Rooms Full of People
September 28, 2020 12:09 PM Subscribe
Is Palantir's Crystal Ball Just Smoke and Mirrors? Peter Thiel-backed surveillance giant Palantir Technologies (previously) is set to go public September 30. Long controversial for its secrecy and involvement with the more unsavory parts of the national security state (e.g., ICE, CIA, NSA), Palantir is under scrutiny for its financial woes -- it posted a $600 million loss in 2018 and in 2019 -- and for whether its product even works as advertised. Palantir portrays its software as like its namesake — a crystal ball you gaze into for answers... But the truth is that it still appears to take a lot of manual labor to make it work, and there’s nothing magical about that.
Vampire Boi over-promised and under-delivered. Huh. Imagine that.
posted by SPrintF at 12:27 PM on September 28, 2020 [7 favorites]
posted by SPrintF at 12:27 PM on September 28, 2020 [7 favorites]
Palantir was founded in 2003 which was near one of the nadirs of fashionability for AI. This was a time when people who started their careers in AI research were desperately trying to rebrand and claim they were doing anything but artificial intelligence.
When Palantir was founded, they made a big public deal out of the fact that they were not doing AI, because as everyone in Silicon Valley knew and agreed, AI doesn't work, but rather making an interface for human users to aggregate data and make judgments. They may have toned down this messaging but I don't think they have ever claimed to be an AI company. It was always intended to be Rooms Full of People as the end users -- this was a major selling point.
What's been suspected for a long time is that they understate the extent to which getting the software to work, for a new company or department, also requires Rooms Full of People from Palantir. The idea of selling software is that you make the software once and sell it -- but it seems that there may be a lot of development effort required for each new contract.
posted by vogon_poet at 12:34 PM on September 28, 2020 [8 favorites]
When Palantir was founded, they made a big public deal out of the fact that they were not doing AI, because as everyone in Silicon Valley knew and agreed, AI doesn't work, but rather making an interface for human users to aggregate data and make judgments. They may have toned down this messaging but I don't think they have ever claimed to be an AI company. It was always intended to be Rooms Full of People as the end users -- this was a major selling point.
What's been suspected for a long time is that they understate the extent to which getting the software to work, for a new company or department, also requires Rooms Full of People from Palantir. The idea of selling software is that you make the software once and sell it -- but it seems that there may be a lot of development effort required for each new contract.
posted by vogon_poet at 12:34 PM on September 28, 2020 [8 favorites]
*sets up altars to every conceivable deity* Please let this be the next Theranos. PLEASE.
posted by Gin and Broadband at 12:38 PM on September 28, 2020 [14 favorites]
posted by Gin and Broadband at 12:38 PM on September 28, 2020 [14 favorites]
I'm sensing a very distinct whiff of an exception to Betteridge's Law.
posted by flabdablet at 12:40 PM on September 28, 2020 [4 favorites]
posted by flabdablet at 12:40 PM on September 28, 2020 [4 favorites]
The thing is... horribly invasive technologies don't have to actually work, they just have to have the appearance of working.
See: Breathalyzers, how almost none of them have open source code, and how when that source code eventually gets scrutinized, it's some of the worst code imaginable, without error correction or basic comments to describe the code and just generally doesn't follow best coding practices.
See also: AI video monitoring systems that can see "faces" except black people's faces, or that just produce numerous false-positives when dealing with dark skin tones. Those exist and are widely used, despite all the evidence that the technology is biased.
Does it work?
Well, all it really needs to be able to do to "work" is allow the rich to continue to subjugate the poor, privacy laws be damned. So, it probably works.
posted by deadaluspark at 12:46 PM on September 28, 2020 [8 favorites]
See: Breathalyzers, how almost none of them have open source code, and how when that source code eventually gets scrutinized, it's some of the worst code imaginable, without error correction or basic comments to describe the code and just generally doesn't follow best coding practices.
