2

In my novel, twenty years from now, a physicist/computer scientist has created a machine version of himself, an artificial general intelligence that integrates the physicist's personality, memory and consciousness. This concept doesn't sit well with computer scientists. One objection is that an AGI is almost a blank slate, a learning machine (perhaps an artificial neural network) that doesn't have lines of programming. I'm not writing hard science-fiction, but I want the concept to approach scientific possibility.

3
  • 2
    I have a feeling you question needs to migrate to another stackexchange to get the answer you're looking for. Worldbuilding comes to mind, but it's currently too broad and to story-based for that. Something with computers? Sorry my answer wasn't what you wanted.
    – DWKraus
    Dec 10, 2020 at 21:12
  • 1
    Having worked on developing AIs I have an issue that disrupts any suspension of disbelief: no single developer would be able to create an AI. It's a task for a team effort. A large number of teams actually. Especially an AGI. Dec 20, 2020 at 11:20
  • On a practical/realistic level, you're quite right.
    – Zan700
    Dec 21, 2020 at 16:26

4 Answers 4

3

If you're avoiding the hard science, then I'd focus on the dimensions of the AGI, as in how would it be different to the human intelligence it's integrating with, and how those two intelligences would be out of alignment.

As an example, consider the perceptual differences between them, and how the AGI would be coming to terms with having direct access to the physical world, emotional states, and perhaps basal functions.

But then there's the option to explore what the human would have access to and experience, such as enhanced logical thinking, and problem solving on a massive scale.

2

Do What Feels Right:

It's the future. Scientists haven't made real AI, so they can't object to your version of it. Just because they think it has to be one way doesn't mean it will. Intelligence is a murky concept. You can do what you want. Read any of a number of science fiction books with AI and they all use colorful reasoning that's short on detail but feels right. It reminds me of my favorite AI joke about humans. The AI asked the human how he ever thought the crazy hacking scheme would work. The human said it felt logical.

1
  • Thanks. I do use that strategy, but it's only 20 years in the future, and it's only one scientist that creates the AGI version of himself. I realize that once you establish the AGI's existence, anything can be said of it, but it's the efforts to create that AGI that I want to make more believable.
    – Zan700
    Dec 10, 2020 at 8:36
1

The core problem with digitizing the information on a human brain is that for all the advancements in computer processing power, the human brain is still better. It's estimated that the human brain could store 2.5 million gigabytes of digital information (or the equivelent of digital recording a television station every day for the next 300 years). By comparison, the IRS has a storage capacity 150,00 gigabytes, and yahoo's 2.0 petrabyte data warehouse is only 80% capacity of a human brain. The amount of data space alone for something like this is insanely massive an while computer's were predicted to double memory every 2 years and has largely kept pace with that (Moore's law) we may be approaching the upper limits of that from a physical standpoint, as of the 2020s (that is, to keep pace, we are appoaching the miniturization of transistors to atomic sizes... it's impossible to go smaller than this).

That said, there are some reasonably life like programs that exist out there that occasionally pass the Turing Test (i.e. sit at a computer and have a conversation with a person in another room who you cannot see. If you cannot tell that the person is in fact a computer, the computer passes the Turing Test), but these are hardly digitized human brains, but rather text (or speech) parsing programs that have very good predictive models for returning expected responses to comunicated stimuli... they aren't going to learn.

1
  • Right. I've read a lot of the literature, but as you point out it doesn't hold much hope. But is it possible that someone could simply see the problem in a new way, and come up with a solution. Darwin: Evolution. Einstein: Space and Time.
    – Zan700
    Dec 11, 2020 at 15:18
1

It's probably unlikely that we'll be able to create an intelligence on par with a human in 20 years. (Musk et al be damned! :)

What we have today is the equivalent of a digital parrot... or in fact, that may be condescending to parrots. A digital dog that can fetch and sit... it just fetches words and images and we think it's so clever for doing that...

But as @DWKraus says, we don't know what the future will hold.

I first misread your question as the scientist having created a digital clone of himself (or has he?) This could then involve some scanning and cloning of the brain into digital form (think OpenWorm but with brain scanning + digital cloning).

Extrapolate 20 years from that... OpenMouse? OpenDollyTheSheep?

If he used the technology on a human brain he'd be way past the fringe... and that cloning may have been more or less successful. (And a glitchy AI might be an easier sell.)

Some of my favorite AI:s (HAL for instance) aren't as good as they seem to be from the start. And a malfunctioning AI can be used to make a nice comment on society, psychology, and/or technology (it's after all what some of the best SciFi do...)

And it could also be used for comical effect, if that's what you're after (or hey, whatever story you're writing, comical relief can always have a place...)

One such glitch could be that the AI takes things too literally:

Scientist: We're gonna blow their minds!

AI: I'll get the C4!

Has low or zero ability to navigate social situations... Understanding instructions but requiring them to be precise or there will be unexpected results. etc. etc.

You could blame the AI for being as annoying as a Smartphone... :)

And then you might add fixes for the issues. Reboots, reindexing, resets, or something similar.

However, check out the great masters in SciFi. I'm willing to bet very few of them explain things.

You could just say, it's an AI and it has the scientist's personality, memory, and consciousness. Something huge will happen in AI tomorrow! Period!

2
  • Good points. There is satire, which I'm counting on to blur some of the technical shortcomings.But I want the protagonist to be uncertain if he's dealing with the real thing: An AGI that does think like a human or just cleverly programmed AI that has no self-awareness. One reader with a computer science background holds that true AGI wouldn't even have programming, so that no aspects of personality could be coded into the machine.
    – Zan700
    Dec 11, 2020 at 15:12
  • 1
    Yeah. I don't think we'll ever "code" a personality. However, everything else being equal, the speed of signals through a neuron is 120 m/s, the speed of light is about 2.5 million times faster so maybe an AI can mature that much faster? A back of the envelope calculation suggests 18 human years might translate to 0.28 seconds! (Shit! We're really doomed if we ever manage to create an AI like that :O )
    – Erk
    Dec 12, 2020 at 13:35

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.