Hi
everyone,
First
of all I would like to wish you all the best in the New Year :)
I
found for you a very interesting topic. I know that AI subject has been already
discussed in this semester, but recently the issue of AI has become so popular
that I took the liberty to publish yet another installment of the AI saga;
namely “The top A.I. breakthroughs of 2015”
Which
of these achievements do you think is:
1
- the best ?
2
- the most promising ?
3
- the most interesting ?
and
the question asked by the author:
-
“ (…) even more exciting are the developments we don’t expect.” – what
these could be in your opinion ?
This article was very, very interesting and amazing. I love modern technology but I have never heard about such advanced works about developing new skills in AI. When I finished reading this article first I thought about the movie called “Terminator”. Everybody knows that movie and I will never forget about Arnold Schwarzenegger’ role as “bad” AI. After the first delight about breakthroughs described in the article I felt fear about humans’ future. I am sure that in the future people or robots will use AL technology against other people as a weapon or even something worse. Therefore development in AI technology is amazing for me. I cannot imagine our world in 2100. This is almost in the bag that future will be connected with the AI in all parts of human life. Very different skills are described in this article and without deep cognition itis difficult to choose the best or most promising solutions. In my opinion all the five categories are very interesting and will be useful in a lot of issues in the future. For example Dreaming Up Visions AI will be able to create new clothes according to our suggestions, AI from Intuitive Concept Understanding part will be able to search criminals via Internet etc. According to the question asked by the author: “ (…) even more exciting are the developments we don’t expect.” I care about future a little and I generally don’t like unexpected situation and things. Personally, I think we must be very, very careful about rampant development in AI technologies.
ReplyDeleteI agree with you; I share your concerns. It was mentioned already in past blog articles on the AI and you probably know the opinions of Elon Musk or Stephen Hawking on AI. Below I include Bill Gates interesting opinions on this topic http://qz.com/335768/bill-gates-joins-elon-musk-and-stephen-hawking-in-saying-artificial-intelligence-is-scary/
DeleteThis comment has been removed by the author.
ReplyDeleteThis comment has been removed by the author.
ReplyDeletePlease have a look at our post
ReplyDeletehttp://konwersatorium2-ms-pjwstk.blogspot.com/2016/01/week-6-04-10012016-openai-as-way-to.html
It look like that 2016 can be even more exciting than 2015 in terms of IA.
Going back to 2015 The top A.I. breakthroughs. For me:
1 - the best ? - Intuitive Concept Understanding
2 - the most promising ? - Abstracting Across Environments
3 - the most interesting ? - Creative Abstract Thought
But the most the real challenge in AI is to build something really useful for ordinary people.
Following Mark Zuckerberg thoughts
(https://www.facebook.com/zuck/posts/10102577175875681):
“challenge for 2016 is to build a simple AI to run my home and help me with my work”
I think for Mark Zuckerberg to create a smart home will not be particularly demanding task, given the current development of smart homes idea. To create an entity like Jarvis is a matter of a very short time. 3 days ago the Consumer Electronics Show in Las Vegas took place, during which a lot of attention was payed to the idea of smart houses, est. below
Deletehttp://www.cnbc.com/2016/01/06/ces-smart-homes-of-the-future.html
This comment has been removed by the author.
ReplyDeleteA.I. is very exciting and interesting. In this area we can achieve significant results that will give us possibility to create specialized solutions. Let’s imagine factory. Director have to cooperate with departments to oversee business. Given that this is time-consuming (data collection, reports) we can use A.I. efficiency to deal with issues. I do not want to decrease experience and importance of knowledge in management. I just want to take a look on A.I. as a another tool, which we can use in decision making process. A.I. based solution has an access to real time data without big effort. We should hold back on a moment on the previous sentence. We can find there a role of system user. People have to ensure proper data quality, or acceptance threshold otherwise decisions, which will be suggested by system are based on incorrect data.
ReplyDeleteCategories of intelligence described in article have an areas where they are better then the other. Each of them is promising and interesting. This one, which will decrease my effort will be the best.
“ (…) even more exciting are the developments we don’t expect.” – what these could be in your opinion ?
