[virgilp]

Where by "worked fine" you mean it couldn't tell a cat from a swan. Nowadays speech to text actually works, translation is actually usable, iPad unlocks with your face etc. Did you put in any effort to remember what it looked like 20 years ago, before saying "it worked fine"?

[scottlocklin]

Speech to text and machine translation worked about as well 20 years ago as it does now. Dunno where you were then. IPad unlocking with your face: what amazing progress. I feel so futuristic. Obviously this great boon to humanity is equivalent to say, those delivered in the decade between 1900 and 1910 (off the top of my head; safety razor, zeppelin, radio, air conditioning, powered heavier than air flight, electronic amplifiers, cellophane, mass produced automobiles, Haber process, talkie movies).

I think I'll take removing the last 20 years of environmental destruction and go back to using a 6 digit pin. Or a thumbprint reader.

FWIIW Dweeb Learning still has problems with image recognition, and people still don't understand how it works. Pretty sure that "progress" was mostly people writing frameworks for video cards.

[virgilp]

> Pretty sure that "progress" was mostly people writing frameworks for video cards.

Pretty sure you either don't know as much as you claim, or are knowingly claiming things that are blatantly false

[scottlocklin]

Cool story bro. Enjoy your iphone face recognition; that's pretty much all you got in the last 20. Seems like the Haber–Bosch process slightly more important.

If you have an actual example of a past 20 years developed machine learning idea which has pushed the needle for normal people in the same way the Haber-Bosch process did (making an extra 6 billion lives possible); I'd love to hear about it!

[Gibbon1]

I have about 50 years of solid experience. As a reference I have my grandmother who lived through most of the 20th century.

There is something hollow about the technological advances of the last 40 years.

[virgilp]

No I don't but I didn't claim that.

Whereas you did claim that speech to text and translation were about as good 20 years ago. Basically, to use your chemistry analogy... you're saying "Haber Bosch was nothing new, Cavendish produced the same kind of results hundreds of years ago!"

[scottlocklin]

Please characterize the amazing improvements the last 20 years have brought to these fields.

Your analogy is faulty: Haber Bosch was obviously something very different and new. Nothing in use today differs appreciably in accuracy or capability in speech to text or machine translation as it was 20 years ago.

[virgilp]

Care to tell me one single speech to text product that had at least a horrible 50% accuracy for Romanian 20 years ago? Because today I can dictate my messages to Google - that, plus the occasional correction, is faster than typing.

[raducu]

Why are you guys even arguing about this?

IF ML works great, yeah it could have a big positive impact, but not much for the average person so far, meanwhile it does work for the chinese dictatorship, hedge funds, the military, the scummy corrupt politicians to influence ellections; self driving cars will be a HUGE benefit, but guess who will take the lion's share of that benefit -- the 0.5% that will own the tech; maybe a little bit for you and I, but the milions of professional drivers will be crushed.

I'm not against tech, but the way it will be use to further re-feudalize the world and promote the accumulation of wealth to the 0.5% -- which will be terrible for the overall progress of humanity.

[scottlocklin]

I guess we're arguing because I deny that there has been appreciable progress. Most of what passes for "progress" in tech and ML is simply adoption of old ideas, or improvements in lithography. Along with forgetting vastly more important ideas. IMO the most important real world idea in ML in current year, defined as "being used and changing how people solve problems" is not deep learning (God Bless Yann), it is Boosting; a 30 year old idea. For that matter, the most important stuff for future progress isn't deep learning either; it's a basket of ideas that hasn't congealed into a name yet (stuff like Volodya Vovk and pals, as well as Gunnar Carlson work on).

All that said; I 100% agree with you, the fruits of supervised learning will entirely accrue to oligarchs, governments and other such dirtbags who have the databases to fit against, do not have the well being of average people in mind. If we notice at all, it will be via blackmail, creepy ad tracking and late night knocks at the door.

[bryanlarsen]

Appreciable progress often comes through the adoption of old ideas. Something that only works in the lab if you squint at it carefully isn't useful; something that only 3 people know about isn't useful; but something that works reliably in the field is useful, even if their are no appreciable differences between them.