Google Puts Artificial Intelligence into the Cloud With New API

82 15 Loading
Google-learning-API

Prediction API could be a black box of AI. Data goes in, smart analysis comes out.

Ok Google I have an addendum to your unofficial motto: “Don’t Be Evil…And Don’t Create Skynet”. The Silicon Valley giant has made its machine learning software available via ‘the Cloud‘ – that is, the Internet. (Or at least the distributed network of millions of servers that forms its backbone). The Google Prediction API will allow third party developers to access the machine learning capabilities via other programs, possibly enabling a new generation of smarter, better Apps and websites. This sort of artificial intelligence is narrow in the sense of what it can learn, but it is absurdly general in how it can be applied. Figure out which products your customers are likely to buy, sort incoming emails as friendly or hostile, or determine if a Facebook status update carries newsworthy information. The Google development video below discusses the possibilities of the Prediction API. Looks like cloud-based AI is going to be very useful tool.

As befits a search engine company, Google has to wade through massive amounts of data and find the kernels of really important information. They also need to tailor the search results you receive to your location and preferences. The latter may be the inspiration behind code like the Prediction API, which learns how to predict which data best fits a situation according to examples. The former is probably the basis for Big Query, the other software discussed in the Google Developers video below. Both of these programs have been used inside Google, in various forms, for years. Now they could be available to you.

If you’re lucky that is. Right now the access to Big Query and Prediction API is limited. (You can sign up for Prediction API here and Big Query here.) According to Technology Review, the number of current developers is only in the hundreds. But that’s likely to change after Google finishes its test runs for both programs.

Soon, every software developer may have access to massive data analysis and learning machine code. What will that look like? Well, you’ll have to upload data to Google Storage, then spend some time training the Prediction API to give you the kind of results you want (by feeding it the right examples). After that you’ll be able to call on the API through your own App or website. It’s basically that simple. Working with Big Query would be very similar only with less emphasis on learning and more on filtering big sets of data.

Suddenly your social networking App is better at blocking spammers, or your bank App can better guess which transactions might be the work of an identity thief. This will greatly level the playing field for new companies. Instead of needing millions of dollars to develop your own state of the art learning machine approach to a problem, you can just use Google Prediction API.

Google, in turn, benefits from a learning machine that is constantly tested and improved by large numbers of users. The Prediction API code should be able to find ways of applying the lessons it learns in one application to another. Again, this is narrow artificial intelligence – it’s not going to be able to solve any problem ever, nor will it suddenly become self aware. It should however, become really really good at performing the tasks it is taught.

And that learning code is going to be freely available over the internet, distributed and mirrored thousands of times over so it can’t be lost, growing in processing power as servers and computers are added, and increasing in sophistication over years. This a very powerful situation. As I’ve said before, narrow AI applications like Google Prediction API won’t spontaneously develop into a general (human-like) artificial intelligence but I do think they are laying the groundwork for its creation. Right now this kind of machine learning is relatively simple but if it continues to be developed without major setbacks then in a few decades it could become something much more.

We could probably trust it with anything short of nuclear weapons.

[image credit: Google]
[source: Technology Review, Google Labs]

Discussion — 15 Responses

  • Philip K August 26, 2010 on 11:53 am

    And we count the combined power of google servers (some 20-100 petaflops) and learning based on the whole world’s information, the AI doesn’t seem like a distant future now, but more like a long megaproject. Akin to highways or power stations.

    I wonder how applicable this AI could be for confidential data. If it learns by a confidential sets, then the decisions made could reveal the original data.

  • Philip K August 26, 2010 on 7:53 am

    And we count the combined power of google servers (some 20-100 petaflops) and learning based on the whole world’s information, the AI doesn’t seem like a distant future now, but more like a long megaproject. Akin to highways or power stations.

    I wonder how applicable this AI could be for confidential data. If it learns by a confidential sets, then the decisions made could reveal the original data.

  • LaurentB August 26, 2010 on 3:05 pm
  • LaurentB August 26, 2010 on 11:05 am
  • Joey1058 August 27, 2010 on 1:10 am

    “Don’t Be Evil…And Don’t Create Skynet”. Don’t you think it’s a little late? It’s the old “We’ve got them right where they want us” scenario.

  • Joey1058 August 26, 2010 on 9:10 pm

    “Don’t Be Evil…And Don’t Create Skynet”. Don’t you think it’s a little late? It’s the old “We’ve got them right where they want us” scenario.

  • Nathan Waters August 27, 2010 on 2:13 am

    We recently talked about recommendation engines and where it is taking us this decade: http://hive45.com/shows/episode-28-recommendation-engines/ (ignore the thumbnail, despite it’s hilarity).

  • Nathan Waters August 26, 2010 on 10:13 pm

    We recently talked about recommendation engines and where it is taking us this decade: http://hive45.com/shows/episode-28-recommendation-engines/ (ignore the thumbnail, despite it’s hilarity).

  • Ian Parker August 27, 2010 on 10:16 am

    The motto is from Dumas the 3 musketeers. This could be translated +the language designated by a look up. Google has been able to find quotes like that for yonks.

    Why is GT unable to discriminate genders in Hebrew. http://groups.google.com/group/google-translate-general/browse_thread/thread/d1bdf1df34669018 There are women drivers in Israel?

    Why does the Google team call us all fools when we point this out?

  • Ian Parker August 27, 2010 on 6:16 am

    The motto is from Dumas the 3 musketeers. This could be translated +the language designated by a look up. Google has been able to find quotes like that for yonks.

    Why is GT unable to discriminate genders in Hebrew. http://groups.google.com/group/google-translate-general/browse_thread/thread/d1bdf1df34669018 There are women drivers in Israel?

    Why does the Google team call us all fools when we point this out?

  • Ian Parker August 27, 2010 on 7:48 pm

    There is one addition point that I failed to make first time. That is hype and share price. Google wants the share price and hence the value of the company to remain high. The share price of any high tech company, Google included is boosted by hype.

    My point about Alexander Dumas is apposite for hype. Google wants to convince us that it has Near Human Quality translations. If it is (in essence) using a human translation, the three musketeers has long been translated into English. In fact I know the story and the motto IN ENGLISH.

    Look at https://docs.google.com/Doc?docid=0AQIg8QuzTONQZGZxenF2NnNfNzY4ZDRxcnJ0aHI&hl=en_GB and you will see the REAL quality of GT, not the hype quality.

    I must say Google is also showing itself to be arrogant.

  • Ian Parker August 27, 2010 on 3:48 pm

    There is one addition point that I failed to make first time. That is hype and share price. Google wants the share price and hence the value of the company to remain high. The share price of any high tech company, Google included is boosted by hype.

    My point about Alexander Dumas is apposite for hype. Google wants to convince us that it has Near Human Quality translations. If it is (in essence) using a human translation, the three musketeers has long been translated into English. In fact I know the story and the motto IN ENGLISH.

    Look at https://docs.google.com/Doc?docid=0AQIg8QuzTONQZGZxenF2NnNfNzY4ZDRxcnJ0aHI&hl=en_GB and you will see the REAL quality of GT, not the hype quality.

    I must say Google is also showing itself to be arrogant.