SoylentNews is people
SoylentNews
from the do-you-want-skynet?-This-is-how-you-get-skynet dept.
The European Patent Office has rejected two patent applications filed on behalf of an AI by researchers. The AI is named DABUS ('device for the autonomous bootstrapping of unified sentience'),
DABUS created two unique, usable ideas that were submitted to [the] patent office: the first was a new kind of beverage container; and the second was a signal device to help search and rescue teams locate a target.
One of the researchers, Ryan Abbot of the University of Surrey, argues that this should have been handled differently
'If I teach my Ph.D. student that and they go on to make a final complex idea, that doesn't make me an inventor on their patent, so it shouldn't with a machine,' he said in October.
He believes the best approach would be to credit the AI as the inventor of the patents, and then credit the AI's human owner as the assignee given license to make decisions about the patent or draw benefit from it.
The EPO rejected the patent applications on the grounds that "there was no human inventor." This is a constraint built into European Copyright law, but until now not part of European Patent law.
Also at Techdirt
(Score: 2) by DannyB on Friday January 10 2020, @03:32PM (3 children)
Imagine if a machine were to "evolve" or "monte carlo" a sequence of machine instructions that can pass all unit tests. Any successful instruction sequences can then be further evolved to optimize toward some particular goal. (eg: faster, smaller, more obfuscated) (or bizarre goals like same binary runs on dissimilar architectures, regardless of binary size or runtime speed)
Even without obfuscation as a goal, it may simply be the case that the working instruction sequence is completely incomprehensible and impenetrable to human analysis. In fact, something like this could be used as an "optimizer" in the code generation stage of LLVM or some compiler.
Trying to understand the code could put a human into a catatonic fugue like state from which recovery is impossible. (That could be another optimization goal of the AI. But my fellow Java programmers and I don't need an AI to generate code that puts the readers into a catatonic state.)
To transfer files: right-click on file, pick Copy. Unplug mouse, plug mouse into other computer. Right-click, paste.
(Score: 4, Insightful) by PiMuNu on Friday January 10 2020, @04:43PM (1 child)
As any tester will tell you, it is not possible to test every input for any reasonably complex piece of code. Passing unit tests is not enough.
(Score: 3, Funny) by DannyB on Friday January 10 2020, @08:09PM
That's what end users are for.
To transfer files: right-click on file, pick Copy. Unplug mouse, plug mouse into other computer. Right-click, paste.
(Score: 3, Interesting) by Mojibake Tengu on Friday January 10 2020, @05:11PM
I suffered similar catatonic states when reading a Haskell language code from Haskell libraries on Hackage, until I understood the real culprit of it, by reading the true code:
Haskell language itself is but a hierarchy of four funny interpreters stacked on top of each other: full pure Haskell built upon the core language, that upon STG interpreter which is a specific implementation of abstract G-machine, that itself disguised as lambda execution model translatable to weak version of C, but built on bloody trickery shuffling little memory blocks (frames and atoms) around heap.
And all this stuff is marketed to unsuspecting innocents as a compiler.
So much for code purity.
Last time I checked, an ingeniously crafted degenerate type declaration could still drive the safest compiler on the planet into infinite loop crash. Not long ago.
I say: if humans could create such contraption without being ashamed on themselves, synthetic intellects could do that much better, for sure.
Respect Authorities. Know your social status. Woke responsibly.