General > General Technical Chat

Why does OpenAI ChatGPT, Possibly Want to disagree/annoy and change my eating...

<< < (18/22) > >>

MK14:

--- Quote from: Zero999 on January 04, 2023, 06:21:32 pm ---Computers are already powerful enough to achieve it

--- End quote ---

I hadn't properly noticed that bit, until another poster, had mentioned it.

But, I don't think we know, what computer power level, would be needed (as a minimum), when proper/useful general intelligence, is invented/created, which might never happen.  We will then know (or eventually find out), what computer power level, is needed, to create it, at various, useful (equivalent to human) IQ levels.

I.e. I don't think we know if the existing available computing power levels, e.g. a big server, or set of server racks, full of lots of servers.  Is currently powerful enough to do it.  It might be, but then again it might not be.

Taking the human brain as an example, is not necessarily going to give accurate results.  E.g. Computers have been powerful enough, to beat, even the best/world (human) Chess champion, in a real time game of chess.  For a long time now.

EDIT2: On the other hand, me mentioning that the initial problem, of having computers, powerful enough to beat even the best human (at the time), at chess, seems to have already been solved.  I suspect, we don't really know (world-wide), how much extra computer power (if any), is needed, to be able to create a more general type of AI.
Looking at the existing human brains computer power, seems to me, like looking at human muscles, to attempt to predict how powerful/fast a car is going to be, before cars have been invented.
It would give an extremely rough, potentially very inaccurate indeed results.  Because a machine (car), is significantly different to a human being.
Analogy, to further make the point.
Maybe we humans can remember, a few bytes of information, when remembering a quick, temporary number.  E.g. Please remember (without writing it down, or cheating) the 10 digit number, 9746641372, while continually jogging, singing out loud, and answering simple mental arithmetic questions, to be tested on that number, in around 3 hours.

So, call it 4 or 5 bytes of storage (temporary/quick, one-off), in the human brain, when remembering long numbers, without training or being a human memory champion/expert.

But that doesn't tell us how much, ram a computer would need, in theory, before computers have been invented.  Modern computers, easily have billions of times that number.

In other words, some approximations, are unwise to do, and use of them, could lead to extremely misleading results.

EDIT:  Also, it depends on if the solution while being used, has to respond in real-time, like a real human would.  Or if the generally intelligent AI machine, is allowed to spend many minutes, or hours, days, weeks, or even months.  Before churning out the same or a better answer, to something a human would have taken a considerably shorter period of physical time, to answer.

E.g. If an AI machine, could write a 100% working/tested, to specifications, program, which is as good as (or better), than something a decent programmer could have written.  Even if a humanly programmed project, would have taken, perhaps an hour to achieve (write the program).  Even if it takes, a few weeks to create it, by computer.  That would still be both an amazing achievement, and potentially useful, for people.

Zero999:

--- Quote from: tom66 on January 04, 2023, 07:05:18 pm ---
--- Quote from: Zero999 on January 04, 2023, 06:21:32 pm ---Going by the rate of progress, I can't see AGI being developed any time soon. Computers are already powerful enough to achieve it, but whether it's possible for them to think in that manner is another thing.

--- End quote ---

What makes you believe computers are able to achieve AGI already?
--- End quote ---
The fact that, by all objective measures, computers can both store and process more data much faster than the any human can and have been able to do so for a long time.


--- Quote --- The human brain is 'estimated' to have somewhere around the equivalent of 10^15 MIPS capability -- the difference is the brain is really an interconnected analog computer rather than a digital logic circuit. The fastest processors are still around 5-6 orders of magnitude away if we look at the capability of typical neurons and how they are connected, and how they can change their connections, in the human brain.  Even specialised processors, e.g. vision neural net processors, are around 3-4 orders of magnitude away.  Not to forget that the human brain uses about 20 watts to do that yet the Tesla autopilot computer can maybe pilot a vehicle autonomously on close to 10x that.  We still have a long way to go before we're at the same level.  And we'll probably need semiconductor die about 30cm in diameter with way more than one layer.
--- End quote ---
There's no point in making a computer which can simulate the human brain. Much of what the brain does is just survival and processing low level signals from the muscles and pain receptors. In the animal kingdom, brain size correlates better for body size and whether the animal is warm or cold blooded, than intelligence. For example, crows have comparable intelligence to many grate apes, yet have much smaller brains.

