Hi everybody!
I would like to share with you an article that seemed extremely interesting to me!
It tells how our brain analyzes the remembered information, seeks analogies and explanations for the phenomena we encounter every day (e.g. how a sequence of events "spilled fluid"->"broken vase"->"dog" is remembered. According to the theory of imagination, such a sequence should be remembered exactly the opposite).
Article (shortened version):
https://deepmind.com/blog/article/replay-in-biological-and-artificial-neural-networks
The above research will result in a better understanding of human intelligence and the creation of an algorithm that brings us closer to its more faithful reproduction.
I am interested in your answers to the following questions:
1) Have you ever remembered a situation that did not happen (and you were sure it happened)?
2) Are you convinced by the presented theory of imagination in which the brain distorts the remembered information in order to keep it consistent? Why yes/no?
3) What else do you think is not reflected in the artificial intelligence algorithms currently being processed?
4) Does it make sense to introduce such improvements (distortions between events observed by the AI algorithm)? Why yes/no?
1) I think that some of my childhood memories are a mystery to me, because I remember certain situations very precisely, e.g. telling some imaginary stories to my friends in kindergarten, but when I think about it as an adult, I have doubts whether a few years old child would come up with such precise, complicated stories about dark castles and assassins and actually told them to someone later. I also remember a few realistic dreams that felt like they were real, but I know they were just dreams.
ReplyDelete2) Yes, it convinces me. The brain is a complex mechanism and every second it receives different stimuli that it must process and analyze. We know from neurobiology that the brain processes only part of the data from the environment and rest is completed from memory so that it does not boil over the amount of data processed. We can't remember all the events very accurately, so it makes the most sense to distort the data to keep the events consistent. It's like summarizing a book in several points. It is easier to remember and the rest can be added based on stereotypes and certain patterns.
3) The brain. It is still an unsolvable puzzle for us. We are able to program several neurons connected into a single structure, but we are not able to model how the entire brain works.
4) I think so, because we are trying to create algorithms that are as close to the actual behavior as possible. Adapting AI to disturbed, distorted events so that it is able to draw correctly flowing conclusions from them is very important. There is increasing talk of using data with noise to teach AI. I think that by learning on diverse data sets that do not only contain specific patterns / events, AI learns to better adapt to unexpected situations and to better generalize its knowledge.
Thank you very much for your comprehensive answer. I also have similar insights to my childhood memories :)!
DeleteAs you mentioned, the brain is a mystery to us. I am curious how we will use the potential of the simulated mind (in the future when it will be possible).
1) Have you ever remembered a situation that did not happen (and you were sure it happened)?
ReplyDeleteRecently, I had this situation when I was shopping at a cosmetics store. Among other things, I wanted to buy a make-up powder in the store. The next day, I wanted to use this cosmetic and reached into the shelf where I keep my cosmetics. The powder was not there and I was sure that I bought it and started looking in the flat thinking that I had put it in some other place. Unfortunately, I couldn't find it in my flat, so I thought maybe it fell out of the bag when I was returning from the store, but the bag did not have a hole. I did not take the receipt, but fortunately all my purchases are saved in the application of this store and I checked that I did not actually buy this powder but was convinced that I did.
2) Are you convinced by the presented theory of imagination in which the brain distorts the remembered information in order to keep it consistent? Why yes/no?
The human brain is a very complex organ of the human body. Research on the human brain has been going on for years and will certainly continue for a very long time. There are still areas in the brain that have not been studied by scientists. The theory presented in the article is surprising. The authors of the article mention that the obtained results can be a promising style of reproduction imagination in continuing research on artificial intelligence. The results seem shocking, but not everything has been discovered yet, and science should shock at times.
3) What else do you think is not reflected in the artificial intelligence algorithms currently being processed?
I think it sometimes happens that scientists use artificial intelligence like a black box into which they put data and that black box returns some data to them. Artificial Intelligence is based on advanced math that not every scientist understands in detail. Scientists do not always try to explain why these results came out and not others, for example why we obtained such weights and what they mean for the model. Some people just quote the end results. I think image recognition could be given as an example.
4) Does it make sense to introduce such improvements (distortions between events observed by the AI algorithm)? Why yes/no?
I am not a specialist in this field, but it seems to me that if this approach gives promising results, why not use it for further research. Sometimes not everything in science is obvious at once. At first, it may seem nonsensical, and after another analysis, it turns out to be understandable, explainable and groundbreaking.
Nice story with this powder :D!
DeleteIt only shows the scale of distortion of our memories.
I just want to add that I am curious how such a mechanism would translate into modern ML approach? I mean, if algorithms started to "add" elements that were not in the picture, what would be the trust for such a detection algorithm?
Alternatively, this mechanism could be used to force the ML algorithm to focus on a specific fragment of the image depending on the context.
1) Have you ever remembered a situation that did not happen (and you were sure it happened)?
ReplyDeleteYes, I think I experienced this situation, but as you know it’s blurry that’s why I’m not sure that did happened. I can recall more than one memory that could be a movie scene or really happened to me. Another similar thing is connected with photos. I’m not sure if I remember the original situation that was capture on a camera strip or I imply this memory by listening to a story told by my relatives while looking through family albums.
2) Are you convinced by the presented theory of imagination in which the brain distorts the remembered information in order to keep it consistent? Why yes/no?
We are still learning about how human brain works, I think it’s really interesting and deep subject. Not every theory might have sense at the beginning but in case of the one presented in the article it seems reasonable that human brain distorts information to make a better sense from it. I think it’s easier to remember things that way and learned from experiences that have logic consequences.
3) What else do you think is not reflected in the artificial intelligence algorithms currently being processed?
Even though that artificial intelligence is being built with intention to mimic human brain we still have much to work on to reach that goal. Theoretically while we know more and more about brain we could implement more sophisticated algorithms, but it’s not always so easy to do. AI works with so less complexity than human brain, so I can point out a few differences for example: abstract thought that could find not obvious connections, emotions that change the remembering process, learning from past experiences that are not necessarily connected to current problem.
4) Does it make sense to introduce such improvements (distortions between events observed by the AI algorithm)? Why yes/no?
Definitely yes, I think we should use our knowledge about how human brain is functioning to improve performance of AI algorithms. While this is proven to be important pattern of human learning process it might also bring big benefit to working of AI.