5 ms·

Some of the best and most impactful AI research comes from Google DeepMind or Facebook AI research. Those ad clicks are subsidizing a whole lot of scientific ad

by typon 7y ago

Some of the best and most impactful AI research comes from Google DeepMind or Facebook AI research. Those ad clicks are subsidizing a whole lot of scientific advancements.

lanrh1836 7y ago
Google acquired DeepMind...
- SquareWheel 7y ago
  Sure, but Google also acquired Android, Apple acquired Siri, and Microsoft purchased DOS. They were still developed for years under their new companies. At some point, it does become their own project.
  - cameronbrown 7y ago
    To be fair, DeepMind is still fairly isolated from Google.
meowface 7y ago
You're not wrong, but that argument can kind of be used to justify almost anything. "Yeah, what we're doing to make a ton of money right now is skeevy and not adding any actual value the world, maybe even creating negative value, but we promise we're going to spend all of that money on research to advance humanity." Google and Facebook actually did end up doing that, to good effect, but 1) was it worth it? (maybe, especially if we get benevolent superintelligent AGI directly out of it and Google / FB actually agree to not monopolize or abuse it), but 2) how many times should we accept that excuse from companies? If in the next 10 years, a hot up-and-coming new startup begins to grow rapidly and looks like it'll end up in the FAANG tier, should we also trust them when they say their questionable tactics are worth it because they're gonna spend that money on cool science stuff?
- typon 7y ago
  What's the alternative? Increase government R&D spend to the same levels as FAANG companies? But will they have enough infrastructure to support that? For example, Tensorflow runs on Google's global datacenter infrastructure - this is the only way these latest monster transformer models can be trained, using thousands of GPUs or TPUs.
  - meowface 7y ago
    The alternative would be to create a tech or hard science company which doesn't choose advertising as their core business model. Many large and successful tech companies already do this, and are contributing a lot to ML research and other fields. It's just that Google, Facebook/Instagram, and Twitter, some of the largest companies in the space, are inextricably tied to that model. Apple and Netflix don't share the same problem, and that gives them a lot more freedom in other areas, too (but perhaps resulting in them having less total user acquisition potential than Google or Facebook; but of course it's needed less, since every customer is directly giving them money). Uber also doesn't share that problem, and they've contributed quite a bit to the community, though of course they have other problems. The main problem isn't tech companies or monopolies, IMO; it's the tech companies that survive only by cannibalizing both non-paying users and non-paying non-users who happen to visit just about any website on the Internet (because odds are any given website is using Google Analytics or has some sort of Facebook or Twitter integration). It creates bad incentives. Of course, there are also separate debates to be had over general issues of monopolization and working conditions (like Amazon fulfillment workers not being able to take bathroom breaks without risking accumulating points which may result in them getting fired), and control over different media platforms (like Google owning the only real video platform in the world and Twitter owning the only real microblogging / "town square" platform, and issues that may come from how they regulate users and content). The advertising stuff is just the cherry on top. I do think trying to regulate or break them up over the content policing stuff would be a huge mistake, though. Staunch conservatives say they want them to relax their standards and stop being biased against conservatives; staunch liberals say they helped the spread of disinformation, racism, and fascism and should have tighter standards. I think if either side has their way, to any degree, it will be a disaster.
samsonradu 7y ago
The broader theory of Peter Thiel is that most of these advancements have been done in the rather unregulated ‘world of bits’, while areas related to physics, chemistry etc. have seen less advancement. Playing Go against an AI opponent while in a 1970s, crumbling, subway system would make for a good analogy.
buboard 7y ago
It wasn't "basic" research though, as the basic research in AI happened a few decades ago. Google is industrializing Hinton's work, but hinton created most of it in toronto. That said, industry can definitely do basic research work (e.g. Pharma does). But what google is doing is acquiring any company that might be a real threat to them.
- WhitneyLand 7y ago
  Are you suggesting Google doesn't have real academic style research groups? Surely you jest. Didn't they even crossover into biology on one paper? They're a profit minded entity yes, but no way is it all shaving nanoseconds off of click throughs.
  - buboard 7y ago
    Sure they have, and they even do basic research. But they didn't make "the most impactful" basic research in ANNs.

by typon 7y ago

Some of the best and most impactful AI research comes from Google DeepMind or Facebook AI research. Those ad clicks are subsidizing a whole lot of scientific advancements.

lanrh1836 7y ago

Google acquired DeepMind...