See also: AI video monitoring systems that can see "faces" except black people's faces, or that just produce numerous false-positives when dealing with dark skin tones. Those exist and are widely used, despite all the evidence that the technology is biased.
Does it work?
Well, all it really needs to be able to do to "work" is allow the rich to continue to subjugate the poor, privacy laws be damned. So, it probably works.
posted by deadaluspark at 12:46 PM on September 28, 2020 [8 favorites]
FWIW, there have been rumors in the industry about Palantir's tech being less than it appears for a while; at least a few years, probably longer. This is the first time I've seen an exposé in the mainstream press, though.
Their stuff demos really well, and they use a lot of buzzwords to try and describe how it works, but... there are questions, which they don't seem to want to answer. Mostly about the backend. It always seemed to me that there were two explanations: either they're geniuses, and have made some really astonishing advancements in AI/ML that nobody else has, or they're pulling a Wizard (as in "they're off to see the...") and stretching a really nice UI over a ton of manual analysis, and chalking up that human effort to their software.
That's not to say that building a good UI for extremely complex datasets is trivial, quite the opposite. But if that's what their product does, they should at least be upfront about it. But there's at least a chance that their 'secret sauce' is really a bunch of 23-year-olds getting about $70k a year as "Operations Analysts" doing the heavy lifting, not some amazing algo.
The importance is in the scalability: hiring a shitload of analysts to sift through data and massage it into a format that you can display with your neat UI is fine, if you're the US government and have near-bottomless pockets for that sort of thing. But you can't Ctrl-C/Ctrl-V that success like you could if it was software. They are, if this theory holds, essentially a services business—not a technology company.
posted by Kadin2048 at 12:49 PM on September 28, 2020 [9 favorites]
Their stuff demos really well, and they use a lot of buzzwords to try and describe how it works, but... there are questions, which they don't seem to want to answer. Mostly about the backend. It always seemed to me that there were two explanations: either they're geniuses, and have made some really astonishing advancements in AI/ML that nobody else has, or they're pulling a Wizard (as in "they're off to see the...") and stretching a really nice UI over a ton of manual analysis, and chalking up that human effort to their software.
That's not to say that building a good UI for extremely complex datasets is trivial, quite the opposite. But if that's what their product does, they should at least be upfront about it. But there's at least a chance that their 'secret sauce' is really a bunch of 23-year-olds getting about $70k a year as "Operations Analysts" doing the heavy lifting, not some amazing algo.
The importance is in the scalability: hiring a shitload of analysts to sift through data and massage it into a format that you can display with your neat UI is fine, if you're the US government and have near-bottomless pockets for that sort of thing. But you can't Ctrl-C/Ctrl-V that success like you could if it was software. They are, if this theory holds, essentially a services business—not a technology company.
posted by Kadin2048 at 12:49 PM on September 28, 2020 [9 favorites]
It's not going to be the next Theranos or Enron. The software is real and it works and is apparently useful. From all reports it is a nice user interface on top of complicated databases. It's also apparently overpriced and apparently expensive to develop and sell, but it's not a fake product.
posted by vogon_poet at 12:51 PM on September 28, 2020 [2 favorites]
posted by vogon_poet at 12:51 PM on September 28, 2020 [2 favorites]
It's also apparently overpriced and apparently expensive to develop
Those two assertions clash rather brightly.
posted by chavenet at 1:06 PM on September 28, 2020 [1 favorite]
Those two assertions clash rather brightly.
posted by chavenet at 1:06 PM on September 28, 2020 [1 favorite]
"The thing is... horribly invasive technologies don't have to actually work, they just have to have the appearance of working."
