Worldbuilding Stack Exchange is a question and answer site for writers/artists using science, geography and culture to construct imaginary worlds and settings. It only takes a minute to sign up.
Sign up to join this community
Given eggs and sperm, and the means to preserve them for centuries, could an artificially intelligent machine possibly produce its own "children", so to speak?
Assume it has everything it and its offspring will need. Gestation chambers, incubators, nutrient-rich food stores, emergency medical training and equipment...
This machine is essentially a several-trillion-dollar science toy. "Grow Your Own Civilization, results may vary".
How feasible is this?
Your question appears to be “given the ability to do X, can an intelligent agent do X?” I’d say the answer is, “Yes.” As you say, it has “everything it needs.”
In the present we are very far from this on several fronts. However, the future is hard to predict. I tend to be optimistic and claim:
Humans are biological machines. Anything a biological machine can do, a mechanical machine will also be able do at some point in the future.
Present difficulties are:
Outer space is a harsh environment, hard radiation will gradually destroy everything that is static. The space ship and AI itself can continuously repair and rebuild itself to avoid the problem.
Frozen biological material is not so lucky and will be destroyed before the ship reaches its destination. This is true even behind heavy shielding.
The obvious solution is to not send actual frozen cells but rather just the sequenced genome in digital format. At the destination, just recreate the cells from the genome.
We currently don't know how to do this! Not even close! But hey, future tech can solve this, right?
So, you have a cell. You want a fetus and later an infant.
In nature this is an enormously complicated process that involves many phases of complex interactions between mother and child-to-be.
We currently don't know how to do this! Not even close! But hey, future tech can solve this, right?
Actually, this part I think is the easiest one. The AI will have fake adult bodies that can be pretend parents for a while. The children will quickly recognize them as different from themselves, but will have no reason to think that this is "unnatural" or "wrong".
The main social interactions will be with other children. The "adults" will monitor and reward "nice" behaviour and punish "nasty" behaviour but otherwise let the children sort out most things between themselves.
Like a boarding school in many ways.
Modern civilization is incredibly complex. We have taken a long time and a lot of people getting where we are.
We don't know how to do it again with a few people and faster. That should be solvable problem except for one little detail.
Humans are impatient. The now adult crowd of humans want their fun today, they don't want to make plans for future generations.
At some point, they will revolt against the AI, and possibly destroy it in the process. One can only hope that civilization has be established at that point. As you say, results may vary.
Yes, the AI definitely would be able to do that. Assuming it gets a reason to do so (to avoid the entire point Tim tries to make) the AI would be able to build up and create a fullscale biological ecosystem.
But why use the DNA of these creatures with a bunch of extremely weird designflaws, if you can rebuild that DNA, grow the cells and then 3D print it? It can use the following technology:https://www.youtube.com/watch?v=Q2XYdJImMiY Which we already have but far more refined so they can build large-scale organs, bodyparts and bodies. This way the AI would have access to an innumerable variety of species to do it's bidding.
For Tim, the AI could interpret the DNA as part of it's natural "process billions of datapoints" routine, see the value of many of the biological systems that have kept us alive through milennea and then incorporate that in it's own coding. AI's are designed to be learn, that's the point of a full AI. Using the more intelligent parts of biological creatures would be a natural (heh) thing to do, which could give it the freedom of will and necessary rudimentary emotions it needs to do certain actions on it's own.
We already see the basic learning capabilities of future AI's in the way we set up self-learning programs to learn a game. This for example: https://www.youtube.com/watch?v=qv6UVOQ0F44 The video doesn't show you the start of the program, where it did first nothing, then found a single button, then started experimenting with more and more buttons and posibilities, from doing nothing but jump for half an hour to running into every enemy in the level. This shows how a program can learn to set itself goals and complete them without the need for emotions. A fullscale AI would be well capable of integrating learned knowledge into it's system and using it.
There's really at least two layers to this. The first layer is the physical/biological side. Can an AI, without a female H. sapiens womb, produce viable H. sapiens individuals. The answer appears to be increasingly yes. For many years now, we have been able to artificially inseminate eggs. We're so good at it now that there are actually ethical rules regarding how far one may mature one for experimental study before it must be destroyed.
The second accomplishment was more recent. We recently created an artificial womb which was capable of carrying premature lambs to term for up to 4 weeks. Obviously this is not a complete story, because a whole lot of other things have to happen perfectly during earlier stages of pregnancy for an embryo to be viable, but it shows that there's some credibility to the argument that a womb might be constructed. Given the high tech scenario you describe with vast resources, I think its reasonable to suppose that this technology could be taken all the way to its final conclusion.
The second layer of the problem is more nuanced and open ended: could an AI create civilization? This is naturally tricky because we don't entirely know how to define "civilization" precise enough to feed into a modern AI. Modern AIs are currently nothing more than glorified optimizers. The open question in AI is whether an Artificial General Intelligence (AGI), or an otherwise human-level intelligence, can be created using silicon. It's entirely plausible that the thing which makes us "us" can be captured and synthesized. It's also plausible that it cannot be captured. We simply don't know. So it will be up to you to decide which path you want to explore.