I am waiting for this developments, but I can agree with ZC - we should be careful. I hope results will be usable. (take care about security, improve quality of life, or just new entertainment source)
Thank you for your opinion, I am also waiting for development of AI :)
DeleteI think the best and most interesting achievement mentioned in the article was Planarian Regeneration Model Discovered by Artificial Intelligence. I really believe that in this area will be most useful. By the way mentioned example is really impressive. Machines can really help us in learning process based on exploring large collections of data because they have incredible power to do that. It's difficult to predict whether it will be the most promising but I think it can be. Everything depends how they will continue research. Moreover I really like the article 'Training artificial intelligence to write letters' its really intereseting new approach closer to the human method of learning. I think we can expect that researchers will use the neural networks which recognizes images in other incredible ways too.
ReplyDeleteRegeneration Model Discovered by Artificial Intelligence seems to be a very interesting area. I hope that soon we will be able to read about the progress in this area of science.
DeleteThanks for the article, it's a really impressive summary.
ReplyDeleteI actually agree with Przemysław when it comes to the most interesting achievement. I've also read the corresponding article about solving flatworm regeneration mystery (http://motherboard.vice.com/read/artificial-intelligence-solved-the-mystery-of-flatworm-regeneration). AI making its first scientific discovery and solving a problem which wasn't tackled over years of regular human research in 42 hours is a huge thing!
I also really liked the AI built on DeepMinds technology which was able to learn to play many different Atari games well. Still it's hard for me to image what can come next. I'm really looking forward into AI improving our daily lives.
This is yet another very interesting article about AI. Thank you very much for the link and your answer :)
DeleteArtificial intelligence refers to the development of machines and software that simulate human intuition. The field can be divided into two broad categories: efforts focused on interpreting sophisticated input (such as speech, emotions, or coordinated movements) and those aiming to recreate higher-level cognitive capabilities like learning and decision-making. Already, AI-powered systems from antilock brakes to Siri have come to surround us in everyday life, and recently these examples have grown in ambition and scale to include drones, self-driving cars, and humanoid robots.
ReplyDeleteI agree with you, the current development of technology AI is very dynamic; Siri is still unperfected but day by day is more useful and the same situation is in the area of self-driving cars. However, you did not answer my questions with regards to which technology you find the most interesting and promising. Maybe you think that the idea of self-driving cars is the most promising?
DeleteThis comment has been removed by the author.
ReplyDeleteHi,
ReplyDeleteIt is very interesed article , I am read this with passion but
I am agree with Marcin Krysinski about
the best, the most promising AND the most interesting achievements .
Thank you for your opinion :)
DeleteHi,
ReplyDeletefor me most interesting is Deep Mind's network, but as far as I know, DeepMind plays games since at least 2014. For long neural networks were able to process one type of signal or even one set of training data, for example handwritten digits (MNIST) or pictures in number of categories categories for supervised learning (CIFAR). In my research I'd like to model a network that will handle not only multiple types of video games but that will handle multiple signal types, namely visual and auditory stimuli.
I think that now that we have a lot of GPU computing power useful in AI computations and there's a lot of people around the world doing AI research, next year will come with more development in this area. I'd guess that further improvement of deep neural networks with reinforced learning.
I share your opinion about deep learning and in particular the use of GPU augmented processing. However, for your information there is already a neutral networks technology, called CNN (convoluted neutral networks) that allows both - multiple input signal types and looped in feedback for reinforced learning. Maybe you should look for article on CNN.
DeleteWhich of these achievements do you think is:
ReplyDelete1 - the best ?
Time will tell. I don't want to judge these achievements. On one hand they belong to different categories, on the other many of them could be used together to reach a new frontier. One thing I must admit: all of these achievements seem promising.
2 - the most promising ?
I don't think it will be a surprise to anyone. I think the most promising is the one I posted few weeks ago: "human-level concept learning through Bayesian program induction from a single example".
3 - the most interesting ?
I don't think I can answer this question. I can't see clear winner in this category. I want to follow-up and read on most of these advancements.
4 and the question asked by the author:
- “ (…) even more exciting are the developments we don’t expect.” – what these could be in your opinion ?
Can we expect the unexpected? :)
I think we can expect someone to combine some of these into one single research. In some cases (for instance Bayesian...) extrapolating the achievements onto new fields might lead to a breakthrough.
