With coding partner AI tools like Githubs Copilot arriving how long before Alexa or Siri...

With coding partner AI tools like Githubs Copilot arriving how long before Alexa or Siri can help us write games?

Copilot -> GitHub's Copilot Is an OpenAI-Powered Coding Partner (pcmag.com)

Hint @Unity DOTS it can write boilerplate code for people!

Will different knowledge domains especially those with bespoke API systems and even engines need seperate dedicated AI coding tools for maximum benefit?

Imagine working with a partner coding AI that knows the code from every game released and can help you get your game started with the best architecture and designs in the industry?

What would this mean for the Asset store and programmers in general and in the future?

Could we see the number of programmers needed to write games drop in the future from hundreds on large projects to handfulls with AI?

Could there be a new golden age for solo + AI developers of games and apps in the near future?

 

For me personally, the proceduralism introduced by Roslyn source generators makes me more than happy. The best thing about it (aside from being super useful) is that I don't have to train it to eventually replace me.

Can we actually use this tech to improve the world and just replace billionaires with AI? please?

 

skynet

 

That's not all it can write! It's been drawing all its data from code in public repos regardless of license, which may be putting them in some potentially groundbreaking legal hot water. When I say "groundbreaking," I mean "this sort of problem could set legal precedent that changes how we approach these sorts of things forever."

https://www.theverge.com/2021/7/7/22561180/github-copilot-legal-copyright-fair-use-public-code

 

Yeah, the license thing is pretty interesting, because that's going to force someone, somewhere to legally define what "learning" means and how it is different to "copying".

When I write a for-loop in my code nobody accuses me of copying it from whatever long-forgotten source I originally learned about for-loops from. It is accepted that I understand them and can generate them originally, even if they're 75% similar to other for-loops. But a computer doesn't "understand" stuff, it's largely matching patterns it's seen before against the pattern it's seeing now.

So it's a legitimately interesting philosophical question to ask: at what point do those patterns stop being copies of the other peoples' stuff it learned from and start being the foundation of another entity's own knowledge?

If I can learn from other people's stuff and then write original things of my own based on that knowledge then it stands to reason that a sufficiently complicated machine could do the same. But where is the line?

- - -

Of course, in this case the interesting philosophical question may be putting the cart way before the horse, as CoPilot has been suggesting obviously recognisable fragments of other peoples' code which it very clearly does not "understand".

 

That's why I was thinking will Unity need it's own domain specific Co-pilot or even genre specific versions, assuming that the narrower the specific domain the better at writing code for it an AI will be.

It would also massivly reduce the training needed from all genres to just Adventure games, RPGs, Simulations, FPS etc...

Or would a Knowledge tree style set of AI's be better, with basic C# programming as the root and then branches of the AI trained towards different domains e.g. a leaf AI could specialise in "2D isometric twin stick shooters".

Or would AI's specialise in a similar way to human programmers along technical and APIs, graphics, networking, databases ect.

Have we got hybrid symbiotic AI's yet where one AI can be added as a higher level to a more basic AI e.g. Programming, Design, Architecture layers, after all this is the way the human brain works?

 

This is how I see, this would need to work. One of options.


A. Each dev team owns and trains own set of rules for copilot. This way there is no issue of copy rights, as all stays internally to the team/company.


B. People using copilot AI agree, that if they contributing to the AI training, or using tool, while writing the code, will wave off the rights to legally chase after copy rights, if code is generated by the tool.

In the case of any legal actions, here need to be proved that code could not be generated at the time with that tool and due to fact, could be leveraged as copy right case.


C. People using this tool are declared as contributors for general open to public resources. Any resources trained by AI are in public domain. So the relevant code parts can not be case for dispute.

I would suspect, this over the time, would generate more unified scripts across the globe.

 

In one example I saw it was outputting verbatim code that required an attribution license and then, upon generating a new license itself, used LGPL.

Basically, I wouldn't use ANY of these services in anything that would see a public release in any capacity.

 

It's going to be easy for a lot of people to say "well, those things shouldn't have been posted in a public repo to start with," which is correct, but that's a fundamental distraction from the fact that Microsoft should know better than this too and they're the ones who released a product.

