Hello!
I have an interesting and a bit light paper, entitled "The Surprising Creativity of Digital Evolution: A Collection of Anecdotes from the Evolutionary Computation and Artificial Life Research Communities". It's indeed a collection of anecdotes, gathered from over 2 dozens of researchers, exposing surprising results of evolution in action: https://arxiv.org/pdf/1803.03453.pdf
After reading it, I came up with some questions:
- Do You think we can draw parallels between such surprising results of digital processes and real biological evolution? Or is it just our over-interpretation?
- Is there any point in research on safety of AI systems, or we will be outsmarted anyway?
- Does it mean that humanity is doomed to extinction, when faced with evolution of other entities that could, in future, be accelerated practically indefinitely?
1. Do You think we can draw parallels between such surprising results of digital processes and real biological evolution? Or is it just our over-interpretation?
ReplyDeleteIt is difficult for me to answer unequivocally, but probably the interpretation is somewhat tightened for the purposes of the article. Some of the results may actually be the result of a real revolution, but probably not all of them
2. Is there any point in research on safety of AI systems, or we will be outsmarted anyway?
Whatever problems/tasks/objectives we assign to advanced AI systems probably won’t exactly match our real-world objectives. Unless we put in an (enormous, multi-generational) effort to teach AI systems every detail of our collective values (to the extent there is overlap), realistic systems will need to rely on imperfect approximations and proxies for what we want.
3. Does it mean that humanity is doomed to extinction, when faced with evolution of other entities that could, in future, be accelerated practically indefinitely?
We're probably going into the field of science fiction. I think that some people may find a replacement for chemical or nuclear weapons in order to destroy humanity. But people also want to strive for infinity at all costs, and try to find ways to prolong life.
DeleteThanks!
When reading Your answer for question 2, somehow a short clip "Rick and Morty - Keep Summer Safe" (cartoon violence!) popped into my mind.
I agree that we will need best proxies for goals.
Thank you for this article, it does not deal with artificial intelligence and I did not know that its development can be compared to the evolution of animals. Investing in security should be proportional to the degree of development and applicability of artificial intelligence. The content of the article shows a quite high level of modern solutions and this is a sign that security issues become important and we should think about appropriate regulations.
ReplyDeleteThanks for Your comment!
DeleteEvolution, in its most fundamental meaning, is just a massively parallel search thru solution space, guided by changing fitness function :-)
1. Do You think we can draw parallels between such surprising results of digital processes and real biological evolution? Or is it just our over-interpretation?
ReplyDeleteWe should draw line between them for now of course. Digital process are not even close to biological evolution, sometimes researchers can find someting new and exiting but it would not evolve. Even when it comes to Turing complete personal assistants.
2. Is there any point in research on safety of AI systems, or we will be outsmarted anyway?
Not in our lifetime. I know that there is a lot of fuzz about AI in general, but this branch of science is going formward very slowly.
3. Does it mean that humanity is doomed to extinction, when faced with evolution of other entities that could, in future, be accelerated practically indefinitely?
Not for now. Maybe in future everything depends from us.
"Not in our lifetime" refers to us being outsmarted, I guess?
Delete1) Do You think we can draw parallels between such surprising results of digital processes and real biological evolution? Or is it just our over-interpretation?
ReplyDeleteI think overinterpretation. We can not compare biological processes with digital processes today. However, on the other hand, neural networks are probably a combination of digital and biological evolution.
2) Is there any point in research on safety of AI systems, or we will be outsmarted anyway?
I think that we (our generation) will easily live to the end of our days without artificial intelligence. Our children will have more difficult ;-)
3) Does it mean that humanity is doomed to extinction, when faced with evolution of other entities that could, in future, be accelerated practically indefinitely?
No today, no tomorrow, but the future will show if there is such a possibility.
So there is no point in sort of "preemptive" research? The void will be filled by Sci-Fi authors.
DeleteDo you think we can draw parallels between the real world and the world? Or is it just our over-interpretation?
ReplyDeleteOf course, we can compare, however, it will always be a simplified model for a simple reason that we only represent the reality to a certain extent and we do not create it. We have been doing this for many years, until we have computing machines at our disposal.
Is there any point in the AI systems research, or we will be outsmarted anyway?
As I wrote above, it will always be just a simplified model. With time, we will be able to accurately reproduce it, but it will not be 100% our model.
Does it mean that humanity is doomed to extinction, when faced with the evolution of other entities that could, in future, be accelerated practically indefinitely?
This is a difficult philosophical question, we can only digress and I can not answer this question. It seems to me that humanity as such will not be lost but will be transformed and improved as evolution has done for many millions of years.