It is an exciting time for Machine Learning.
In general I agree with your statements. However, as I mention above there are several more advanced NN technologies, for example CNN and CNN with feedback, that are surpassing achievements of Bayesian networks. I advice you to read on it, if you want to :) as far as expecting the unexpected, we could be very suprised by unexpected early spontanious arrival of true AI.
DeleteKinga, I know about CNN's, RNN's and DNN's in general. It seems to me that you think that when I mention Bayesian Program Learning, you see some kind of HMM or some kind of Bayesian Classifier. Sure, properly trained NN's are way better than simple Bayesian classifiers, or even Hidden Markov Models (or mixtures of HMM with something else, say Gausian Mixture Models, that is GMM-HMM).
DeleteHowever, when it comes to BPL, it is not necessary so. BPL is a new kid on the block and it show performance on pair with DNN's and in some cases surpassing any DNN (even with discriminative training, and so on). It is especially the case with One-Shoot Learning, which was presented in the paper I quoted.
Obviously, what I mean by extrapolating (the BPL) here is, can we find the NN architecture inspired by how BPL works? If we do, I believe that would be the major breakthrough.
This comment has been removed by the author.
DeleteIt is great to find an expert on the subject :) Thank you very much for pointing out a couple of new things to me. Can you send same links on BPL? Also I am looking for an article or example of using NN as a prediction tool ( not just a classifier).
DeleteWhich of these achievements do you think is:
ReplyDelete1 - the best - Creative Abstract Thought
2 - the most promising - Dreaming Up Visions
3 - the most interesting - Abstracting Across Environments
- “ (…) even more exciting are the developments we don’t expect.” – what these could be in your opinion ?
Yes I totally agree with Maricn that the most challenging taks will be building somethin for everyone of us.
And this will be great when you can mix it with open source and you can build it by your own. This kind of experiment was conducted by Massimo Banzi (one of co-founders of Arduino).
On 3d printer they print every furniture and they use Arduino for Home Automation.
Combining ML/AI methods with 3D research and printing appears to have a very profound implications, not just in furniture printing. Recently scientists from University of Michigan developed an ML/GPU based software tool that allows to a design and 3D print new materials on the molecular level. They used computer simulation to discover the fundamental principles of how nanoscale systems of building block self asseble, then they discovered how to control the assembly process to engineer new materials. If you are interested please follow link below:
Deletehttp://devblogs.nvidia.com/parallelforall/developing-new-materials-gpu-accelerated-supercomputers/?mkt_tok=3RkMMJWWfF9wsRojuajOZKXonjHpfsX%2F4%2B0pXaG%2Biokz2EFye%2BLIHETpodcMS8NmN6%2BTFAwTG5toziV8R7HAKs1v3NsQXBXgHi%2F6k7Xsbp9HcLpn9cU8rfn9
Very interesting article. The topics is very timely. It is difficult to assess what will be best. The best in what respects, the most useful, the most progressive, the most reliable? Similarly I can not say what will be the most promising area of research because it is very wide and the target groups are very diverse. It is similar to what is most interesting it all depends on a narrow target group specific actions.
ReplyDeleteMy opinions are similar to speakers.
I suppose that the best and most interesting and the most promising development mentioned in the article was Planarian Regeneration Model Discovered by Artificial Intelligence.
In general I agree with you. Since the subject is very broad and the area of knowledge is dynamically grawing in every direction possible, the only reasonable approach is to subscribe and follow a few selected scientific newsletters or e-magazines like New Scientist or Kurzweil AI news. I recomend specially the last one because it covers very wide spectrum of scientific research.
DeleteThank you, it for sure...
DeleteVery interesting article. The topics is very timely. It is difficult to assess what will be best. The best in what respects, the most useful, the most progressive, the most reliable? Similarly I can not say what will be the most promising area of research because it is very wide and the target groups are very diverse. It is similar to what is most interesting it all depends on a narrow target group specific actions.
ReplyDeleteMy opinions are similar to speakers.
I suppose that the best and most interesting and the most promising development mentioned in the article was Planarian Regeneration Model Discovered by Artificial Intelligence.
