JudoP
Funditor
Offline
Rocks away!
Posts: 936
UK
|
Oh boy. Before I got into this field my thoughts lined up more with the Sam Harris's and Elon Musk's, but now having been pretty close to AI research for a couple years, and with many colleagues in it now I am much less concerned. Demystifying the tech makes it a lot less scary indeed.
Firstly, as far as I can see general AI is a far off pipe dream, hell, nobody even seems to bother working on it, the only place I hear about it is the internet, literally I can't even remember seeing any AGI work in academia ever, unless I deliberately search for it, and then it's usually just a theoretical perspective, rarely anything with even the most rudimentary function. Also smart devices are certainly not general AI, they have multiple types of narrow AI in them, eg- face recognition, speech to text etc. A phone with general AI could learn all those things and continue to learn a wide variety of tasks as well as making plans, reasoning and other stuff.
Pretty much all AI today is essentially tools that process data and move around their parameters according to the data they process. These are highly effective for certain tasks (mostly data processing) and could certainly change the world. However, they are certainly not liable to suddenly overwrite their programming or do anything that requires agency. AI like this is not an entity, it's just a method of computing and data processing. The chances of a cutting edge neural network sprouting agency are exactly equivalent to your dishwasher sprouting agency, that is to say- zero.
I have to just quickly go for the two AI's 'talking' to each other story too, the AI that facebook allegedly shut down because they were 'scared'. The hype around this springs from what can only be described as abysmal, trash tier reporting and talking about the AI like they are making their own decisions or trying to deceive their masters. You will be glad to know that this story is absolute rubbish. The truth is when you have two programs which are programmed to try to spew sentences back and forth you (like with the vast majority of failed experiments) often start to generate gibberish. To treat this as if the AI's were deliberately conspiring is irresponsible reporting in the extreme. Again, these things crunch numbers according to well known rules, they have the agency of your dishwasher.
If you want a good laugh read the Harry Potter AI generated chapter, it illustrates how little the AI understands the actual text and how it's basically just a data spewer trying to replicate the patterns that we feed into it. It captures JKs style... somewhat, but has the structure and narrative of an acid trip fever dream.
|