Ad1
ReplyDeleteBiological evolution is a process that takes billions of years. It shall be based on survival statistics with regard to the conditions under which it takes place. Digital evolution is created at a completely different level of reference. Indeed, we can emulate basic behaviours on a global scale, at the same time there is a possibility to simulate behaviours of the simplest organisms or cells. But this takes place under certain conditions. We do not know yet all the indications which determine the evolution, I am afraid that there may be infinitely many of them. The process of digital evolution responds to human demand, which is a clear indicator of the direction of digital processes.
Ad2
It is difficult to answer this question. Because on the one hand neural networks are algorithms and we know how to behave. On the other hand, neural networks can create new neural networks. The level of complexity may be too great to understand what is going on in them. In addition, even neural networks distribute the information on the basis of which they are learned. I think that information control is the safety point.
Ad3
Evolution is an ongoing process. To date, 15 human species have been described as extinct. Homo sapiens' evolved from these species. Throughout history, our species will also die out, or evolve, into another, better adapted to current conditions.
But "digital" evolution (and evolution of purely digital entities, not approximated models of biological ones) can be run many times faster than biological. Don't you think this may give digitals a chance to catch up with complexity?
Delete
ReplyDeleteDo You think we can draw parallels between such surprising results of digital processes and real biological evolution? Or is it just our over-interpretation?
In my opinion present development of AI has nothing in common with unbelievable complexity of biological evolution.
Is there any point in research on safety of AI systems, or we will be outsmarted anyway?
It is very long way to this point. But probably one day it will be new area of research.
Does it mean that humanity is doomed to extinction, when faced with evolution of other entities that could, in future, be accelerated practically indefinitely?
I think humanity will lost control over AI as they had lost control over nuclear power - in Ukraine or Japan for example.
But is complexity the only difference that sets those apart? Or is there something other?
DeleteOr maybe complexity is *all* the difference?
Do You think we can draw parallels between such surprising results of digital processes and real biological evolution? Or is it just our over-interpretation?
ReplyDeleteI think that it is over-interpretation. We can't compare (yet) digital processes and real biological evolution which are much more complicated.
Is there any point in research on safety of AI systems, or we will be outsmarted anyway?
It's really hard to say. I don't know the answer.
Does it mean that humanity is doomed to extinction, when faced with evolution of other entities that could, in future, be accelerated practically indefinitely?
Human evolution could make us more addaptable to new conditions.
If we are given enough time to adapt :-)
DeleteThanks!
1. Do You think we can draw parallels between such surprising results of digital processes and real biological evolution? Or is it just our over-interpretation?
ReplyDeleteIt is difficult to say, I am not very much on the subject. However, I think this may be a misinterpretation, often scientists see what they want to see.
2. Is there any point in research on safety of AI systems, or we will be outsmarted anyway?
I don't think so. This is a very distant future, if at all, AI will be able to compete with real intelligence.
3. Does it mean that humanity is doomed to extinction, when faced with evolution of other entities that could, in future, be accelerated practically indefinitely?
At the moment, it is more fiction than science. However, I do not rule this out completely, but it will certainly take a very long time.
Yes, I also had seen examples where (in my opinion) researchers attribute too much to results of simple constraints (like geometric)
Delete1. Do You think we can draw parallels between such surprising results of digital processes and real biological evolution? Or is it just our over-interpretation?
ReplyDeleteI think that this is overinterpretation but I am not an expert in this field. The model is too simple comparing to biological evolution.
2. Is there any point in research on safety of AI systems, or we will be outsmarted anyway?
It is hard to say. Safety should always have high priority so there is always a point to make a research of safety even if it is just in case.
3. Does it mean that humanity is doomed to extinction, when faced with evolution of other entities that could, in future, be accelerated practically indefinitely?
I think it is rather scenario of science fiction story.
Do You think we can draw parallels between such surprising results of digital processes and real biological evolution? Or is it just our over-interpretation?
ReplyDeleteI am not an expert in biological evolution topic, I can't say 🙂
Is there any point in research on the safety of AI systems, or we will be outsmarted anyway?
Research is always a good option, thinking like that we could sit here and do nothing, waiting for a catastrophe.
Does it mean that humanity is doomed to extinction, when faced with the evolution of other entities that could, in future, be accelerated practically indefinitely?
I think that there is always a possibility like that, still: I don't know.
Thank you for the paper, I'll read it more thoroughly later. Now I can say that comparing biological evolution to optimal outcomes from simulations looks like over-interpretation. After all environment used for that digital process/reinforcement learning to be more specific is not complete and takes into consideration much fewer variables than are in the real world.
ReplyDeleteI don't believe that any reaserch should be artificially constrained because of the fear that we discover something that can perform better than we. For the same reason we could stop invention of a steam machine as it outperformed human workers. Maybe if we had strong AI then would became much better place rather than changing into the hell where humans are prays of machines.
