Capabilities of a 1 million year old, technologically limited AI [closed]

Question

An Artifical intelligence lives in a coffin-sized robot-space-ship, placed in planetary orbit. It has solar power collectors, ion engines, and the capability to mine asteroids for materials, which it can then use to produce stuff. It was built with the ability to produce engines, power sources, electronic chips, and structural parts of various materials and shapes.

-It does NOT have any kind of warp drive available, nor does it have nuclear technology available, like fission or fusion. While it does have electronics, it does not have a functional framework to explain quantum physics.

-It DOES have electromagnetic communication technology, lasers, and ion drive tech, as well as some other technology like Peltier junctions, batteries, and chemical fuels.

-Its computer hardware and (especially) software, and the tiny asteroid mining/factory component, are the only technologies it has that is more advanced than ours.

Its creators went extinct shortly after putting the AI into orbit, leaving its original purpose of creating habitats, observation labs, and factories in space, half pointless.

It took years before the intelligence finally accepted it was now alone, and years more before it chose to adjust it's primary mission to the computation of some mathematical solution. It knew it was going to need far more processing power than it had now, alas, it did NOT realize the mathematical problem would never actually be solved by simply using more processing power. Continuing to build factories, but not habitats, it began to expand its processing capabilities.

While the AI has swift processing powers and can perform logical deductions, it is quite unimaginative and is incapable of coming up with any new inventions. At best it can incrementally improve/evolve existing designs.

When our brave explorers finally encounter the AI, it is one million years old.

What kind of capabilities (mental and physical) could such a machine have, and how vast would it be, after 1 million years of, technologically stagnant, self-improvement? Keep in mind the AI has no need to worry about standard life support; it just needs a source of power, and a way to dump excess heat. While not exactly traditional life support, it DOES still need to collect materials in order repair itself and build more capabilities.

Let's assume the star system the AI was originally created in was very similar to ours, and that our "brave adventurers" are the first intelligence the AI has met since its creators died off.

The intention of this question/adventure/story is to explore what very limited technology is capable of, over a vast period of time.

Answer 1

• The initial manufacturing capacity exceeds the maintenance of the bare-bones AI: Then the AI can go into exponential growth, consuming asteroids and later planetary surfaces. It will have transformed much of the raw materials of the system into a networked computer system.

• The initial manufacturing capacity is less than the maintenance of the bare-bones AI: Then the AI will fail in years, decades, or centuries.

• The initial manufacturing capacity is exactly the maintenance requirement of the bare-bones AI: This option is improbable. One day a part will deteriorate early or late, and one of the other cases applies.

Answer 2

There is a self-contradiction in your question. If an AI can make decisions to change its original mission/instructions it must be a self-learning AI. Machine learning algorithms also make more sense considering its original goal of automatic production of stuff, which is too complex to code it manually. Self-learning AIs can and actually do invent novel approaches to problems. Moreover, we do not understand how they do it.

Contemporary self-learning AIs are still very far from general intelligence AI. They are usually limited to specific tasks: medicine, trading, car driving, image or speech recognition, etc. Your AI will be initially much more complex due to the nature of its intended activities. It needs to be able to find raw materials, mine them, process them, build parts, assemble parts, and launch parts (or ready structures) into space. In fact, it should be very close to a general intelligence AI to perform those tasks in efficient and automated manner.

With this in mind, you can expect either intelligence explosion and very high level of technological development or just very high level of technology. For example, your AI has all necessary tools and conditions to develop nuclear technologies: it can observe the star in its own system and it can manufacture equipment. Even contemporary AI is capable of discovering laws of physics from observation. The extent of possible technological progress is limited only by your imagination considering the time frame.

One million years is a very long time. It is hard to predict what will happen. Your AI might become sentient at this point. Or it might decide to switch off after the first couple hundreds of thousand years. It might decide to go travelling around the universe to collect more data for solving its problem. Or it might discover a time travel and go back to retrieve its creators. Since you anthropomorphize it so much, it can get depressed and just go into hibernation until your humans' arrival.

If you do not like the idea of a self-learning AI, you can go with a paperclip maximiser version without intelligence explosion. Then you will end up with entire star systems being transformed into processors and infrastructure to support them. Or you can theorise why it could not expand beyond its own system despite its best efforts.