cigar is just a cigar" - S. Freud

Sometimes a search is just a search.

[FranticFish]

Information sites of any sort, your days are officially numbered.

[aakk9999]

... but Google Fellow and SVP Amit Singhal says Google doesn’t understand the question. “We cross our fingers and hope someone on the web has written about these things or topics

But this is exactly what search engine should do - find results relevant to query.

If Google wants to answer the question itself rather than give search results that answer the question then this is the whole new ballgame.

[FranticFish]

Indeed, and it does raise the question as to where the information that powers those answers is going to come from.

If Google is going to build its own answer database (like Wolfram Alpha, mentioned in the comments on Mashable) that's one thing.

However my reading is that they want to develop AI that would allow them to answer questions they weren't expecting.

If they're going to crawl the web, and use information that others have generated or collated to power their own 'answer engine' then that really does give the lie to any notion that there is some sort of 'partnership' between them and the webmasters whose information powers their results.

[LostOne]

So many phd's at the plex. They need something new to do. Before too long their jobs will be in jeopardy. Just let the machines do it.

<Zager and Evans>

In the year 2025 a man will still be alive(in the plex), got nuttin to do but push buttons to survive.

[bwnbwn]

All I can say is wow, what a task Google has put for their future growth. I have no doubt Google is working with defense funds to help feed this animal. If you assemble the type of minds from around the world Google has put together they have the ability to really put this together.

Just a thought from a long time ago. But the ---- came down to see the city and the tower that the men were building. 6 The ---- said, “If as one people speaking the same language they have begun to do this, then nothing they plan to do will be impossible for them. 7 Come, let us go down and confuse their language so they will not understand each other.”

Think about the above for a minute and you will see this will happen.

What it will do for organtic search is a plain guess. I don't see Google developing their own database of content they can't keep it updated so their search will still be dependent on what we produce.

One thing I do see if outdated content is going to be dead unless it continues to have a bearing on the search.

[Rockyou]

Information sites of any sort, your days are officially numbered.

People need good content not Dog content by Google. I can only laugh at this statement. This topic is not worth wasting time here.

[rlange]

Apparently I'm in the minority, but I don't have a big problem with Google providing direct answers to simple questions (e.g. "How tall is the Empire State Building?" "What's the elephant population of Madagascar?"), with proper attribution, of course.

--
Ryan

[FranticFish]

Attribution doesn't pay the bills.

[rlange]

FranticFish wrote:
Attribution doesn't pay the bills.

I'm pretty sure the height of the Empire State Building isn't what's paying the bills, either...

--
Ryan

[driller41]

They would love to scrape other peoples work and then plaster some adverts around the result. Hopefully we will not let them succeed.

Diversity is what makes the web interesting, panda and it's brand centric focus is chipping away at that diversity which makes the web less interesting.

[Maurice]

Well its a big ask and its one of those NP compete problems eg if I saerch for "crimson lake" do I mean the colour Carmine the lake in canada or somthing else.

If your job title on Google+ is art director you could guess that I meant the paint but how do you scale that sort of knowledge.

Its like real AI or Fusion Power it is always 10 years away :-)

[jmccormac]

Plundering Star Trek. Hardly an original thought between the lot of them!

Regards...jmcc

[zuko105]

"and what should you know about those entities" -

That's a pretty scary thought... how do we leave it up to a for profit company to decide what we "should" know.

[FranticFish]

I'm pretty sure the height of the Empire State Building isn't what's paying the bills, either

Maybe, maybe not.

If someone comes to your site to read your answer to a question there is a chance that they'll click on one of your ads, or do something else. There is zero chance of that happening if they never get to your site.

[loner]

If Goo can answer my questions, maybe they can bump that up a notch and ask them too? That way there will be no intervention or interaction on my behalf and I'll be free to pursue offline interests, like looking for a real job.

I agree with FranticFish- Ultimately, a webmaster's only function will be to volunteer research for Goo.

[ergophobe]

>> pure text match btw?

It doesn't and can't differentiate between different meanings of the same word, it doesn't understand context, it can't find articles on iMac computers if I search on Mac computer or Apple computer. Pure text match is handy to have as a option ("verbatim") which I use a lot (I may actually be looking for misspelling of a word), but it's not usually what you want.

Long term, what I want is the Star Trek computer.

Me: Computer?
Computer: Working.
Me: Which planets are within 20,000 light years and have a gaseous atmosphere with an atmospheric nitrogen concentration of higher than 18% with an atmospheric methane concentration of lower than 22%?
Computer: There are seven such planets. The closest one is Aldebran II-a, 256 light years away with an atmospheric....

Not only is this what I *want* long term, I'm quite sure that something like this (though with a much better interface) is what I'll *get* long term. Or at least the kids in college now will live to see this.