From the point of view of its inhabitants, the simulated environment is complete :->
DeleteI also agree there should be no artificial constraints on research.
Przemek,
ReplyDeleteThank you for interesting topic. I like such crazy scientific approaches and comparisons ;)
I am open to such conclusion. It can be possible, that such digital and technology evolution plays with the traditional one. I believe the whole world works in pulsing rhythms, cycling again to the same points again and again. The history shows such tendencies, that everything must go in the same order, the difference is put in the detail only.
As for the second question, there is always point in conducting research, because in the science history there were many surprising results. And in my opinion, we will not be outsmarted anyway, I hope so and I keep on believing in that.
And I don’t think we are doomed to extinction. I truly believe in a deep sense of the cycle of life we are in. I hope there is higher order deciding and interfering when it is needed and justified.
BR,
Marta
As for me, I don't know if I should hope we won't be or we'll be - but only time will tell.
Delete1. Do You think we can draw parallels between such surprising results of digital processes and real biological evolution? Or is it just our over-interpretation?
ReplyDeleteYes, why not? These are just parallels…Sometimes the course of certain processes will be even identical (because that's just the way things are).
2. Is there any point in research on safety of AI systems, or we will be outsmarted anyway?
There is such a probability but there are methods that improve security and cannot be outsmarted in a short period, for example entering complex passwords or multi-stage authentication. However, I suppose that in the future AI will paradoxically create security methods against other AI systems.
3. Does it mean that humanity is doomed to extinction, when faced with evolution of other entities that could, in future, be accelerated practically indefinitely?
No, I don’t think so. To my mind various entities can exist together. Evolution of human being theoretically could be accelerated indefinitely but practically it doesn’t happen. I am rather full of hope in this matter.
Thanks! I haven't thought of this aspect of safety.
DeleteThanks for such interesting article.
ReplyDeleteDo You think we can draw parallels between such surprising results of digital processes and real biological evolution? Or is it just our over-interpretation?
In my opinion we can. These are models and parallels seem to be good assumptions. However, it is difficult to say because I am not a pundit in this field.
Is there any point in research on safety of AI systems, or we will be outsmarted anyway?
In my point of view it will not happen in the near future. This field of research is moving really slowly.
Does it mean that humanity is doomed to extinction, when faced with evolution of other entities that could, in future, be accelerated practically indefinitely?
I don't think so. It seems like a doomsday scenario but I don't think it will happen. It is more a fiction than a science.
Do You think we can draw parallels between such surprising results of digital processes and real biological evolution? Or is it just our over-interpretation?
ReplyDeleteHi, I think we are far away to draw parallels between them. So the answer from my perspective is over-interpretation. We try to watch standard evolution and find mathematical patterns included in AI - but we are not able to understand a lot of processes that occurs in nature.
Is there any point in research on safety of AI systems, or we will be outsmarted anyway?
Unfortunately I am not able to give you the answer for this question because - I do not know. I only can assume that point exists.
Does it mean that humanity is doomed to extinction, when faced with evolution of other entities that could, in future, be accelerated practically indefinitely?
It sounds like a scenario from Sci-Fi movie ;). I will be a little bit optimist - and I think that we will survive. :D
We can probably distinguish whole genres of movies - the topic is very catchy
Delete1. Do You think we can draw parallels between such surprising results of digital processes and real biological evolution? Or is it just our over-interpretation?
ReplyDeleteI think that this is our overinterpretation. Digital processes are a simplification of biological ones. Just look at artificial intelligence.
2. Is there any point in research on safety of AI systems, or we will be outsmarted anyway?
In films AI outwitted human and sought to destroy him. Research is important and the topic of AI is interesting. It is difficult to answer your question. It's easy to overlook the point where you should stop.
3. Does it mean that humanity is doomed to extinction, when faced with evolution of other entities that could, in future, be accelerated practically indefinitely?
This is a black scenario, but I don't think it will come true.
Do You think we can draw parallels between such surprising results of digital processes and real biological evolution? Or is it just our over-interpretation?
ReplyDeleteI think we can to some extent. It reminds me a little bit of history related to Fibonacci and Golden division.
Is there any point in research on safety of AI systems, or we will be outsmarted anyway?
I've heard stories of quite serious visionaries that are catastrophic. If we're talking about artificial intelligence. Personally, I'm skeptical.
Does it mean that humanity is doomed to extinction, when faced with evolution of other entities that could, in future, be accelerated practically indefinitely?
Personally, I'm skeptical.
You mean like the spirals, "popping out of nowhere" in things like sunflower seeds patterns, while being consequence of one trivial rule of growth to spawn new element as far from the others as possible?