1 - the best ? Abstracting Across Environments and DeepMind’s seminal DQN, because it can play games. The idea was simple as well, we are teaching our AI with basics, like in an example by playing pong our AI can easier learn how to play tennis.
ReplyDelete2 - the most promising ? Intuitive Concept Understanding actually shows us how complex our brains are. We can easily distinguish human emotions when we are looking at someones face. We can learn that with simple examples and so on, for AI it's far more complex, but with proper algorithms we can make software to prevent certain attacks or something like that based on human emotions.
3 - the most interesting ? Agile and Dexterous Fine Motor Skills, especially for people with disabilities.
In principle I agree with you and everything I had to say I wrote in response to Wieslaw answer above :).
DeleteHello, thank you for your article recommendation. My picks are as follows:
ReplyDeleteThe best of them is hard to define, I guess I'll have to pick just overall best for my likings which is the first one which uses DeepMind DQN because I think is the one that brings us closer to the singularity among those that were presented. I find this as the most promising too for the same reason.
As the currently most interesting though I find the one that deals with imagining pictures based on patterns, something that untill now was done but in the reverse.
Very interesting line of thoughs :) I wonder what is your opinion about Singularity? One day we should discuss it.
Delete
ReplyDeleteWhich of these achievements do you think is:
1 - the best ?
2 - the most promising ?
3 - the most interesting ?
It is really hard to tell which of the achievements is the best, the most promising or the most interesting. I'd say that all these achievements were. The time will indicate the best ones.
For me the most interesting are GPU algorithms as well as most promising.
and the question asked by the author:
- “ (…) even more exciting are the developments we don’t expect.” – what these could be in your opinion ?
We do not know. I think that ingenuity of people is unlimited so I hope that the developments will make us excited.
GPU based algorytms are definitely very interesting, specially the work presented by NVIDIA corporation. You can read more of their exellent research in NVIDIA weekly newsletter
DeleteFor me answer for all 3 questions is: Abstracting across environments because it gives a possibility of transferring skills and knowledge learned in one field into another just like real intelligence does
ReplyDeleteI agree that Abstracting across environments is the most fundamental feature of AI and the "Holy Grail" of AI research. As soon as this problem is solved, the practical emergence of true AI will be possible.
DeleteI'm not really into AI, but Nowadays the technology which surrounds us depends on AI. As the answer for the first question I wouald say that I have seen many examples of AI achievements and it is really hard to choose which is the best, but I can tell which i like the most - the japanese and korean robots.
ReplyDeleteFor the second question, Like I said on the begining the technology depends on AI, but maybe i can say: The current technology, on any field, is the representation of our work on AI - so the question should be more specific.
What is the most interesting? - personally I think the most interesting is the image recognition.
In this case I strongly suggest to follow several links that I have published in this conversation :)
DeleteAll of the achievements mentioned in this article are impressive. In my opinion Creative Abstract Thought is the best, the most promising, the most interesting achievement. Using 20 kinds of reasoning, like deduction, induction, temporal reasoning, and path finding by a system gives an enormous field for future developments and improvements.
ReplyDeleteIt looks like Planarian Regeneration Model Discovered by Artificial Intelligence is just a start.
PR model is definitely one of the most promising approaches, but it represents only a small part of research in AI development. To read about other researches and achievements in this area I suggest to follow links mentioned above.
DeleteAll of the achievements mentioned in this article are impressive. In my opinion Creative Abstract Thought is the best, the most promising, the most interesting achievement. Using 20 kinds of reasoning, like deduction, induction, temporal reasoning, and path finding by a system gives an enormous field for future developments and improvements.
ReplyDeleteIt looks like Planarian Regeneration Model Discovered by Artificial Intelligence is just a start.
ReplyDelete- “ (…) even more exciting are the developments we don’t expect.”
