Another example of the Google mission to build ‘Skynet’ (with the military).
“While I am proud of what has been accomplished in search over the past decade, there are important areas in which I wish we had made more progress. Perfect search requires human-level artificial intelligence, which many of us believe is still quite distant. However, I think it will soon be possible to have a search engine that “understands” more of the queries and documents than we do today. Others claim to have accomplished this, and Google’s systems have more smarts behind the curtains than may be apparent from the outside, but the field as a whole is still shy of where I would have expected it to be.”
I sense him being dishonest in that last bold highlighted bit. Surely, he of all people is familiar with the Law of Accelerating Returns (LoAR). He said that in 10 years “computers will be 100 times faster still, and storage will be 100 times cheaper”. LoAR concludes that in 10 years computing power would exponentially increase well over 1,500 times, while data storage would be even far greater.
That’s following a doubling every 18 months as has happened over the past four decades, while not even contemplating revolutionary computing advances such as quantum processors and optical circuits. Call it a hunch, but I’m quite sure that Brin is well aware of that.
And so on:
The past decade has seen tremendous changes in computing power amplified by the continued growth of Google’s data centers. It has enabled the growth and processing of increasingly large data sets such as the web, the world’s books, and video. This in turn has allowed problems once considered to be in the fantasy realm of artificial intelligence to come closer to reality.
Google Translate supports automatic machine translation between 1640 language pairs. This is made possible by large computer clusters and vast repositories of monolingual and multilingual texts: http://www.google.com/intl/en/help/faq_translation.html. This technology also allows us to support translated search where the query gets translated to another language and the results get translated back.
While the earliest Google Voice Search ran as a crude demo in 2001, today our own speech recognition technology powers GOOG411, the voice search feature of the Google Mobile App, and Google Voice. It, too, takes advantage of large training sets and significant computing capability. Last year, PicasaWeb, our photo hosting site, released face recognition, bringing a technology that is on the cutting edge of computer science to a consumer web service.
Just a few months ago we released Google Flu Trends, a service that uses our logs data (without revealing personally identifiable information) to predict flu incidence weeks ahead of estimates by the Centers for Disease Control (CDC). It is amazing how an existing data set typically used for improving search quality can be brought to bear on a seemingly unrelated issue and can help to save lives. I believe this sort of approach can do even more — going beyond monitoring to inferring potential causes and cures of disease. This is just one example of how large data sets such as search logs coupled with powerful data mining can improve the world while safe guarding privacy.
Some other neat quotes I’m just now noticing in their documents:
“We strive to hire the best computer scientists and engineers to help us solve very significant challenges across systems design, artificial intelligence, machien learning, data mining networking, software engineering, testing, distributed systems, cluster design and other areas.”
“I mentioned earlier how we are striving to make Google really understand your query and all the information in the world. To do that, we will have to make Google smart, and that requires artificial intelligence. We are particular believers in large-scale AI that involves both a lot of computation and a lot of data. We’re looking to build the best center for this kind of work in the world.”
“Search is a really hard problem. To do a perfect job, you would need to understand all the world’s information, and the precise meaning of every query. With all that understanding, you would then have to produce the perfect answer instantly. We are making significant progress, but remain a long way from perfection.”
Thus, computer systems will have greater opportunity to learn from the collective behavior of billions of humans. They will get smarter, gleaning relationships between objects, nuances, intentions, meanings, and other deep conceptual information. Today’s Google search uses an early form of this approach, but in the future many more systems will be able to benefit from it.
Traditionally, systems that solve complicated problems and queries have been called “intelligent”, but compared to earlier approaches in the field of ‘artificial intelligence’, the path that we foresee has important new elements. First of all, this system will operate on an enormous scale with an unprecedented computational power of millions of computers. It will be used by billions of people and learn from an aggregate of potentially trillions of meaningful interactions per day. It will be engineered iteratively, based on a feedback loop of quick changes, evaluation, and adjustments. And it will be built based on the needs of solving and improving concrete and useful tasks such as finding information, answering questions, performing spoken dialogue, translating text and speech, understanding images and videos, and other tasks as yet undefined. When combined with the creativity, knowledge, and drive inherent in people, this “intelligent cloud” will generate many surprising and significant benefits to mankind.