The Tao of Gaming

Boardgames and lesser pursuits

Keldon, Google, Evolution and Language

I have — with no malice aforethought — been near some great inventions. While I’m not an expert I think this year has been a simply stunning technically and after reading the NYT profile on Google Translate I can’t wonder if it is the inflection point. (Not in a “SF Singularity” way, but something softer).

Ever since Keldon I’ve appreciated that neural nets can handle more than you expect — after all, we are neural nets and sometimes clever. But seeing some of these translations and the quality involved is still stunning.

Anyway — its worth a read.

 

Advertisements

Written by taogaming

December 20, 2016 at 9:26 pm

Posted in Artificial Opponents, Non-Gaming

Tagged with

2 Responses

Subscribe to comments with RSS.

  1. NLP though is still extremely primitive – it seems to me really no further along than it was 10, 15 years ago, we’re just throwing more computing power at it. We’re good at problems where we can bring huge amounts of data to bear, but haven’t moved the needle much on the harder problems. Even big data problems where the learning algorithms are non-trivial seem elusive, we don’t seem to have been able to usefully deploy big data against problems in health care or education for example, wher there are potentially huge wins.

    I’m actually bearish right now on whether software will improve our quality of life over the next 10 years. We’re starting to see the first really large negative social impacts from algorithms and software (fake news, political siloing, use of big data to segregate disfavored classes of people politically or economically); additionally the tech industry’s big ideas recently have all been targeting and degrading blue-collar jobs – Über and Air BnB being the prime examples of course. If tech is moving downscale, crushing lower-wage jobs while employing tiny numbers of people themselves, instead of creating opportunities for everyone, we could be on a path to serious trouble.

    Chris Farrell

    December 21, 2016 at 2:40 am

  2. > NLP though is still extremely primitive – it seems to me really no further along than it was 10, 15 years ago, we’re just throwing more computing power at it.

    Well, quantity has a quality all its own. I don’t think the neural nets are much different than the ones I learned about, but the qualitative value appears (to this monolinguistic soul) quite improved. OCR is similar, its always seemed to me to be “about five years” from being ready, but my employer is investing heavily in it. The error rates are there, but cutting humans out of the loop for 98% is now cost-effective.

    > I’m actually bearish right now on whether software will improve our quality of life over the next 10 years.

    Well, an inflection point can go either way. I personally think the Internet is re-creating the Reformation. Suddenly people who thought “Eh, he thinks mostly like me, just a few differences” realize that they think nothing alike. Just like Gutenburg revealed differences, so does the internet. Social media has exacerbated the trend. I do think we’ll see mass changes in the next two decades. 2 million drivers out of a job (if Google cars and drone deliveries work). Lord knows how many knowledge workers out of a job. (Radiologists are an example the article mentions).

    But … if a computer can do radiology better, the price should come way down. It’s like the thought experiment of a $10 pill that cures everything and lets you live to be 120. Great for humanity, but terrible for a lot of people on the way. ( https://medium.com/@russroberts/the-human-side-of-trade-7b8e024e7536#.mzhgo5joj )

    taogaming

    December 21, 2016 at 5:49 pm


Comments are closed.

%d bloggers like this: