Palantir’s God’s-Eye View of Afghanistan
wired.com
wired.com
They build a federated search engine with a simple UI to make it accessible to a non-technical audience. That’s it. Have a whole lot of data in different databases/data lakes? Palantir will let you search across all of it quickly and simply. They don’t source the data, they even hold the data, they install their servers on premise with where the data is already being stored, and import it.
It’s a specialized Google with very robust and configurable access controls that you can buy and run because you’ve collected more data than you know what to do with.
Now you may still be legitimately unhappy with them because of who they work with or the implication of making those entities more efficient in processing data, but given that HN is a technical audience I’d really hope that we can get over this trend of thinking Palantir has some magical data summoning abilities.
> In 1937 IBM's tabulating equipment enabled organizations to process huge amounts of data. Its clients included the U.S. Government, during its first effort to maintain the employment records for 26 million people pursuant to the Social Security Act, and Hitler's Third Reich, for the tracking of Jews and other persecuted groups, largely through the German subsidiary Dehomag. The social security-related business gave an 81% increase in revenue from 1935 to 1939.
In a counterfactual world where IBM did not sell them the equipment, the Nazis still would have figured out a way to mass murder people. The Japanese and Stalin and Mao did not have access to IBM tech, but they still mass murdered people.
The parent post from a former employee suggests that Palantir is in a similar position:
> They build a federated search engine with a simple UI to make it accessible to a non-technical audience. That’s it.
That's not a value proposition that is beyond the realms of possibility for other companies to develop. So why does Palantir appear to engage with more unethical projects that other companies if it's not by having a low ethical standard? (Obviously I can only comment on what is public knowledge.)
If the technology is not truly required by the unethical client, then you are doing a good, by taking their money and providing technology they don’t need. Money they give is money they can’t spend for other stuff that they truly need.
It just so happens that some of this modeling is done to optimize indexing and searching data on humans our government would prefer dead, and no specific work has ever been done by Palantir employees to optimize anything for this use case, right?
You produce a search engine that helps identify who to kill, but don't pull the trigger so you have no ethical responsibility, right?
This must be how arms dealers think when they sell to dictators.
Don’t blame those that build the software they use. To me that makes every open source project a potential ethical minefield since they could all be getting used to do immoral acts without even knowing about it.
As a concrete example, most of Palantir’s indexing and searching back when I was there was done using lucene. Should the lucene authors feel some moral conflict over this? If we’re shifting blame/responsibility down the tech stack why stop at the productized form? Why not the libraries? Why not the language?
I understand that the case is different since Palantir is more purpose built for this type of work, but you have to admit the lines get blurry pretty quickly.
That type of whataboutisim is at the heart of every major engineering effort of silicon valley and surveillance capitalisim that is used abusively on the masses.
We can say no. We -must- say no.
Making alcohol should be legal. Specifically selling it to a kid is not legal, because then you are specifically helping them commit a crime.
Likewise I don't blame steel mills for producing steel that is later used in guns wielded by dictators on their own people.
I sure do blame the people that sell the guns to said dictators with full knowledge of their likely use case though.
We as engineers have an ethical responsibility if we make a nuke with full knowledge the buyer probably intends to drop it on a city.
I once built a company that produced a very effective social data aggregator, classifier, and sentiment analysis engine with an easy search interface. It quickly identified sentiment trends across a dozen networks and could of been amazing for brand managers or stock traders. My investors decided they wanted to repurpose it for political propaganda and market it to a specific party. I could of said "well this is really cool tech, and I am making good money... Not my problem what they do with it". I thought that for a bit. I even helped build an MVP for political candidate tracking.
It felt wrong, so I quit. I didn't help them replace me. The company ended and all my years of work and research with it. So be it.
I don't always get it right, but looking back, -not- continuing that project remains one of the proudest choices of my life.
Yep. I've got a feeling this is a sentiment shared among all of the engineers working on shady tech. I'd also add, "and even if I quit, someone will quickly take my place."
I was in a similar situation as the GP. I was hired to work on a software that technology-wise we’d all agree was ethically neutral. But if you looked at who all our customers were, and what they used it for, I’d guarantee many HN posters would have ethical concerns. I ended up quitting for a number of reasons, but the ethics of who our customer base was ranked pretty high in that decision.
This seems reasonable.
> Likewise I don't blame steel mills for producing steel that is later used in guns wielded by dictators on their own people.
So far, a steady train of thought
> I sure do blame the people that sell the guns to said dictators with full knowledge of their likely use case though.
This is a bit of a jump. Up to this point, you had a gradually growing line of argument, but then jumped straight to selling guns to dictators.