SquareWheel 7y ago
Sure, but Google also acquired Android, Apple acquired Siri, and Microsoft purchased DOS. They were still developed for years under their new companies. At some point, it does become their own project.
- cameronbrown 7y ago
  To be fair, DeepMind is still fairly isolated from Google.

meowface 7y ago

You're not wrong, but that argument can kind of be used to justify almost anything. "Yeah, what we're doing to make a ton of money right now is skeevy and not adding any actual value the world, maybe even creating negative value, but we promise we're going to spend all of that money on research to advance humanity." Google and Facebook actually did end up doing that, to good effect, but 1) was it worth it? (maybe, especially if we get benevolent superintelligent AGI directly out of it and Google / FB actually agree to not monopolize or abuse it), but 2) how many times should we accept that excuse from companies? If in the next 10 years, a hot up-and-coming new startup begins to grow rapidly and looks like it'll end up in the FAANG tier, should we also trust them when they say their questionable tactics are worth it because they're gonna spend that money on cool science stuff?

typon 7y ago
What's the alternative? Increase government R&D spend to the same levels as FAANG companies? But will they have enough infrastructure to support that? For example, Tensorflow runs on Google's global datacenter infrastructure - this is the only way these latest monster transformer models can be trained, using thousands of GPUs or TPUs.
- meowface 7y ago
  The alternative would be to create a tech or hard science company which doesn't choose advertising as their core business model. Many large and successful tech companies already do this, and are contributing a lot to ML research and other fields. It's just that Google, Facebook/Instagram, and Twitter, some of the largest companies in the space, are inextricably tied to that model. Apple and Netflix don't share the same problem, and that gives them a lot more freedom in other areas, too (but perhaps resulting in them having less total user acquisition potential than Google or Facebook; but of course it's needed less, since every customer is directly giving them money). Uber also doesn't share that problem, and they've contributed quite a bit to the community, though of course they have other problems. The main problem isn't tech companies or monopolies, IMO; it's the tech companies that survive only by cannibalizing both non-paying users and non-paying non-users who happen to visit just about any website on the Internet (because odds are any given website is using Google Analytics or has some sort of Facebook or Twitter integration). It creates bad incentives. Of course, there are also separate debates to be had over general issues of monopolization and working conditions (like Amazon fulfillment workers not being able to take bathroom breaks without risking accumulating points which may result in them getting fired), and control over different media platforms (like Google owning the only real video platform in the world and Twitter owning the only real microblogging / "town square" platform, and issues that may come from how they regulate users and content). The advertising stuff is just the cherry on top. I do think trying to regulate or break them up over the content policing stuff would be a huge mistake, though. Staunch conservatives say they want them to relax their standards and stop being biased against conservatives; staunch liberals say they helped the spread of disinformation, racism, and fascism and should have tighter standards. I think if either side has their way, to any degree, it will be a disaster.

samsonradu 7y ago

The broader theory of Peter Thiel is that most of these advancements have been done in the rather unregulated ‘world of bits’, while areas related to physics, chemistry etc. have seen less advancement. Playing Go against an AI opponent while in a 1970s, crumbling, subway system would make for a good analogy.

buboard 7y ago

It wasn't "basic" research though, as the basic research in AI happened a few decades ago. Google is industrializing Hinton's work, but hinton created most of it in toronto. That said, industry can definitely do basic research work (e.g. Pharma does). But what google is doing is acquiring any company that might be a real threat to them.

WhitneyLand 7y ago
Are you suggesting Google doesn't have real academic style research groups? Surely you jest. Didn't they even crossover into biology on one paper? They're a profit minded entity yes, but no way is it all shaving nanoseconds off of click throughs.
- buboard 7y ago
  Sure they have, and they even do basic research. But they didn't make "the most impactful" basic research in ANNs.