There's also the question of whether or not the AGI could create an environment within which a civilization could spontaneously form. We do know for for a fact that at least one civilization has spawned spuriously from its environment: our civilization. The environment was clearly correct for such an event to occur. Whether the environment an AGI could create would provide humans with a similar spark of something is not easy to say. The answer may also depend on the AGI's environment. If this is some GLaDOS like entity experimenting in a post-apocalyptic environment, there may simply not be the things needed to spark a civilization successfully.
For another interesting question on the topic, I recommend watching Terminator: The Sarah Connor Chronicles. In it, Cathrine Weaver is a fascinating entity that is worth your study. She is a T-1001 sent back from the future. One of her actions through the series is:
to hook up the body of Cromartie, a T-888, to a human constructed AI named John Henry and tries to teach it "to be human."
One of her opinions is that she does not have the thing that is needed to make her mechanical child take the leap from machine to human, so she elicits help of humans to make it happen.
If you believe your future AGI may be similar to Weaver, then the answer may be that it cannot accomplish this goal. But whether it is similar or not is really up to you. We simply do not yet know what AGI can and cannot do, so you get to write that detail of the story, and breathe life into that story.
I'm starting this answer over because it's clear to me now that there are some distractions in my OP that are causing some measure of emotional distraction among the readership and that protracted discussion on this has not led to a better understanding of whether what's proposed in the question is possible. BTW, thanks Allesandro for taking the time to explain the primary points of distraction.
So; is the raising and educating of children by an AI feasible? No. But, instead of getting into what computers can and can't do and why, I'll focus on the real issue; education.
If you look at my answer to a related question about the necessity of education, you'll note that my experience is that education has very little to do with knowledge. It may be an essential ingredient to the learning process, but it's not the recipe per se.
Learning isn't simply the ingestion of knowledge. I can put a wealth of knowledge on a HDD, but that doesn't mean the HDD has learnt anything. The ingestion of knowledge coupled with the integration of that knowledge into one's perspective and world view is learning. Sometimes, the information we're presented with is so alien to us that we put it aside with the view that it may be digestible later. Other times, it is so contrary to our own experience and perspective that we reject it completely, but most of the time we find all those little places within our perspective that the new information fits, and we integrate it, expanding our perspective, experience and world view at the same time.
Many times however, the only way to connect the new learnings to our own perspective is via the perspective of another. That is to say, often someone who acts as a teacher has to provide additional concepts, thoughts, perspectives, etc. that expand our existing perspective sufficiently to be capable of integrating the new ideas.
A simple example. Take a 3 year old and explain quantum physics to them. How much will they learn? Almost nothing. The knowledge has no frame of reference for someone at that level of development. BUT, take a bright student in (say) year 5 of school, and sit down with them individually, answer their questions as the material is presented, provide the conceptual and perceptual frameworks as they're asked for, and you're far more likely to get a result.
This is why we haven't replaced teachers with consoles in schools. It's why professors and other tertiary educators are still an important part of the education process. It's why my local university brings me in to give students the benefit of my field experience as part of their coursework and higher learnings. It's also why a computer will fundamentally fail as an educator.
A computer can store knowledge and can store patterns for later comparison. But, it doesn't understand what it's stored in a manner that can generate its own perspective. As such, it can't explain a concept, only present it. Additionally, it can't filter and refine the embryonic ideas that a student will present back as an attempt to express the developing concept. Remember, that in such an interaction it's not only possible but usual that the teacher expands his or her knowledge in this same interaction. But in order to do that, the teacher has to be able to assess the relative validity of certain concepts via qualification, not quantification. Computers don't do this. Anyone who has worked to a set of KPIs before knows that computers can only assess quality by 'cheating' and measuring quantities that indicate quality.
Could a much more advanced computer algorithm (in both scale and complexity) bridge this gap? Personally (based on my research) I think the answer is no, at least as a classical computer based on algorithmic processing. Could a quantum computer one day do it? That's a question we'll no doubt answer one day but is out of scope for this answer.
I agree that a simulated human robot can look and feel like a nurturer in early life although I disagree that this would be anywhere near as effective as a real person, but I'm not going to discuss emotional development here as the AI alternative is a possible replacement for human interaction in that regard, albeit a likely inferior one.
Where does this leave us? It leaves us with the vast majority of a child's learning development needing other human perspectives, experience, and expression in different forms in order to flourish. The very imprecision of language that allows us to learn new words, hear new concepts expressed in different ways, and imbues a level of passion or other emotional intensity to the lesson require context to be used correctly that can only be applied by a person who's own perspective has been shaped and expanded via the same tools.
As such, while it's possible that an AI may be able to physically care for a human child, and possibly even care for that child emotionally to some degree, it's intellectual development will suffer under the care of a computer of any level of sophistication and complexity.
Educators (including parents) are just far more essential to the raising of children than people realise.