AI is a hard topic. Most of the problems related with it are only theoretical. From my point of view we`re far from coming close to something related with AI. Why ? Because we don't fully understand our brain and how it work. We know how it wors in the very basic level but sstill we don't understand it fully. Until then we never create a true AI
I disagree completely :(. The best example of the modern and very sophisticated understanding how our brain works was published last year by Ray Kurzweil in his fundamental book "How to Create a Mind: The Secret of human Thought Revealed". This work gave a new and revolutionary view on the fundation of construction of human brain simulation. I strongly recommend this book to anyone interested in the subject. :)
DeleteHi Kinga! Thank you for an interesting article. Personally I am not a big fan of "top something" articles, but this one is very inspiring. It is difficult to say which of these achievements are the most important or promising. The A.I. field is really broad and everyone can choose something slightly different, depends on their research fields or their interests. I am not a specialist in this area, but I agree with previous comments that Creative Abstract Thought is the most impressive.
ReplyDeleteEmilia thank you for your comments :). If you are interested in learning more about AI and you will find time to read all the literature already mentioned by me in this conversation then you a very welcome to follow all the links I have provided above. No doubts, in the process you will find more links and materials relating to the subject.
DeleteAll of those achievements have little in common with AI. This is all about learning and inferences which are only a small part of our intelligence and identity. I can't understand why nobody is interested in how does our imagination work or what are the causes of consciousness? What are emotions, feelings, and thoughts? How do we manage to think at all? I think those "features" are more important for human intelligence and until we haven't found answers for those questions I would stay with term Machine Learning and stop using "Intelligence" as a descriptor because it sounds pathetic for now. From the list of articles I would vote for all non-learning ones. We know something about it while other areas seems to be untouched, so those topics are promising for me.
ReplyDeleteThis is a common misconception that ML is not a part of AI. Quite the opposite! The ML is an integral part of AI development and comprise more than 50% of the current research effort of all involved scientists.
DeleteI would follow Marvin Minsky or Ian H. Witten - the creator of WEKA (I think most popular ML and ANN framework). Both says that it shouldn't be classified as an "artifitial intelligence". Please read their arguments, its hard to not agree..
DeleteAll of those achievements have little in common with AI. This is all about learning and inferences which are only a small part of our intelligence and identity. I can't understand why nobody is interested in how does our imagination work or what are the causes of consciousness? What are emotions, feelings, and thoughts? How do we manage to think at all? I think those "features" are more important for human intelligence and until we haven't found answers for those questions I would stay with term Machine Learning and stop using "Intelligence" as a descriptor because it sounds pathetic for now. From the list of articles I would vote for all non-learning ones. We know something about it while other areas seems to be untouched, so those topics are promising for me.
ReplyDeleteA really nice pick as for article. Frankly speaking I’m not so sure what could be the best or the most.. perhaps due to issues with picking up the top something. But I would have to say AGI by DeepMind made the milestone. It opens up the entirely new opportunities for many day to day solutions.
ReplyDeleteok :) please subscribe to the Kurzweil AI newsletter in order to find more information of the subject on AI. I guarantee you will enjoy it. Write an email to
DeleteI don't know which achievements to choose for each category. They all look promising and interesting but I have no idea according to which conditions I should pick the best one. Choosing most promising is also difficult, in that point I have to agree with Wiesław, all achievements listed differ among the groups of interests, I don't feel being a member of any of them. " (…) even more exciting are the developments we don’t expect." - sounds like really deep thought. I understand it as we have no idea what challenges and discovers will come with time.
ReplyDeleteMaybe reading more about a subject of AI could help :) One good reading is "Singularity is near" by Ray Kurzweil. It is a fascinating book, you will like it.
DeleteGreat article, thanks! I'm not sure which one is the best, most promising, and most interesting but I found the concept of "Abstracting Across Environments" exciting and impressive. It definately makes sense to apply the knowledge from one domain into another. There was for sure a lot of progress in the AI during 2015, probably thanks to strong cooperation between academia and industry (Microsoft and MIT, Facebook and NYU).
ReplyDeleteThank you for your opinion. You are definitely right about the fact that 2015 was the year of the biggest progress in the AI development so far. Everybody think that 2016 will be even greater year of research and some profound breakthrughs in the AI development will happen.
Delete1 - the best ?
ReplyDelete2 - the most promising ?
3 - the most interesting ?
> God.. I don't know what was the best or most promising. None ? Like one said, it wasn't a revolutionary year but we saw the evolutions.
From the article, I found "dreaming up visions" the most exciting one. Endless opportunities for the cinema.