Jill Lepore makes this point very convincingly on a recent Talking Politics podcast (and presumably too in the book she's talking about) -- highly reccomended for a thought provoking look at the psychology of this kind of thing.
posted by tomp at 1:10 PM on September 28, 2020 [5 favorites]
Jill Lepore makes this point very convincingly on a recent Talking Politics podcast (and presumably too in the book she's talking about) -- highly reccomended for a thought provoking look at the psychology of this kind of thing.
posted by tomp at 1:10 PM on September 28, 2020 [5 favorites]
Palantir is a 17-year-old company that loses $600 million a year. This isn't some B2B start-up with millions of customers and hockey-stick SAAS growth potential. This company has literally 125 customers, with just 3 customers making up 30% of revenue.
This is a bad business.
But Palantir's isn't *just* a bad business, it's also a bad business with leadership that's totally tied to Trump.
The timing of this IPO is gross.
posted by a complicated history at 1:19 PM on September 28, 2020 [11 favorites]
This is a bad business.
But Palantir's isn't *just* a bad business, it's also a bad business with leadership that's totally tied to Trump.
The timing of this IPO is gross.
posted by a complicated history at 1:19 PM on September 28, 2020 [11 favorites]
> This isn't some B2B start-up with millions of customers and hockey-stick SAAS growth potential. This company has literally 125 customers, with just 3 customers making up 30% of revenue.
I worked for a while on a B2B product/service that had maybe a couple dozen corporate customers at most. The biggest event in the year-plus that I worked there was the one time they had added one new customer. There were more people involved in developing the product than there were end-users. It was one of the most profitable divisions of its generally-profitable parent company, because it had secured a niche and was highly functional in that niche. It wasn't even the most expensive product/service among its competitors.
Everything about Palantir is bad, and I'm wrestling with myself over whether it would be worse for Palantir to be an outright fraud or legitimate. I hope it crashes hard and doesn't take too many bystanders with it. But what I learned on that job I mentioned is that in the corporate world it can be hard to tell how something is making money if all you have is the publicly available information. 125 customers can represent an absolute shit-ton of income... or not.
posted by at by at 1:48 PM on September 28, 2020 [6 favorites]
I worked for a while on a B2B product/service that had maybe a couple dozen corporate customers at most. The biggest event in the year-plus that I worked there was the one time they had added one new customer. There were more people involved in developing the product than there were end-users. It was one of the most profitable divisions of its generally-profitable parent company, because it had secured a niche and was highly functional in that niche. It wasn't even the most expensive product/service among its competitors.
Everything about Palantir is bad, and I'm wrestling with myself over whether it would be worse for Palantir to be an outright fraud or legitimate. I hope it crashes hard and doesn't take too many bystanders with it. But what I learned on that job I mentioned is that in the corporate world it can be hard to tell how something is making money if all you have is the publicly available information. 125 customers can represent an absolute shit-ton of income... or not.
posted by at by at 1:48 PM on September 28, 2020 [6 favorites]
My google-fu is failing me, but does anyone else remember someone hacking either Thiel or Alex Karp and making their WoW handle public?
posted by The Ardship of Cambry at 1:51 PM on September 28, 2020
posted by The Ardship of Cambry at 1:51 PM on September 28, 2020
It's also apparently overpriced and apparently expensive to develop
Those two assertions clash rather brightly.
Not necessarily, they could both (1) have chosen a methodology that is expensive to do/maintain (a room full of people doing all the heavy lifting instead of a computer) and (2) stack up poorly with other competitors vis-a-vis price [as a consequence of (1)].
posted by axiom at 1:54 PM on September 28, 2020 [3 favorites]
Those two assertions clash rather brightly.
Not necessarily, they could both (1) have chosen a methodology that is expensive to do/maintain (a room full of people doing all the heavy lifting instead of a computer) and (2) stack up poorly with other competitors vis-a-vis price [as a consequence of (1)].
posted by axiom at 1:54 PM on September 28, 2020 [3 favorites]
The next Theranos is Nikola, although they aren't nearly as high profile.
