Could advanced AI decide to kill all humans?
This is a very, very controversial topic. But here's my opinion on the topic:
AI won't harm humanity (at least, to the level that a lot of people think it will).
As Kartik mentioned in his answer, AGI (Artificial General Intelligence) is the concept of AI that's become so "generally" intelligent, that it can do everything we as humans can, and more, but do it better, more efficiently, more accurately, you name it.
I'd say that AGI is (so far) a fictional scenario. Why? Because of the physical limitation of computing that wouldn't let us get there. In order to simulate natural, physical intelligence, you wouldn't be able to use gradient descent, you would need to simulate physical interactions of neurons. Simulating even a single molecule on the fastest supercomputers takes a very long time, because the overhead introduced with taking the real, physical world, transposing it onto mathematics, and running those mathematics on electronics, are too much. So simulating the complete working of just a small cluster of neurons would be almost infeasible.
Then you might ask: "what if we were to get enough computing power in the future?". It's a valid question, however, there's one problem: we're already (possibly) quite close to the physical limit of how small we can get our transistors before Quantum Tunnelling kicks in.
So I'd say it's impossible because of a few key reasons:
- Current algorithms are intrinsically guided by human-generated data (even reinforcement learning algorithms need human-engineered rewards to understand what's good and what's bad). Therefore, there's no such thing as a self-guided objective for machine learning today.
- Computing power isn't there yet, and most probably, will never get near the level required for AGI.
- Humans have an innate bias to "humanize" non-living entities. Language, for example, is a great representation of the way our minds work on a fundamental, subconscious level. When we say "my computer isn't co-operating with me", or something along those lines, we don't realize that computers are big calculators with a screen and lots of predefined operations to run.
One thing I don't disagree with: Machine Learning will replace a lot of jobs - but not nearly as many as you may think! It'll take time - if Tesla eventually comes out with a fully-autonomous semi-truck, it's not like the whole industry will adopt to it overnight. Humans, and our genome, will have time to adapt and evolve overtime. We'll free up more people, over time, to get into fields like healthcare, education, and entertainment. When we don't have the option to get jobs like cashiers, drivers, etc., we'll be forced to get more advanced jobs.
This is just how technology has worked, and will continue to work for as long as humanity exists.
People, however, can use AI/ML for negative purposes and bad things. Face swapping, automated social engineering, etc., are all great examples of this. But with powerful unregulated/illegal technology, comes even more powerful regulated/legal technology.
As pointed out in the comments, transistors are not the only solution. There are alternatives like Quantum Computing, that enable you to remove that mathematical overhead, and directly run your operations on qubits. However, there's still a lot of work to be done to see if technology like AGI or self-guided machine learning is possible even with quantum computers.
None of this is certain - anything I've just mentioned can change in the next few weeks, months, or years.
Artificial General Intelligence (AGI) is machines being able to do any intellectual task that a human being can do, probably better than human beings.
Such machines could possibly out-compete inferior humans in every field possible may it be in jobs, arts, politics. Like Yuval Noah Harari often mentions, this will create a new class of people called useless class, who cannot perform anything better than machines. This is definitely a very big challenge for human beings that co-exist with machines that have AGI. But, like always, it is possible that human beings will evolve and can embrace this challenge.
There are lot of predictions on how the society will be with the arrival of AGI, and the predictions span many possibilities. Some are not ready to agree that AGI is possible. Some agree that AGI is possible, but not in near future (may not be in this century). Some predictions are very scary though saying AGI will be a very big scare to mankind. Anything is possible, and we have very limited information to predict the future at this point of time.
Coming to the question of whether AGI will decide to kill all humans? It is quite possible. It all depends on how our thought and technical leaders decide on how we want to pursue AI. That is the reason why a lot of discussions/debates are happening now a days about ethics in AI.
Many good movies/series also tried to depict how societies will look like with AGI. Some of the movies i like and able to relate with knowledge i have are
The Matrix Trilogy.
Person of Interest.
Seeing these movies and giving deeper thought, you can find your own answer. Also watch videos and read books by Yuval Noah Harari, for more information.
But, one thing i can surely predict is that, we are a long way before we can realize AGI. We are still making baby steps in that direction. But, fact is we have already started in the right direction.
Experienced people on this subject sometimes start explaining the Fermi Paradox. How is it possible that given the apparent huge size of the universe, we still have no evidence or other human life?
There are many ways to respond to this "paradox". One of them is that always that a civilization reaches some level of technology, it destroy itself.
On the other hand, there are personalities such as Ray Kurzwail (director of technology of Google), that mantains that we may reach the point in which, instead of destroying ourselves, we create something more intelligent than us, which at the same time, could create an intelligence bigger than itself (see his book, Singularity, which talks about the hypothesis with the same name). Given that, we could just replace natural selection by conscious improvement, and expand ourselves spread in the universe very fast.
We can not know which is the future, but in my opinion we are not in a stationary point but going towards one of these destinies. The autodestruction of the humanity is plausible, but who knows if putting good constraints on artificial intelligence, and using it appropiately, we can follow rather the pathway defended by Kurzwail.
In the case in which artificial intelligence becomes one of the causes of humanity aniquilation, this could happen because some AI with enough resources to do it becomes uncontrolled. But I think that it is more plausible to think that it would happen because of a human decision, and rather the AI be just the tool employed to do it. I think that because there is a debate that uses to go along with this one, and it is whether technology will be enough to overcome human needs, or population growth will always take the lead (see the Malthusian trap), and till now, the last option is the one which has occured.