There's a step right before, and that's manufacturing the gun itself — is that acceptable in your view? I think most people agree that selling guns to dictators is bad, but whether or not the gun should be manufactured at all, and sold to police and militaries of countries with republican governments is debatable.
Palantir does not work with "dictators". At least according to their IPO S-1 filing, as a matter of policy they only work with US and her allies. Whether you think that NATO states constitute "dictators" is an orthogonal and controversial claim. If you agree that NATO defense agencies and local police departments don't constitute "dictators", then you would agree that what Palantir is doing here, per your analogy, is "making alcohol", but not "selling it to a kid".
I wouldn't conflate the US as a dictatorship (maybe a broken two-party corporatocracy), but in light of the BLM protests, dealing with local police departments to better surveil their own citizens is unethical. In light of Snowden's revelations, dealing with ICE to better surveil their own citizens is unethical.
I'm not sure that this is so clear cut. This also one of those things where it's fairly uncontroversial that the George Floyd (and all of the other comparable incidents) were unquestionably bad. But is that representative of all policing? Are police departments, as a concept, unethical? I'm not sure that this is clear cut or obvious.
And finally, all of these police departments are funded and run by republican forms of government (which BTW are predominately one-party democracies, just not the party that's popular to hate), that can democratically change. It's not a dictatorship. Any more than 21st century Italy is a "dictatorship" because it briefly had an authoritarian Prime Minister (Berlusconi).
> In light of Snowden's revelations, dealing with ICE to better surveil their own citizens is unethical.
Sure, and I think that it's up to the public to elect leaders that change these policies, which is possible because the US is a Federal republic, not a dictatorship. Unfortunately, in light of right-wing extremism, domestic surveillance is increasingly developing bipartisan approval. Is it unethical to serve the needs of the people even if some subset of those people don't approve of domestic surveillance? Unless it's not about doing business with dictators, it's about doing business with democracies that enact policies we don't like, but that's a different line of argument entirely, and one much less straightforwardly bad than "selling guns to dictators".
In their current state, yes. Tough on Crime, War on Drugs, and the prison-industrial complex are unwarranted and destructive.
> (which BTW are predominately one-party democracies, just not the party that's popular to hate), that can democratically change. It's not a dictatorship. [...])
State and local governments that don't change party-wise are much worse, as their actions are even less accountable; there's no reason to improve.
> in light of right-wing extremism, domestic surveillance is increasingly developing bipartisan approval
It used to be assumed we had domestic privacy; it's Snowden's revelations that let us know we don't have such rights. Now we've found ourselves in violation of Article 12 of the Universal Declaration of Human Rights.
But this presupposes that those things are unambiguously bad. "Tough on Crime"? The American public broadly agrees that crime is bad and should be punished. I agree with you that the War on Drugs is bad, but that's not representative of policing as a whole. I understand the "ACAB" outlook, but you have to understand that this isn't an objective moral position, it's purely subjective.
> State and local governments that don't change party-wise are much worse, as their actions are even less accountable; there's no reason to improve.
You're conflating dictatorships where the people have no mechanism by which to enact change with democracies where the people enact policies you don't like. As a software engineer at Palantir, you're essentially selling weapons to democratically elected governments whose constituents broadly agree with and vote for policies you don't like. That's obviously not great, but it's also not as obviously bad as the punchy "selling weapons to dictators".
> It used to be assumed we had domestic privacy; it's Snowden's revelations that let us know we don't have such rights. Now we've found ourselves in violation of Article 12 of the Universal Declaration of Human Rights.
Sure, and it's been 10+ years since Snowden's revelations. We've had 2 presidential elections and 2 mid-term elections since then to address the problem using the levers of our republican government. A far cry from "dictatorships", and by no means is it obvious that it's unethical to provide software for an elected government that has enacted policy you and I both don't like.
This is a slippery argument.
There is a difference in building a car that people can drive and one that is specifically to be used as a weapon.
This sounds like an equally slippery argument: one whose premise is that weapons should never be built, and that there is no justified use for them. The unfortunate reality of the world is that we do have and require militaries with intelligence agencies.
Based on the premise of your argument, the only non-slippery, internally consistent framework is that building weapons is universally and unambiguously bad.
It is a matter of degree which becomes a matter of kind. There is a difference between a bow and an arrow, a semi automatic and the atom bomb.
This is something I keep pondering over, for all the brilliant minds involved in the Manhattan project, did they ever think about the effects of what they created.
When do we get to vote on the "no killing people based on shoddy intelligence" line item exactly?