With near zero interest rates and the money printer still running (although less than earlier this year), this is very much the year for this kind of company to go public, but they're a few months late for a real good pump and dump.
posted by MillMan at 1:54 PM on September 28, 2020
With near zero interest rates and the money printer still running (although less than earlier this year), this is very much the year for this kind of company to go public, but they're a few months late for a real good pump and dump.
posted by MillMan at 1:54 PM on September 28, 2020
So this doesn't seem much better or worse than most software competing in this space. "A sleek product, but not necessarily commercially viable" is pretty normal except for the Thiel involvement and its creepy fascination with intelligence and surveillance*. And as for sleek presentations that elide the underlying work with the sleek UI, well, I swear I spent 25% of the 2010s in meetings trying to get vendors to say in plain English what kind of data collecting & preparation was needed so a gung-ho executive would stop expecting magic.
I was actually supposed to attend a meeting with Palantir for biotech data, of all things, at one point--around 2011 or 2012. Someone knew someone or something, they were "geniuses," they had a nice UI. I was glad it got dropped on its own for once. It would have made sense in some ways if they were really that much better than everyone else, but I think pharma probably has more experience and customers with this sort of thing and I suspect was less impressed. Never heard anyone else working with them on that.
* OT, but Australian economist John Quiggin prefers the word "propertarian" for self-described libertarians like Thiel, because the driving motivation for the modern breed is use of state power to guarantee property and property rights only. Actual freedoms and a protections for individual citizens' rights outside property don't matter. It's a better word and should be used.
posted by mark k at 2:35 PM on September 28, 2020 [11 favorites]
I was actually supposed to attend a meeting with Palantir for biotech data, of all things, at one point--around 2011 or 2012. Someone knew someone or something, they were "geniuses," they had a nice UI. I was glad it got dropped on its own for once. It would have made sense in some ways if they were really that much better than everyone else, but I think pharma probably has more experience and customers with this sort of thing and I suspect was less impressed. Never heard anyone else working with them on that.
* OT, but Australian economist John Quiggin prefers the word "propertarian" for self-described libertarians like Thiel, because the driving motivation for the modern breed is use of state power to guarantee property and property rights only. Actual freedoms and a protections for individual citizens' rights outside property don't matter. It's a better word and should be used.
posted by mark k at 2:35 PM on September 28, 2020 [11 favorites]
So look Gotham and Metropolis are two *very* different types of software packages, and Foundry is really where they are making their in-roads into companies. I'll be honest, Gotham isn't the scary software package because ostensibly, it is contained to one government. I'll be honest, Foundry is the scary one - because it it *good* and sort of useful - as in - I could see companies wanting to use it.
And it isn't just, what someone knows that is important, but how they know it, how they learn it, and how they act upon said information. And Thiel's platform seems very willing to collect and aggregate that informationabout I mean FOR their corporate customers. But, then I come back to the looking glass, and I realize it is being held by Theil, and he is funneling all that data through his black box, and he is seems very willing to hand whatever in that black box over as a product for Gotham.
posted by Nanukthedog at 2:37 PM on September 28, 2020
And it isn't just, what someone knows that is important, but how they know it, how they learn it, and how they act upon said information. And Thiel's platform seems very willing to collect and aggregate that information
posted by Nanukthedog at 2:37 PM on September 28, 2020
The timing of this IPO is gross.
Perhaps only one of many pump-and-dump schemes since November 2016. Maybe Thiel sees the writing on the wall and is bailing out while there's still some money left to grift.
posted by They sucked his brains out! at 2:41 PM on September 28, 2020
Perhaps only one of many pump-and-dump schemes since November 2016. Maybe Thiel sees the writing on the wall and is bailing out while there's still some money left to grift.
posted by They sucked his brains out! at 2:41 PM on September 28, 2020
If I’m right — if, in fact, Palantir is loved in the way the Kardashians are loved — well, the Kardashians are not going to be famous forever.
Horrible analogy. Whomever does the Kardashian PR and marketing is a pure genius. Kendall Jenner is worth over a billion and they managed to make Travis Scott (Kendall Jenner's boyfriend) a name overnight with the first marketing deal with McDonald's. When was the last time McDonald's and "successful viral marketing campaign" was said in the same breath? Kardashians are not going to be famous forever and Google will not be the search engine of choice forever.
