Hello,
this week I would like to talk to you about automatic text generation by neural networks. Natural Language Processing is a rapidly growing field of science. It touches on problems like speech recognition, langage understanding, machine translation, but the one I personally find the most interesting is natural language generation. Recent years the neural networks techniques seems to be the most promising in this area. In 2019 OpenAI group released very impressive language model called GPT-2 based on Transformer architecture. An example of what it can create might be this text about unicorns.
Articles that tell more about the model:
https://openai.com/blog/better-language-models/
https://arxiv.org/abs/1706.03762
If you would like to try it by yourself, check TalkToTransformer or TabNine tools.
What do you think about generating texts in natural language by AI?
Is it a chanse or threat?
What practical applications of this technology do you see?
1. What do you think about generating texts in natural language by AI?
ReplyDeleteI think this technology is amazing. AI is a powerful tool. Generating a good quality tekst by AI could be dangerous because we can easily produce fake news. For some people this model could be helpful in generating text about some topic as a source to improve own text. We need less time to write about something. It’s a good idea to not public the big model because it could be used for bad purpose.
2. Is it a chanse or threat?
I think it depends how we will use AI and for what purpose. In wrong hands it could be used to generating fake news. In a right hands we could used AI for automatic support. AI could answer our qestions about problems or we can chat with AI.
3. What practical applications of this technology do you see?
I think this technology we can used to create automatic technical support assistant or a chatbot to talk with him about different topics.
I agree with you. There are always both sides of the problem. Automatic text generation could be a tool of disinformation, but also a great time-saving solution.
DeleteWhat do you think about generating texts in natural language by AI?
ReplyDeleteThe use of AI for testing has been used for a long time. The first serious attempts to use Artificial Intelligence methods for natural language analysis were made in the late 1960s. This was the peak of the Cold War and the US Secret Service wanted to use AI to quickly and efficiently translate the Russian language texts acquired through operational intelligence activities into English. Today, the AI is so advanced that it is able to write the whole book. In my opinion, this is a problem because during election campaigns, international conflicts or induced social unrest very often methods of manipulation or simply lies, euphemistically called "fake news" are used. When these phenomena cause real problems, we start talking about "cyber-threats" or "information war". It is much worse when behind such actions there are specialized AI algorithms, which are able to generate thousands of "news" every day based on a mixture of real and false information. Then even the best team of analysts is not able to analyse them. People, even those with higher education, are very poor at recognizing false messages. Making the society sensitive to media manipulations and lies is a very important educational task. Researchers at Stanford University have conducted research with about 8,000 students to determine their ability to assess the reliability of online information. The results proved to be devastating. The vast majority of students do not have the basic skills to recognize reliable information and cannot distinguish sponsored content from real articles.
Is it a chanse or threat?
Just like I mentioned above. There's a danger. Because the basis of a single sentence can generate texts on an extremely wide range of topics: from gossip from celebrity life, to current political events and scientific discoveries, to prose, and they are linguistically correct and extremely stylistically coherent. The greatest threat is the excellent Fake News. GPT-2 does not make mistakes in sentence syntax and does not create nonsensical information, unless this is the convention of the source text. So we make mistakes to fool around with the assumption. And that's knowledge only from the interlude pages because. The algorithm of the program was created on the basis of the analysis of 8 million websites and 10 million articles from various fields. If the base is expanded, the quality of the tests will certainly improve as well.
3. What practical applications of this technology do you see?
The applications are unlimited. Perhaps the most useful thing is to translate difficult issues through the neural network into more simple ones.
Thank you for an interesting historical view. I agree with you that false information (or fake news) is a significant problem nowadays. The research you mentioned is a bit disturbing, as we can assume that students are a group of people who should be fluent in searching and interpreting information.
DeleteIn such conditions language model as gpt-2 might be a dangerous tool. Although I hope it rather will be used for good purposes.
1. What do you think about generating texts in natural language by AI?
ReplyDeleteThis is a very interesting technology. Looking at the current development of AI and the great interest in what intelligence really is, I think that the development of such technology is very necessary. Maybe one day it will be used in robotics to build robots that will imitate us even better.
2. Is it a chanse or threat?
As I said earlier, I see great potential in developing this technology. Perhaps it will allow us to better imitate man. But as others have said, it can also be a field for abuse and will allow the creation of fake news, which will be even more difficult to recognize.
3. What practical applications of this technology do you see?
In combination with speech synthesis technology, I see great possibilities for use in robotics and in various types of voice assistants.