Hell, we don't even get to vote for who we want to be president (the electoral college does that and it's certainly not one person, one vote).
https://news.ycombinator.com/item?id=24740791
https://news.ycombinator.com/item?id=25542949
Sometimes I fear people don't see the difference ethically based on HN discussions on other issues.
...used to kill innocent people
Thank you.
So I get a lot of downvotes and lose some imaginary internet points; that’s fine. I’m not attacking anyone, and for the most part nobody is attacking me, so I’d still call this healthy if contentious debate.
You are a person I would gladly grab a beer with when such things are a thing again.
They were actually incredibly open with what they did technically, and a lot of who they worked with it’s just that nobody seemed to care at the time. They had tech talks showing the full system architecture on YouTube that didn’t get more than 10k or so views. A quick search indicates that a lot of those are gone now (maybe just outdated) and the channel is more marketing heavy now, but just as low trafficed.
They also open source a lot of their core technologies. The main database they used was “AtlasDB”, a relational layer on top of Cassandra. It’s open sourced and actively developed, and again, nobody seems to care, with it at a measly 700ish stars - https://github.com/palantir/atlasdb
People just like a good boogie man story, and after 2 decades of being labeled as such eventually you just stop trying to fight it I guess.
Good news is that I was one of those people that had a clearance, so can confirm that those individuals exist at least.
Bad news is because I had a clearance, I can’t say anything more than that. Way too much liability to even speak vaguely about it, even tho I no longer have it. The US government doesn’t play around with that type of stuff.
However i disagree with your conclusion.
Palantir has some magical data summoning activities.here is why
Circa 1990, if you said there was a way you could get any answer you wanted in 10 seconds or less,you would have been called a charlatan.
Then google happened. Google is not worth billions because it just has a simple UI and is accesible. No. They have figured out how to make the web searchable.
Palantir is the same. Just because search is not romantic anymore, doesnt mean they dont have magic to them. If aggregating intelligence data was such a no brainer , google or many of the others may have taken a stab at the problem.
Here lies what i believe is a contradiction. If palantir does not have "magic", then why do you mention one of the most valuable companies on earth to describe palantir's function?
In short, just because the u2 has wings and turbines, it doesn't mean it was just another plane when it came out.
They do a lot more than what you claim. Did you forget e.g. blu/red force tracking and routes? Mission planning?
Governments using private companies to decide who deserves to die based on ML algorithms without judge or jury is morally reprehensible. Full stop.
They live in another country so they don't deserve the same rights we superior americans do, right?
This behavior only gets to continue to happen because of "we require secrecy to do our work" excuses which are code for "we know the public that indirectly funds us would not approve"
I grew up going on jobs stalking people with my private investigator father who mostly worked fraud cases for insurance companies. Public places, license plates, collecting evidence of an injured person walking just fine... then prosecutors can use that evidence in court where it belongs.
I am not saying this tech should not exist, or can't be used to serve justice in the right context. I am saying those with that kind of power must have maximum transparency and a high level of accountability.
I think the NSA does important work too, but power in secrecy is a breeding ground for "the greater good" thinking. It is so easy to overstep and do things like tap lines to Google datacenters to spy on private citizens with no evidence it has ever stopped a single terrorist attack.
After the collapse of the Soviet Union, who took power? The one with the most access to signals intelligence. Now that country is a surveillance state dictatorship disguised as a democracy.
The US keeps trending this way too recently.
Who will be the Snowden of Palintir to give it the public scrutiny required to make sure their powers stay in reasonable bounds?
If a Palantir employee is reading this, maybe it could be you.
It goes beyond that. People don't care Palantir tried to do predictive policing via ML in the US because the suspected criminals were poor POC.
Lets not muddy the discussion, WASPs in Connecticut dont have to worry about Palantir as much as a random brown person in Asia.
But comparatively -- yes, I agree with your point, it is nothing.
I don’t think his color comes into it. It’s as if in WWII we were like, oh noes, they’re wasps (ignore Bavaria for a moment) don’t shoot!
Or with Russians today, oh noes, don’t make them into all powerful boogeymen.
Not everything in the world is about color.
The military does not care who the enemy is, they engage without mercy as is the case in war or conflict.
The problem is, any kind of big data dump inevitably would be met with the same excuse which is made on behalf of all big tech power these days: "it's just a private firm so it doesn't need to abide by the same limitations that would apply to the government."
Governments (or, more accurately, human beings acting under color of law) deciding to assassinate people is morally reprehensible, regardless of how the targeting is done.
If "private company" and "ML" are the keywords it takes to get people outraged about assassination programs, I'm fine with that, but let's make sure we know why we're upset.