Computers have to be much more powerful than necessary to do similar things a human brain can perform easily, because they're not optimised for the same set of problems. Pigeons can perform many visual tasks, better than humans, such as identifying defects in parts and telling which artist painted a certain picture, yet are less intelligent than humans.

MK14:

--- Quote from: Zero999 on January 04, 2023, 09:25:51 pm ---
--- Quote from: tom66 on January 04, 2023, 07:05:18 pm ---What makes you believe computers are able to achieve AGI already?
--- End quote ---
The fact that, by all objective measures, computers can both store and process more data much faster than the any human can and have been able to do so for a long time.

--- End quote ---

But, computers are only doing very simple processing steps, per instruction.  Such as adding, comparing or multiplying a couple of 64 bit numbers.

The human thinking process, seems to be massively more complicated, than that process.  So just because computers seem to be faster, doesn't mean they would be so fast or faster, when performing extremely complicated thinking processes.

Speed is a complex issue, because if you discuss, allowing it to NOT run in real-time.  I.e. it is allowed 1 hour, to emulate/simulate 10 seconds of human thinking capabilities.  That would (in proportion to the ratio), allow, computers to act as if they were vastly faster and more powerful machines.

E.g. A 1 instruction per second (e.g. all relay) computer, can act like a 1,000 MIP computer.  As long as you allow it 1,000,000,000 seconds to calculate each second of the powerful computer.

I.e. If it takes a 1,000 MIP computer 1 second, to calculate PI to a million places.  The 1 instruction per second relay computer, can also calculate PI to a million places, as long as you leave it running for 1,000,000,000 seconds.

Zero999:

--- Quote from: MK14 on January 04, 2023, 09:42:13 pm ---
--- Quote from: Zero999 on January 04, 2023, 09:25:51 pm ---
--- Quote from: tom66 on January 04, 2023, 07:05:18 pm ---What makes you believe computers are able to achieve AGI already?
--- End quote ---
The fact that, by all objective measures, computers can both store and process more data much faster than the any human can and have been able to do so for a long time.

--- End quote ---

But, computers are only doing very simple processing steps, per instruction.  Such as adding, comparing or multiplying a couple of 64 bit numbers.

The human thinking process, seems to be massively more complicated, than that process.  So just because computers seem to be faster, doesn't mean they would be so fast or faster, when performing extremely complicated thinking processes.

Speed is a complex issue, because if you discuss, allowing it to NOT run in real-time.  I.e. it is allowed 1 hour, to emulate/simulate 10 seconds of human thinking capabilities.  That would (in proportion to the ratio), allow, computers to act as if they were vastly faster and more powerful machines.

E.g. A 1 instruction per second (e.g. all relay) computer, can act like a 1,000 MIP computer.  As long as you allow it 1,000,000,000 seconds to calculate each second of the powerful computer.

I.e. If it takes a 1,000 MIP computer 1 second, to calculate PI to a million places.  The 1 instruction per second relay computer, can also calculate PI to a million places, as long as you leave it running for 1,000,000,000 seconds.

--- End quote ---
We don't know how the human thinking process works. It might appear to be complicated, but it might not be that complex in reality. Some seemingly very simple systems can require a lot of power to model, using conventional computers.

Psi:

--- Quote from: Zero999 on January 04, 2023, 06:21:32 pm ---Going by the rate of progress, I can't see AGI being developed any time soon. Computers are already powerful enough to achieve it, but whether it's possible for them to think in that manner is another thing.

--- End quote ---

Yeah, I think AGI will require a quantum computer, or whatever type of computing comes after that

But I do think we'll get something in the next 20 years that "seems" like an AGI but is just a combination of systems linked together.
eg, a system that doesn't really 'think' or have any opinions or aspirations or desires, but we figure out a way to fudge it to a limited degree to aid with human interaction.
Kind of like how chatGPT seems to be more than what it actually is.

Navigation

[0] Message Index

[#] Next page

[*] Previous page

There was an error while thanking
Thanking...
Go to full version
Powered by SMFPacks Advanced Attachments Uploader Mod