Steven Nelson / US News and World Report – 2017-04-09 18:41:08
Anti-Fraud Experts Launch News-Accuracy Site,
Find US Probably Blamed Wrong Side for Syria Chemical Attack
Can statistics replace human reason? Tech entrepreneurs believe it can, but the effort is already controversial
Steven Nelson / US News and World Report
In August 2013, President Obama accused the Syrian government of attacking civilians with chemical weapons.
(December 6, 2016) — Horrifying footage from Damascus suburbs showed rows of dead children and writhing victims following a 2013 chemical weapons attack. With hundreds killed, President Barack Obama asked the American public to support his proposal to bomb Syrian President Bashar al-Assad’s government, which he said was responsible.
The airstrikes never happened amid a groundswell of opposition in Congress, including claims from one back-bench lawmaker who said he reviewed classified evidence and found it was being “manipulated” to justify military intervention.
But were US authorities yet again asking Congress to consider unsubstantiated claims ahead of a war vote? The debate is being reopened with the launch Tuesday of a news analysis website that says it’s probable anti-Assad rebels actually launched the sarin gas.
The new website, called Rootclaim, seeks to use math to replace human biases in news analysis. It’s a pioneering effort likely to attract controversy and close scrutiny amid sustained focus from mainstream media outlets on “fake news” and biased or shoddy reporting ahead of President-elect Donald Trump’s victory last month.
Rootclaim is the Tel Aviv-based creation of online fraud-detection experts Saar Wilf and Aviv Cohen, who worked at PayPal after the online payments giant bought the Wilf-cofounded Fraud Sciences for $169 million in 2008.
The technology seeks to repurpose the fraud-detection approach used by online businesses. In online fraud detection, individual pieces of evidence — such as the type of email address a user enters and where they are shipping items — don’t necessarily point to fraud, but a collection of indicators increases the probability of malfeasance.
In applying the fraud-detection approach, Rootclaim seeks to break news events into similar bite-sized pieces and assign values to the individual pieces of evidence, factoring for uncertainty and source reliability. The individual pieces are then loaded into an algorithm that draws big-picture conclusions.
The current outcome of the Syria chemical attack analysis: a 92.3 percent probability rebels fighting against Assad were responsible.
“We’re not claiming anything, all we’re claiming is that using mathematical probabilistic tools is better than relying on human intuition,” Wilf says. “We [have] found a model that we are very certain is much better than any human methods of analyzing complex issues. The big question is, how will the world react to it?”
“It’s way beyond fake news,” Cohen says. “Our focus is on taking very complex, typically controversial issues where you have lots of evidence in different directions, sometimes conflicting evidence, evidence with different reliabilities — where us humans, because of all kinds of cognitive flaws, just aren’t able to deal with so much information — and coming up with the right conclusion.”
The site is transparent about its methodology, the founders say, and aspires to have a large user base submitting evidence and debating the proper statistical weight applied to evidence and the credibility of sources for various hot-button news topics.
Who actually committed the Syria chemical attack — probably the most controversial issue addressed initially by Rootclaim — remains surprisingly murky.
The US continues to believe Assad’s government was responsible, says Ned Price, a spokesman for the White House’s National Security Council. The position long has been articulated by the US as a clear-cut truth.
“I’m confident in the case our government has made without waiting for UN inspectors,” Obama said during an Aug. 31, 2013, speech in which he said he was asking for congressional approval for an attack, after the U.K. parliament voted against war.
In a speech on Sept. 10, 2013, Obama said “we know the Assad regime was responsible” and offered the following evidence:
“We know that Assad’s chemical weapons personnel prepared for an attack near an area where they mix sarin gas. They distributed gas masks to their troops. Then they fired rockets from a regime-controlled area into 11 neighborhoods that the regime has been trying to wipe clear of opposition forces.
“Shortly after those rockets landed, the gas spread, and hospitals filled with the dying and the wounded. We know senior figures in Assad’s military machine reviewed the results of the attack, and the regime increased their shelling of the same neighborhoods in the days that followed.”
On Sept. 24, 2013, Obama said at the UN:
But the leader of the UN investigation tells US News he actually is unsure of who committed the attack. The UN report completed in December 2013 by Ake Sellstrom’s team determined a sarin attack happened but did not assign blame.
“I still maintain an interest in what may have happened in Ghouta and do hope that the truth may emerge one day,” Sellstrom says.
“I imagine the facts about Ghouta as a big jigsaw puzzle, where many of the key pieces are missing,” he says. “What I hope for is that those pieces of information may pop up as almost circumstantial information that few people see the relevance of, but they may be the pieces of the puzzle you need to understand what the full pictures actually look like. This could be about the source of the sarin, the loading of the rockets, some crucial information exchanged before the event or other.”
Evidence reviewed as part of the Rootclaim analysis includes low-quality video purportedly of rebels loading similar rockets, which is assessed to be unlikely a forgery, the quality of sarin and the apparent trajectory of rockets alongside evidence offered by the US, including a claimed intercepted call by a Syrian official following the attack.
“What we do is we take the evidence that was in the US government assessment and the UN reports and we bring all of the information inside and we let the system do the analysis of all the evidence,” Wilf says. “We have all the evidence the US claims to have, it’s all in there. Some of it is relevant. It’s not like all the evidence goes for one hypothesis and nothing goes for the other. If it was like that, it would be 99.9 percent. There is some interesting evidence there, it is just much weaker.â€
Statisticians contacted for this story say they are intrigued about Rootclaim’s attempt to apply Bayesian statistical analysis to news. But they have questions and aren’t sure the ambitious attempt will work.
“The Bayesian approach to statistics essentially updates probabilistic description of uncertainty through accumulating data,” Mark Glickman, a Harvard University statistics professor, says in an email. “It is therefore a learning model, and in the case of Rootclaim can adjust probabilities of the veracity of news events by incoming news data (and the reliability of the sources — that would be a major factor as well).”
Glickman says one challenge for Rootclaim will be ensuring inaccurately reported “facts” are discounted.
“If the majority of evaluations or data on news items are accomplished by fraudulent sources, then just using a Bayesian approach itself cannot distinguish between truth and non-truth,” he says.
Duke University statistics professor David Banks says he’s skeptical about whether Bayesian statistics can be useful in assessing the truthfulness of news.
“It’s an exciting and interesting thing to try, and if they are transparent about it, that’s great — the research community will be able to look and see what they do and decide whether their conclusions are highly sensitive to delicate assumptions or if the conclusions are pretty robust to a reasonable range,” Banks says.
“There are lots of Bayesian tools out there,” he adds. “They could be using a smart one and they could be using a dumb one. And that’s one face-down card.”
Other topics addressed on Day One by Rootclaim are less controversial, with near-certain probabilities that a common childhood vaccine does not cause autism and that pro-Russia rebels downed a Malaysia Airlines flight over Ukraine in 2014. A lighthearted analysis finds inconclusive evidence to support a theory explaining Trump’s hair, though finds existing evidence leans against a toupee.
Wilf says it is possible that new evidence would tilt the scales away from Syrian rebel responsibility for the infamous 2013 chemical attack but that it “would have to be pretty strong evidence.”
“If you have one strong substantiated argument, it can beat a million people who support some conventional wisdom that may not be substantiated,” Cohen says.
Steven Nelson is a reporter at US News & World Report.
Posted in accordance with Title 17, Section 107, US Code, for noncommercial, educational purposes.