This is premeditated murder, and throwing around terms like "insurgent" and "rules of engagement" are just a way of legitimizing premeditated murder.
Assume for the sake of argument that the United States has a legitimate interest in fighting the Taliban. Within such an exercise I hope we could agree that there is a distinction between lawful violence and unlawful violence. People of goodwill can disagree about where to draw that distinction. But 'rules of engagement' is not a mealy-mouthed weasel word. It's how we describe this distinction.
(It may be likely that you don't accept under any circumstance that the US has a legitimate interest in fighting the Taliban. That's fine but it's a different discussion.)
It is incredibly critical every citizen be capable of choosing -not- to do things that are legal and would benefit them, because they find them unethical.
This is how majority held ethics graduate to becoming laws.
We can't have 21st century discussion about race and gender equality and then ignore this glaring fault. I know some people will come and say 'you know this is how states and secret services and whatnot should operate blablabla'.
Its plain wrong. Justice and equality of human beings simply don't exist if inconvenient. It doesn't matter if its my or your state doing it.
And arguably the worst part - from all I could ever read on it, this different type of racism doesn't even bring any proper safety results for US that would justify it in some twisted optics. It for sure pisses tons of people all around the world though.
Most often and historically governments used poorly fed fellows (companies of privates instead of private companies) to decide who lives or dies. Is decision by ML morally different? Does it have the redemptive possibility of making better more informed decisions?
Some choice quotes below.
Must be super boring:
> Kevin spent hours a day, seven days a week watching all kinds of people go about their lives, with the goal of separating out the insurgents from the civilians.
Terrifying:
> These are actions that allow for that individual to be moved out of civilian status and into insurgent status—to be targeted and killed legally according to army rules of engagement.
> Because we know what you did, we think we know what you are going to do next.
Powerpoint. Also terrifying.
> The 429 package, which allows an insurgent to be killed in an air strike, must meet the legal requirements. The full-motion video gets snapshotted as evidence. While this is going on, the PGSS operator quickly generates a PowerPoint containing all the data, which goes to the S2. The S2 quickly reviews that, then sends the information to the battle captain.
> If there’s not air support available, then the person of interest remains marked for death in the system. “The moment there is a target of opportunity to take him out, I call it in. I don’t have to go back through the approving process,” Kevin says. “The 429 package stands. That’s why it’s called a Target of Opportunity. When you have the opportunity, you strike the target.” You kill the man.
"Should we kill him?"
"I dunno, where's the slide deck? Are the slides convincing?"
...that just seems like an absolutely terrible way to decide between life and death.
I know that on the one hand we're just talking about a file format, and the data has to exchange hands somehow. But it bothers me that the same tool I use to convince my bosses to, say, pay for a monthly subscription, is also being used by someone else to convince their bosses to approve killing human beings.
Perhaps the reason this bugs me is it breaks the illusion of expertise and sophistication. It proves, once again, that people are just people, no matter what job functions they happen to have.
(Also: It raises certain questions. Like: Is a person more or less likely to be killed if their powerpoint deck is ugly? Does typography impact these decisions?)
CHAIRMAN: What was your position.?
CICCI: Well at first like everyone else I, I was a soldier.
CHAIRMAN: What is that?
CICCI: A button you know SENATOR, come on.
CHAIRMAN: No I don't know. Tell me.
CICCI: Well -- when the boss says push a button on I guy -- I push a button. See SENATOR.
QUESTADT: You mean you kill people.
CICCI: I what?
QUESTADT: You kill people at the -- uh -- at the behest of your superiors?
[CICCI's lawyer tells him to say yes.]
CICCI: Yea -- that's right Counselor.(Not a defense of killing people based on slide decks, more of a comment on the historical consequences of war and military occupations)
Are they even a threat to us? Maybe they are, but I frankly think that you gotta wonder why we’re still killing terrorists in the Middle East.
People are usually easily biased by the information they're constantly fed. It becomes easy to concoct a good explanation for why something unacceptable can become acceptable sometimes. "We're fighting for freedom and democracy", "fighting the enemy", "fighting terrorists and pedophiles". All of a sudden a lot of stuff is justified. Then someone else uses it and the argument gets turned on its head. "They're terrorists, enemies, dictators", all others are too "morally corrupt" to wield this power.
Most of those people are able to justify anything as long as they're doing it or at least it's not being done to them... Until eventually it is Because what's ironic is that most don't realize that to maintain power you must keep your enemies in check but eventually also your own citizens.
I’ve been doing work in the realm of Covid for basically the last year. Part of that privilege is access to HHS Protect, as managed by Palantir.