The article rightfully focuses on the "room full of people" which in my industry is usually couched in the term off-shore. The problem is that there's a belief, a hope, that an engineer making $250k might come up with an algorithm or concept that might scale and with the same amount of money invested in labeling or whatever in an off-shore scenario you won't get that. I understand the philosophy, it is like optionality. It never works out in reality.
posted by geoff. at 2:50 PM on September 28, 2020 [2 favorites]
Horrible analogy. Whomever does the Kardashian PR and marketing is a pure genius. Kendall Jenner is worth over a billion and they managed to make Travis Scott (Kendall Jenner's boyfriend) a name overnight with the first marketing deal with McDonald's. When was the last time McDonald's and "successful viral marketing campaign" was said in the same breath? Kardashians are not going to be famous forever and Google will not be the search engine of choice forever.
The article rightfully focuses on the "room full of people" which in my industry is usually couched in the term off-shore. The problem is that there's a belief, a hope, that an engineer making $250k might come up with an algorithm or concept that might scale and with the same amount of money invested in labeling or whatever in an off-shore scenario you won't get that. I understand the philosophy, it is like optionality. It never works out in reality.
posted by geoff. at 2:50 PM on September 28, 2020 [2 favorites]
Anyway, I'm not sure which makes me more nervous, Palantir or this kind of stuff that Palmer Luckey's Anduril is marketing to corporations and governments.
Honestly though, I'm sure the marketing still over-hypes capabilities, as is the nature of most marketing.
Also, it's thoughtful to note this is also a LOTR name
posted by deadaluspark at 2:53 PM on September 28, 2020
Honestly though, I'm sure the marketing still over-hypes capabilities, as is the nature of most marketing.
Also, it's thoughtful to note this is also a LOTR name
posted by deadaluspark at 2:53 PM on September 28, 2020
Palantir or this vimeo kind of stuff vimeo that Palmer Luckey's Anduril is marketing to corporations and governments.
One has a lattice AI core. DO YOU KNOW WHAT YOU CAN DO WITH A LATTICE AI CORE?!
posted by geoff. at 3:04 PM on September 28, 2020
One has a lattice AI core. DO YOU KNOW WHAT YOU CAN DO WITH A LATTICE AI CORE?!
posted by geoff. at 3:04 PM on September 28, 2020
Wow, rare reverse Betteridge.
posted by RobotHero at 3:06 PM on September 28, 2020 [1 favorite]
posted by RobotHero at 3:06 PM on September 28, 2020 [1 favorite]
The S-1 was apparently written by Khan Noonien Singh:
The engineering elite of Silicon Valley may know more than most about building software. But they do not know more about how society should be organized or what justice requires.
posted by RobotVoodooPower at 3:11 PM on September 28, 2020 [4 favorites]
The engineering elite of Silicon Valley may know more than most about building software. But they do not know more about how society should be organized or what justice requires.
posted by RobotVoodooPower at 3:11 PM on September 28, 2020 [4 favorites]
So it's labour-washing/mechanical turkery, much like virtually every other form of “AI” beyond puppyslugs?
posted by acb at 3:25 PM on September 28, 2020 [1 favorite]
posted by acb at 3:25 PM on September 28, 2020 [1 favorite]
>The idea of selling software is that you make the software once and sell it -- but it seems that there may be a lot of development effort required for each new contract.
Well, yes, zero marginal cost. But also, no. We're now learning that businesses need to change to make the most of their tech, and that means training, consultancy and more training.
Think of it like this: you bring on the future with your new software and it's not evenly distributed until you've sold more copies and it becomes normal baseline. If your product leaps too far into the future, you have to train each new cohort of users so they understand enough about your system to gain from it. Paradigm shifts and disruption -- and failing business models -- come from that leap too far ahead.
posted by k3ninho at 3:39 PM on September 28, 2020 [1 favorite]
Well, yes, zero marginal cost. But also, no. We're now learning that businesses need to change to make the most of their tech, and that means training, consultancy and more training.
