[infurl]

https://thespinoff.co.nz/the-best-of/06-03-2018/the-mystery-of-zach-new-zealands-all-too-miraculous-medical-ai/

This article makes for an interesting read.

With AI being so much in the news lately it's only to be expected that it will start to become a feature of scams, along with blockchain and cryptocurrency. It's funny how easy it can be to fool "the smartest person in the room" sometimes.

[Don Patrick]

At first the technology sounded like neural networks. I'm sure one could train a network to take a stab at what notes to output for a bunch of text. Spelling mistakes could have been in the original training material, so that's not immediately proof.

But then it goes right into common crackpot territory by saying it's nothing like anyone's done, while emphasizing technicalities instead of the AI.

it uses a completely different approach to anyone else’s. Most people are using neural networks, and that is all very well – and has advantage of being portable, reproducible, and you can run it on a Windows box. But this runs on its own special hardware. It has its own custom made hardware – custom silicon – which is designed for natural language processing.

While this and the slow response times support the possibility that it's a mechanical Turk, I personally think it all points to a basement-dwelling programmer with scrapyard resources. I'm imagining there's just one computer in use, with a program that's just not designed for multi-thread use, and a programmer who doesn't know how to make a secure API (I don't either, but I do know how to hook into an email-to-php service if I needed a workaround). The description above indicates that the program is not portable, not reproducable, and doesn't run on windows, which could go some way to explaining its limited functionality (i.e. it's an obscure mess). And "liquid nitrogen cooled supercomputer" sounds like someone's boasting about having a server. It all suggests limited skills and hobbyist enthusiasm to me. Not a willful scam artist, but hardly trustworthy.

[Art]

Great find. After reading through the entire article I had to smile.

If they are using supercomputers, the response time would not be up to 20 minutes. IBM's Watson provided answers within 30 seconds.

Emailed responses? Right.

A limited number of requests? For a Super Computer? Right.

This Dr. Seddon Smith who is training Zach stated that they are installing multiple CrayXC50 Supercomputers!
Multiple? The NIWA just spent $18 Million for TWO CrayXC50's! Thay are installing Multiple CrayXC50's? Right.

Somewhere under this dungheap of misdirected AI are actually some humans, somewhat unscrupulous individuals intent on pulling off an almost believable scam.

Like the hidden man under the chessboard...the aforementioned Turk. Things are not always as they appear.

Lesson: "If it sounds too good to be true, it usually is."
(But a lot of us already knew that)

P.T. Barnum would be proud!

[Don Patrick]

I used to own a Supercomputer. It was a laptop with the Kryptonian symbol for hope on the lid.
Nowadays I have a Batcomputer.

[infurl]

Funny thing is I've spent a significant portion of my career using and programming supercomputers. I can vouch for the fact that they are not always used to perform the tasks that can be done quickly. Quite the opposite, because of their speed they tend to be used for the tasks that would take an impossibly long time with slower machines.

I was working in a high transaction commercial environment for more than a decade and one of my notable successes was reducing the time that one of our RS9000s (the same kind of computer that beat Gary Kasparov at chess) took to calculate how to restock a supermarket. I got the time down from a couple of hours to twenty minutes which meant the difference between being able to keep up with demand, and supermarket shelves becoming empty.

Twenty years later I've got several computers on and around my desk that are more powerful than that machine was and I could probably have them doing the same task in seconds.

However, Zach reads like a scam from beginning to end.

[Don Patrick]

Interesting to hear your experiences with actual supercomputers.

As for Zach the AI, there's more news, and I have to admit that it is a scam. I'm used to scam artists who put more effort into writing a credible story, so this one didn't tick the boxes for me until I saw the effort gone into the money-grabbing part.
https://twitter.com/davidfarrier/status/971310467217002497

[infurl]

So yeah, this wasn't a scam, it was just a delusional crackpot and an indulgent parent, a scenario that we're painfully familiar with here on this forum. The only victim was the reputation of the doctor involved but no laws were broken and nothing was stolen so nobody was punished.

https://www.stuff.co.nz/business/118797355/doctor-duped-into-believing-supercomputer-was-real