Slashdot Mirror
Chatbot that Overturned 160,000 Parking Fines Now Helping Refugees Claim Asylum (theguardian.com)
Elena Cresci, writing for The Guardian: The creator of a chatbot which overturned more than 160,000 parking fines and helped vulnerable people apply for emergency housing is now turning the bot to helping refugees claim asylum. The original DoNotPay, created by Stanford student Joshua Browder, describes itself as "the world's first robot lawyer", giving free legal aid to users through a simple-to-use chat interface. The chatbot, using Facebook Messenger, can now help refugees fill in an immigration application in the US and Canada. For those in the UK, it helps them apply for asylum support. The London-born developer worked with lawyers in each country, as well as speaking to asylum seekers whose applications have been successful. Browder says this new functionality for his robot lawyer is "long overdue". He told the Guardian: "I've been trying to launch this for about six months -- I initially wanted to do it in the summer. But I wanted to make sure I got it right because it's such a complicated issue. I kept showing it to lawyers throughout the process and I'd go back and tweak it.
Evidence please? And not "it's been used 160,000 times".
A simple search in /. for "chatbot parking" turned up this previous article, which indicated that it successfully challenged 160,000 out of 250,000 tickets. So, no, not "it's been used 160,000 times". This is a "it's won 160,000 times". And that was as of June of last year. This NPR piece from earlier this year indicates that its up to 200,000 successful cases now in just three cities, and that its overall success rate with parking tickets stands at around 60%.
From what I understand of the current asylum interview process, the key question is "is your life in danger" followed by variations on "prove it." (Sometimes the proof is as simple as pointing to death threats on Facebook.) Does anyone know if coaching this process is what this bot is doing?
Yes...but using that reductive approach, you can say that this is how almost any compliance/vetting process works.
PCI DSS: "Do you handle payment card information securely," followed by variations on "prove it." Yet, accomplishing this is expensive and challenging.
Tax audit: "Have you paid what you owe for taxes," followed by variations on "prove it." The visceral reaction of anyone who has been through a tax audit makes my point here.
Security clearance interview: "Can we trust you with state secrets," followed by variation on "prove it." This gets even more interesting if you get a polygraph exam...which is essentially nothing more than a twisted, mind-fucky variation of the same.
The trick is in the "prove it" part...or more specifically, the overlap between what actual means are feasible for providing proof combined with what the questioning entity defines as acceptable proof. In different situations, this overlap may be subject to negotiation as well (or not), and that is its own area of expertise unto itself in some cases. Almost all of these processes also involve setting legal precedents during their early days as well.
In short: sure, you can use a verbal metaphor to represent the process in an oversimplified manner. But that doesn't make the actual process...as required by anyone who engages with it...simple or easy.
For your security, this post has been encrypted with ROT-13, twice.
A chatbot can not get Congress to fund care for the large increase in veterans Congress decided to create.
Every government form should have chatbot assistance available. The purpose of the government form is to convey information accurately so that help can be delivered to the people that qualify and rule out assistance to people that should be denied - that is the government's intent after all.
If the chatbot helps imrorve the effectiveness of that process, and it seems there is evidence that it does as I have not heard that the 160,000 odd parking citations were overturned improperly, then all for the better.
Nullius in verba