 

There are quite a few people in the chain who should have known better, including anyone who wrote or was involved in putting this little statement on their website (quoted via the Verge article lined earlier):

The "machine learning community" decided it was ok for themselves to use other peoples' stuff and... that's that?

If it were "training" in the human sense that would be fine. But it's not even close to that at this stage.

 

Ehh yeah, because something is puboically available, doesn't meant it publically licenced.
I see now major issues with current training data set approach.

 

No thinking of the very obvious consequences. Not studying the actual problem domain for magic ai.

Sound like the day were we are all turned into paper clip is getting closer.

 

Only first AI will arrive like helpful intellisense code completion and optimisation technology, then as you train it on how you code and the company gathers real world data on coding in different domains and gets better and better at coding/helping until it can develope, write and test entire modules of an app.

So should Github contributors and developers who use AI tools be given shares in the technology companies that use their work?

 

But the problem raised aren't ai efficiency problems, they social and common sense problems. The failure wasn't on the ai, it was in the organization that created the ai.

 

No, they shouldn't use the work without appropriate licensing in the first place.

 

Githubs CoPilot is insane levels of abuse of access.

Some people defend Microsoft.

Some even champion them.

Others think Bill Gates is genuinely philanthropic.

Surely things like this should give them pause...

...consider this:

Microsoft is frequently considered the lesser of the four horsemen of the apocalypse.

Google, Facebook and Amazon are, I'd argue, a result of Microsoft's success, in one way or another.

Which kind of makes you think Microsoft might still be the worst, despite Google being the most powerful.

 

Dunno about AI devs, but solo game developers seem to be going through a golden age in Roblox.

Not my cup of tea, but the amount of growth in usage Roblox as a platform is seeing would tend to indicate a metamorphosis in the means and methods of digital creativity is occurring, right under our noses.

Dreams, on Playstation, might be a hint at what's to come, in a next generational update and evolution to these ideas of user empowerment.

 

When you sign up for github it tells you they are going to mine your public repos.

And if your repos are public, you are obviously already fine with people learning from your code. And let's be honest, half of those people aren't learning as much as they are copying and pasting, (regardless of any license) which is usually a terrible idea, but that doesn't stop people from doing it..

I'm sure people have already built web crawlers that mine and analyze code posted on the internet, including from github and stack overflow and a hundred other places, because anyone can see it, and the people that post it don't care that anyone can see it.

Copilot is kinda like a smart google search for code, which is something that everyone does all the time.
Now you don't even need to leave your code editor.

A pity that it isn't going to be free

 

There's the key difference.

Any "ONE" versus a one eyed monster that's able to see everything, all at once, and project back out to everyone based on its comprehensions of all that it sees.

"Corporations are people, too, my friend."

 

Assuming that's true, where's the line between "learning from" and "directly copying" or "derived work"?

My local library has a bunch of blurays I can borrow. I think it's pretty obviously not ok for me to rip them and put them online. They're also "public". This is the whole reason that licenses exist in the first place.

 

Just the other day you were saying how visual programming is the revolution that will shake up and change the status quo of programming, and you were happy about it.

Copilot is more of a revolution. I don't know about you, but I always want better programming tools, and I always want programming to become more convenient.

The law might draw a line. You'd actually have to get caught and be worth suing for anyone to care, though.

I just find it very funny that the anti microsoft people are almost always the "everything needs to be open source, copywrite is evil, restrictive licenses are evil" kinds of people. So when Microsoft releases a BETA test of a tool that isn't configured properly and occasionally practices the ideology of most of the anti microsoft people, they somehow still find a way to complain about it, and they do a 180 and now they're all for restrictive licences and copywrite and they want to private their code.

 

You misunderstood what was said.

 

Same like most webpages you can access the source code directly, doesnt mean you can go and use copyrighted code from another persons website!

It is crazy that microsoft went ahead with this before at least annoucing what it is to get a feel for the public reaction.

 

Did I? That's not my fault, nobody can comprehend what you said.

 

Perhaps nobody you know.