Just to soothe some unknowns, Covid surveillance is nothing like the kind of military surveillance described in this article. There is no personally identifiable data and certainly no effort to trace individuals with Covid.
The actual data inside HHS protect is not much more specific than what is publicly available. Covid tests/cases/deaths by county, locations of long term care facilities, pharmacies, hospitals. Data on supplies, hospitalizations, staffing. Demographic data comes straight from Census data, also still aggregated.
In addressing Covid, I often wish we had better data. Google and Apple’s earlier plan to anonymously contact trace is far more “creepy” than anything the federal government currently has. As you’ll notice, that tech got basically no traction. Contact tracing has largely been a failure in the US, but that’s a whole other story.
Do you really think that Covid would have hit the US as intensely as it did if we had military level surveillance? I don’t. Many points of inaction are due to unclear or insufficient data.
On the other hand, the raw data available to Google, Apple, Facebook and others is insane to me. Your phone is your tracking device. These companies have it as a unique identifier and most definitely track a large number of users in an individually identifiable way. The carriers do this as well.
I absolutely agree that we need refrain allowing this technology to be used in domestic policing, but in terms of war I see it a net positive. Data drives insight which drives action. Palantir just happened to build one of the best platforms to do this. (You may disagree with any of the above, that is fine)
The cat's out of the bag. Now it's up to us what we do with it (and what we tolerate the government doing with it).
Any clue on why so much pressure was needed to release an otherwise innocuous contract[1]?
This reading doesn't inspire much confidence[2] to be honest.
[1]: https://www.opendemocracy.net/en/ournhs/under-pressure-uk-go...
[2]: https://www.cnbc.com/2020/06/08/palantir-nhs-covid-19-data.h...
I think there is some hypocracy in doing so while simultaneously paying for the bombs voteing and for the people who give the order. The buck stops with the american people, and only they have the power to stop it.
We'll soon need a new word for actual terrorists.
https://reliefweb.int/report/world/decade-global-ied-harm-re...
Your cat isn't a squirrel, drone operators aren't terrorists, and neither are the Taliban fighters who plant IEDs targeting combatants.
I believe the idiom you were actually looking for was "If it looks like a duck, swims like a duck, and quacks like a duck, then it probably is a duck."
At least that might've supported your argument.
That's a critically important role and I think an apt description of knowledge work today. And it has implications for machine learning and predictive analytics.
Here the wrong guy was targeted based on the purple hat which was a sensor processing error based on early light levels.
While integrating all of this data is important and building machine learning models that can pick up specific markers is important, it is clear that you need some type of overall exception processing framework to confirm in this case positive id.
We have a similar problem already with law enforcement no knock warrants.
For example, in a project where failing to predict something would result in costs up to $100 million and wasting two months, and catastrophic consequences, but a false positive would result in an engineer waking up to check, it is acceptable to have false positives. Yes, it may be inconvenient, but compared to having nothing and the engineers telling you it's okay to bother them because one of catastrophic consequences is they may lose their life, the system can cry wolf. In other projects, not so much. And in other projects, you're not dealing with consequences of that magnitude.
Logic also had a story about palantir this quarter. What I found most remarkable about that article is that the UI hasn’t changed in a decade! What looked edgy ten years ago now looks cheesy as hell.
I don’t know how the internals have changed over that time, but at that time, palantir was an entity resolver and a search engine. They also sold and continue to sell a ton of consulting services at mid-tier rates. In that sense their biggest competitor is... Booz Allen although they’d never admit it. BAH at least makes a profit.
So, he is happy to spy on another country's citizens, from within their own country, but he draws the line at it being done back home. Hypocrisy.
I'd argue this is more a case of an aptronym, mentioned in the second paragraph of the wiki entry you referenced. "Nominative determinism differs from the related concept aptronym, and its synonyms aptonym, namephreak, and Perfect Fit Last Name (captured by the Latin phrase nomen est omen 'the name is a sign'), in that it focuses on causality. Aptronym merely means the name is fitting, without saying anything about why it has come to fit."
The whole story is such a direct fable of the danger of powerful tools being used perhaps initially with neutral or good intent but ultimately serving the interest and aims of powerful evil people, how the very power itself will inevitably corrupt the bearer and how some tools are perhaps just better off never ever being used. With a company actually called Palantir, is this just lack of awareness or is it nihilism?
Far too fitting indeed.
on a more serious note, people do choose the names for their companies/products. they could be totally meaningless, merely sound good, or convey something about the brand. it's not a rigorous argument, but nor is it ridiculous to wonder what they were going for here.