Right now, experienced searchers might not even formulate a search such as "the ten deepest lakes in the US" but would probably come at this idea from related searches that stand a better chance of finding a text match

Actually Ted, I think I disagree, though I'm not 100% certain I know what you mean. I think of myself as an experienced searcher, but over the last couple of years, what I have found is that more and more my searches are adapting to natural language searches with articles and pronouns and so forth.

I'll give you an example from this morning. I wanted to find the OBD port for a 2004 Subaru Forester. The thing is, I didn't have the word "port" and my first old-style keyword search failed. So I fell back on what I consider my new strategy. I typed in my query as if I were talking to a human mechanic and typed in "Where do I plug an OBD reader into a Subaru Forester?" Yes, of course I know half of those are stop words, but guess what? An appropriate page, with a diagram and everything, popped up #1.

I have deep uneasiness about what Google knows about me (and Facebook and Visa and US Bank and REI for that matter), but I think search technology is noticeable improved from five years ago and by a lot. The perceived change is much less obvious because if the search tech is 3x better, the SEs are faced with a web that is growing much more rapidly. Estimates vary, but if we believe Google's official statements [source 1] about the number of unique URLs they have indexed (which is not the number of unique pages, of course):

Number of pages in Google index:
1999: 26 million
2000: 1 billion
2008: 1 trillion

So yeah, the SERPs often seem polluted with irrelevant and overlapping garbage. The problem is there is massively more noise in the system than there was in 2000 and so the filters have to be massively better. In fact, what they are is only considerably better, not massively, game-changing better.

So all of this as a round about way of saying - this is why you change something that's working. What works when seaching 26 million URLs simply won't give a reasonable result with 1 trillion unique URLs.

1. Source: [googleblog.blogspot.com...]
In 2005, the "Surface Web" (i.e. the indexable web) was 11.5 billion pages.

[Panthro]

Good read. Thanks for posting.

[Sgt_Kickaxe]

Now you know why Wikipedia ranks so highly, it touches on every related aspect to your search terms. Thing is, even on Wikipedia, that leads to a lot of skimming to find just the tidbits you really wanted to know.

More isn't always better.

[jmccormac]

Now you know why Wikipedia ranks so highly, it touches on every related aspect to your search terms.

I remember Amit Singhal's post after the Panda fiasco on what constitutes a "good" website. He was, consciously or not, describing Wikipedia.

A few years ago, Microsoft made a very bad mistake with its search engine in believing all that academic crap about the Semantic Web. The reality is that the Semantic Web is a great idea but the real world just doesn't use it. I wonder if Google has made a similar error?

Regards...jmcc

[lexipixel]

providing direct answers to simple questions (e.g. "How tall is the Empire State Building?"

...Just like like when you google "empire state building" on [duckduckgo.com...]

[Robert Charlton]

What's wrong with pure text match btw? Why change something that is working?

ergophobe beat me to it... but we haven't seen pure text match for many, many years. With the huge number pages now in Google, even without the ambiguities of words, pure text match wouldn't scale.

Think "phrase-based indexing refined by user satisfaction and AI."

...natural language searches with articles and pronouns and so forth....

...I didn't have the word "port" and my first old-style keyword search failed. So I fell back on what I consider my new strategy. I typed in my query as if I were talking to a human mechanic and typed in "Where do I plug an OBD reader into a Subaru Forester?" Yes, of course I know half of those are stop words, but guess what? An appropriate page, with a diagram and everything, popped up #1.

For a number of different reasons over the last several months, I have been running test searches in various niches, leaving out all but one or two of the target keywords in a long tail query, but including peripheral vocabulary to describe the situation, and I've also been getting some surprisingly appropriate results.

[santapaws]

always seems strange to me in this search for ai in a search engine and gearing up everything to return ai like results in one hit. Im not sure what universe everyone else is on but in mine no intelligent question is answered in one hit. Its called a conversation, truly. Search engines one day will grasp the reality that to do ai they need to ask questions back for refinement and not think it will work with a best guess one hit reply.

[idolw]

easy way to do it is to limit the results number.
working with a limited number of suppliers you can have AI-based search for nearly anything: electronic, travel booking, finnces, etc.
I guess this is how Siri works / is going to work (no Siri access in my country).

Then, all you need is marketing to call it AI :)

[rlange]

lexipixel wrote:
...Just like like when you google "empire state building" on [duckduckgo.com...]
Hah. And I thought DuckDuckGo was supposed to be Not Evilâ„¢. How could anyone suggest DDG as an alternative to Google when DDG is clearly also subduing the hard work of numerous webmasters without any regard for their livelihood. Oh, the humanity...

Gratuitous sarcasm aside, I had a specific Google query (not the question I typed out previously) in mind that returns, at the top of the page:

Best guess for Empire State Building Height is 381 m
Mentioned on at least 4 websites including wikipedia.org, about.com and pbs.org - Show sources - Feedback

The only problem I have with that is that the value is in meters when Google knows I'm in the United States.

But I think this is getting a bit off-topic, so I'll stop here.

--
Ryan