Delete1. Do You think we can draw parallels between such surprising results of digital processes and real biological evolution? Or is it just our over-interpretation?
ReplyDeleteI think that this is overinterpretation but I am not an expert in this field. The model is simple and comparison to biological evolution seems not to be accurate.
2. Is there any point in research on safety of AI systems, or we will be outsmarted anyway?
Safety should be number one priority in every research especially one taht could have big impact.
3. Does it mean that humanity is doomed to extinction, when faced with evolution of other entities that could, in future, be accelerated practically indefinitely?
I do not think that machines is our biggest problems rather ourselves and envrionmental changes
Do You think we can draw parallels between such surprising results of digital processes and real biological evolution? Or is it just our over-interpretation?
ReplyDeleteWe can't and we are still far from overinterpretation to a proper interpretation of results. Maybe if we manage to make quantum computers working as expected. One of our colleagues has posted an article about caffeine particles and so on. Feel free to read it.
Is there any point in research on safety of AI systems, or we will be outsmarted anyway?
Someday, but not in the nearest future. Probably our children will see something more advanced and sophisticated. Maybe you are mentioning https://en.wikipedia.org/wiki/Three_Laws_of_Robotics ?
Does it mean that humanity is doomed to extinction, when faced with evolution of other entities that could, in future, be accelerated practically indefinitely?
We are evolving as well. If you take a look at human history, we have managed to evolve into decent form. However, if we make a mistake and let AI decide whether something is good or bad for us it can lead to a very bad scenario.
No, 3 (4) Laws are much too ambiguous and expressed in human language. Require from the AI some sort of will to obey the laws and to constantly re-evaluate its actions against possible interpretations.
Delete1. Do You think we can draw parallels between such surprising results of digital processes and real biological evolution? Or is it just our over-interpretation?
ReplyDeleteI'm not quite sure. I guess we cannot be 100% sure if the results are really related to evolution or if it is our interpretation or misinterpretation or rather wishful thinking.
2. Is there any point in research on safety of AI systems, or we will be outsmarted anyway?
I agree with the previous speakers, we should live this problem to our children and grandchildren, as we won't live long enough to see the real development of this scientific branch.
3. Does it mean that humanity is doomed to extinction, when faced with evolution of other entities that could, in future, be accelerated practically indefinitely?
This sounds a lot of like science-fiction to me - Terminator scenario.
But is there something more in evolution (its core) than selective pressure + imperfect replication?
DeleteYes, we can draw such line. If we think about evolution of IT systems, AI, etc. than we can see that connection between digital processes and biological evolution is very strong. Of course, there is a point. We can not work on AI and forget about safety of those systems. In terms of IT safety is one of the most important elements. AI may be one of the most dangerous inventions of humanity, we can not forget that such system must be safe. The question is to how we will define safety. Should we make it safe or should we teach AI how to become safe.
ReplyDeleteNow. The strength of humanity is the fact that we can adapt to every situation. I’m sure that we will adapt also to described situation and we will survive.
1. I think that evolution involves everything, even language, or mems, so why not digital processes?
ReplyDelete2. Right now even when I think just about an example of the AI, which is using AI autonomic cars, I can see a problem with that (basic moral dillema - in case of possible collision will it be better to drive the car in such direction that it can cause the death of the person sitting in the autonomic car but it will save a lot of other people, or is it better to save the driver at all costs), so I am sure there is a point in researching on safety of AI systems.
3. I think , when we talk about future , we have common path with AI.
I think sooner we will be involved in immortal bio-cybernetic spieces supported by artificial bodies, then than would be a war with the machines. And from the other side we are not food for the machines, we are not their adversary in fight of resources. Just look at us - we like cats for no reason, even though they are less intelligent than us and we must clean their cuvette and feed them.
Also many of people fight with cruelty to animals - the more the country is developed, the more people are aware of animals rights. Moreover, more and more people give up eating meat.
So maybe there are no reasons why new intelligent conscious spieces would want to kill us.
1. Do You think we can draw parallels between such surprising results of digital processes and real biological evolution? Or is it just our over-interpretation?
ReplyDeleteWe can learn from such simulations. But much of it maybe just projection on part of the researchers, and looking for deeper meaning everywhere.
2. Is there any point in research on safety of AI systems, or we will be outsmarted anyway?
Well, even if we eventually will, it's better to live a little bit longer than shorter.
3. Does it mean that humanity is doomed to extinction, when faced with evolution of other entities that could, in future, be accelerated practically indefinitely?
There is no 'idenfinitely' in nature. But unfortunately, that doesn't preclude we won't be extinct sooner than we could if we didn't pursue some of our discoveries. Especially ones that could have 'a life of their own'.