Think of it like this: you bring on the future with your new software and it's not evenly distributed until you've sold more copies and it becomes normal baseline. If your product leaps too far into the future, you have to train each new cohort of users so they understand enough about your system to gain from it. Paradigm shifts and disruption -- and failing business models -- come from that leap too far ahead.
posted by k3ninho at 3:39 PM on September 28, 2020 [1 favorite]
I don’t know what you can do with a Lattice AI core. Why is that a big deal?
posted by mmcg at 5:57 PM on September 28, 2020
posted by mmcg at 5:57 PM on September 28, 2020
I think the big issue is that whether it works or not, or how badly the company is run, is sort of moot. They've put the big idea out there, and at some point, they'll either figure out how to do it well (bad outcome), or they'll go broke trying, and someone else will build on what they failed to do, and release a better version that works (worse outcome).
Aside from everything, from kitschy naming things after Tolkien things, destroying Gawker, to just being a genuinely repugnant human, I hope a long, long life on Thiel, broke, and removed from any concept of influence, with no one knowing his name.
posted by Ghidorah at 8:50 PM on September 28, 2020 [3 favorites]
Aside from everything, from kitschy naming things after Tolkien things, destroying Gawker, to just being a genuinely repugnant human, I hope a long, long life on Thiel, broke, and removed from any concept of influence, with no one knowing his name.
posted by Ghidorah at 8:50 PM on September 28, 2020 [3 favorites]
I don’t know what you can do with a Lattice AI core. Why is that a big deal?
It is just a stupid marketing slogan they made sound like a thing. Lattice is an FPGA maker. From my experience FPGA chips usually perform more or less similar as general purpose CPUs with much less power consumption so perfect for drones. But they're not promoting that, they're promoting nebulous AI and a core.
Keep in mind with this article, and to its credit the article puts it out there but doesn't really make it the thesis, is that Palantir is an outsider pushing out a system that was inferior in every way. No one is saying the intelligence system had a better system or was the same smoke and mirrors unless I missed a large portion of the article. As the article notes even a very, very well funded outsider like Palantir had a hard time penetrating the military industrial complex. That's a real issue as there's some really large contracts floating around uncontested.
I'm not surprised that software developed for a military contract fails commercially as the military has some highly specific needs that aren't scalable in the way that makes it a commercial success.
posted by geoff. at 9:47 PM on September 28, 2020 [2 favorites]
It is just a stupid marketing slogan they made sound like a thing. Lattice is an FPGA maker. From my experience FPGA chips usually perform more or less similar as general purpose CPUs with much less power consumption so perfect for drones. But they're not promoting that, they're promoting nebulous AI and a core.
Keep in mind with this article, and to its credit the article puts it out there but doesn't really make it the thesis, is that Palantir is an outsider pushing out a system that was inferior in every way. No one is saying the intelligence system had a better system or was the same smoke and mirrors unless I missed a large portion of the article. As the article notes even a very, very well funded outsider like Palantir had a hard time penetrating the military industrial complex. That's a real issue as there's some really large contracts floating around uncontested.
I'm not surprised that software developed for a military contract fails commercially as the military has some highly specific needs that aren't scalable in the way that makes it a commercial success.
posted by geoff. at 9:47 PM on September 28, 2020 [2 favorites]
So it's going to be like SAP, supposedly a drop in business ERP system, but one that now supports a multi-billion $ industry of consultants and customising.
posted by PenDevil at 2:05 AM on September 29, 2020 [8 favorites]
posted by PenDevil at 2:05 AM on September 29, 2020 [8 favorites]
What's been suspected for a long time is that they understate the extent to which getting the software to work, for a new company or department, also requires Rooms Full of People from Palantir. The idea of selling software is that you make the software once and sell it -- but it seems that there may be a lot of development effort required for each new contract.
That is *one* model for software. Another is that you are the only one who can provide the required Rooms Full of People, to Customers Full of Budget.
From another lens, realize - we have Rooms Full Of People. A lot of society is organized around figuring out what the heck to do with all of them.
posted by effugas at 4:56 AM on September 29, 2020
That is *one* model for software. Another is that you are the only one who can provide the required Rooms Full of People, to Customers Full of Budget.
From another lens, realize - we have Rooms Full Of People. A lot of society is organized around figuring out what the heck to do with all of them.
posted by effugas at 4:56 AM on September 29, 2020
The engineering elite of Silicon Valley may know more than most about building software. But they do not know more about how society should be organized or what justice requires.
Here, I'll translate this into other terms. Self driving trucks end about 16% of American jobs. What should those workers do?
posted by effugas at 5:00 AM on September 29, 2020 [2 favorites]
Here, I'll translate this into other terms. Self driving trucks end about 16% of American jobs. What should those workers do?
posted by effugas at 5:00 AM on September 29, 2020 [2 favorites]
geoff. -- Got it, thanks. I couldn't find anything about it when I did some basic research, aside from that it was an FPGA from some manufacturer, so I figured it was either something really cool and new or nothing at all.
posted by mmcg at 5:46 AM on September 29, 2020
posted by mmcg at 5:46 AM on September 29, 2020
effugas: "Self driving trucks end about 16% of American jobs. What should those workers do?"
Learn to code. Duh. /s
posted by SansPoint at 7:22 AM on September 29, 2020
Learn to code. Duh. /s
posted by SansPoint at 7:22 AM on September 29, 2020
So it's labour-washing/mechanical turkery, much like virtually every other form of “AI” beyond puppyslugs?
This is a pretty well known issue by now, yes. The model for most software is: you have high capital costs to develop your software and then no / little marginal cost per unit. Even when you're doing SaaS and therefore hosting the software, that is still the model. That is why the manic drive to revenue and damn the torpedoes attitude to profit that would otherwise make no sense. Doubling sales when you sell cars is very nice but doubling sales when you sell software can be the difference between losing money hand over fist and being wildly profitable.
The problem with a lot of machine learning which seems to be the same as Palantir's problem is that it takes quite a lot manual work to make it all work in practice. I think of SAP as being sort of in between. A lot money spend on the "capital" core SAP software, a lot spent upfront to customise each installation, but then much less annual spend to operate it. With Palantir it seems that even the annual operations costs are quite high.
Fundamentally they succeeded by being better than the alternatives available to the military, but that context is pretty vital - the defence and intelligence establishments have had trouble recruiting their own people to do this kind of work and military contracting is almost impossible to break into.
Out in the big bad world outside though, they have to compete with companies building their own analytics stacks, even with things like PowerBI at the lower end of the scale. When you consider how different the landscape of user friendly analysis software is now (even Excel will plot geodata for you out of the box!) it's not surprising they've been struggling.
posted by atrazine at 7:47 AM on September 29, 2020 [1 favorite]
This is a pretty well known issue by now, yes. The model for most software is: you have high capital costs to develop your software and then no / little marginal cost per unit. Even when you're doing SaaS and therefore hosting the software, that is still the model. That is why the manic drive to revenue and damn the torpedoes attitude to profit that would otherwise make no sense. Doubling sales when you sell cars is very nice but doubling sales when you sell software can be the difference between losing money hand over fist and being wildly profitable.
The problem with a lot of machine learning which seems to be the same as Palantir's problem is that it takes quite a lot manual work to make it all work in practice. I think of SAP as being sort of in between. A lot money spend on the "capital" core SAP software, a lot spent upfront to customise each installation, but then much less annual spend to operate it. With Palantir it seems that even the annual operations costs are quite high.
Fundamentally they succeeded by being better than the alternatives available to the military, but that context is pretty vital - the defence and intelligence establishments have had trouble recruiting their own people to do this kind of work and military contracting is almost impossible to break into.
Out in the big bad world outside though, they have to compete with companies building their own analytics stacks, even with things like PowerBI at the lower end of the scale. When you consider how different the landscape of user friendly analysis software is now (even Excel will plot geodata for you out of the box!) it's not surprising they've been struggling.
posted by atrazine at 7:47 AM on September 29, 2020 [1 favorite]
SAP is a great analogy that I hadn't thought of before. Or maybe Salesforce. These examples make me think Palantir may end up being a success...
I really think it's worth restating, though, that Palantir is not selling itself as an AI or machine learning company. In fact for a long time they sold themselves as "we are very proud of not doing machine learning". As tides have shifted the past few years, they may have backed off on this, idk.
If one wants to understand how a big part of their business involves helping the military and law enforcement build tools that are used to violate people's rights -- I think it's really important to have a picture of how they're different from the many other companies who do try to sell machine learning for the same purposes.
For instance a part of their sales pitch is that by avoiding ML, and instead relying on human operators, they can ensure that only the data absolutely necessary for an investigation is used for that investigation -- minimizing privacy violations, and making a human responsible for anything they view. Worth thinking about how to respond to arguments like this one...
posted by vogon_poet at 9:56 AM on September 29, 2020 [2 favorites]
I really think it's worth restating, though, that Palantir is not selling itself as an AI or machine learning company. In fact for a long time they sold themselves as "we are very proud of not doing machine learning". As tides have shifted the past few years, they may have backed off on this, idk.
If one wants to understand how a big part of their business involves helping the military and law enforcement build tools that are used to violate people's rights -- I think it's really important to have a picture of how they're different from the many other companies who do try to sell machine learning for the same purposes.
For instance a part of their sales pitch is that by avoiding ML, and instead relying on human operators, they can ensure that only the data absolutely necessary for an investigation is used for that investigation -- minimizing privacy violations, and making a human responsible for anything they view. Worth thinking about how to respond to arguments like this one...
posted by vogon_poet at 9:56 AM on September 29, 2020 [2 favorites]
From another lens, realize - we have Rooms Full Of People. A lot of society is organized around figuring out what the heck to do with all of them.
We know what to do with them: infiltrate them. That's the unexpected-upshot of this kind of BS - people behind the surveillance state can have all kinds of motives, including good ones. Be the change you want to see, and all that.
I'm terrified of 'AI' because I understand that term as 'algorithmic intelligence', which is what you get with the current crop of crap (can't distinguish black faces, treat edge cases as halt-and-die apocalypses, etc); I'm terrified of that because it can be used as a force multiplier for smaller-and-smaller numbers of the New Feudalists like Thiel and his ilk. In other words, it won't get you Skynet, but it can get you the autoguns from Aliens, on quadcopter drones. And we are racing towards that. The more 'Rooms full of people' we can interpose, the better.
posted by eclectist at 1:17 PM on September 29, 2020
We know what to do with them: infiltrate them. That's the unexpected-upshot of this kind of BS - people behind the surveillance state can have all kinds of motives, including good ones. Be the change you want to see, and all that.
I'm terrified of 'AI' because I understand that term as 'algorithmic intelligence', which is what you get with the current crop of crap (can't distinguish black faces, treat edge cases as halt-and-die apocalypses, etc); I'm terrified of that because it can be used as a force multiplier for smaller-and-smaller numbers of the New Feudalists like Thiel and his ilk. In other words, it won't get you Skynet, but it can get you the autoguns from Aliens, on quadcopter drones. And we are racing towards that. The more 'Rooms full of people' we can interpose, the better.
posted by eclectist at 1:17 PM on September 29, 2020
« Older Hopper without the melodrama | Just in case you were wondering Newer »
This thread has been archived and is closed to new comments
posted by chavenet at 12:13 PM on September 28, 2020 [4 favorites]