Yes, robotics is one of the fields where NLP solutions could be broadly used. Thank you for mentioning it.
Delete1. What do you think about generating texts in natural language by AI?
ReplyDeleteI think this is a great technology for online assistants (Alexa, etc.), you can generate speech and play it for users. It is also possible to create a thesis, if of course you can set the parameters for generating text. And to create my doctoral work, I think it would also be useful.
2. Is it a chanse or threat?
I think everything can be classified as a threat or utility. It all depends on who will use it and in what hands this technology will be. I think the world is moving forward so fast that it will be impossible to control it and it can be used to the detriment of humanity
3. What practical applications of this technology do you see?
If, in addition to generating the text itself, it will be possible to reproduce it, then the creation of robots is instead of some professions
Thank you for sharing your opinion. As you noticed, NLG has a wide spectrum of applications, depending on the data and parameters provided to the solution.
DeleteHello,
ReplyDelete"What do you think about generating texts in natural language by AI? Is it a chanse or threat? What practical applications of this technology do you see?"
In my opinion, this field of AI is an excellent opportunity to introduce better word suggestion in communication software, which could be used by the people who suffer from neurodegenerative diseases and lock-in syndrome. Even trivial solutions, like BCI interfaces based on the p-300 paradigm, can shift data transfer from a few bytes per minute to higher scores. Basically, we can change them from simple "letter-by-letter" picking for more complex interfaces, if only language models would be good enough to fulfill this need.
Hello Artur,
Deletegreat idea on the application of NLG. Recently I read 'The Diving Bell and the Butterfly' book, written by editor of Elle magazine, who was in lock-in syndrome, and I though about the same, if NLP could help to communicate people with such disease.
1 What do you think about generating texts in natural language by AI?
ReplyDeleteI'm not sure. I know it's all probability, but level of coherence in new models is surprising. Maybe all this probability computations converge somewhere in true understanding of human languages by machines?
2 Is it a chance or a threat?
The dangers are consequence of how important we let social media become. If we kept it a toy, there wouldn't be a threat I think. Now - we can be influenced by texts generated by deep language models in any direction.
3 What practical applications of this technology do you see?
One practical applications I know of is "AI Dungeon 2" game, where text content is generated using deep language model. It's really cool application in my opinion. Allows player to explore infinitely large and never repeating world.
I will answer all your questions at once. This is really a huge technological achievement. The fact that a machine can create even a whole book by itself is impressive. However, as others have already mentioned, the fact that AI is able to create books, texts and articles indistinguishable from what man created is quite dangerous. In the age of the Internet, where everyone has wide access to information, there is a high risk of spreading false news, scientific articles etc. in this way. Pervasive misinformation can occur, and real scientific research can be challenged by artificially generated material. Real research will become unreliable and chaos may occur in societies. AI can create texts that slander politicians or celebrities and harm their career. Any tool is useful, but you must use it carefully. I see many disadvantages of such technologies, but the positive aspect is e.g. generating descriptions of some products in marketing. It is often difficult to write on the websites / online stores descriptions of subsequent products, AI can do it for us. It can also work in writing novels and general fiction. You can never have too many good, addictive books. It can also help in writing film scripts. I think it would also work when writing instructions for all kinds of devices. Especially that most of the instructions are similar and certainly a lot of time goes down when creating new ones.
ReplyDelete1. What do you think about generating texts in natural language by AI?
ReplyDeleteArtificial intelligence is a very dynamic part of computer science that is growing rapidly. What he can do more than once it was possible to find out. He is now writing better and better. However, I would like it not to be too abused by the young generation, because in my opinion it will stop their development of essays, articles and other texts.
2. Is it a chanse or threat?
In my opinion, artificial intelligence is a huge opportunity, because it gives unlimited possibilities and maybe one day it will be possible to build an artificial brain on its basis. Of course, it all depends on what it will be used for.
3. What practical applications of this technology do you see?
I think it could be used, for example, to improve development environments in which people could create applications under the control of people.
1. What do you think about generating texts in natural language by AI?
ReplyDeleteAlmost every business today is looking to embrace AI and reap the advantages of its subsets with an intelligence-driven system that captures, processes and synthesizes data resulting in automated data analysis as well as content management. Tools like Natural Language Processing, Natural Language Understanding and Natural Language Generation and their analytical algorithms are nowadays use for address specific business requirements and goals like improving product recommendations, online search, voice-enabled mobile services, etc. With NLG in place, it is possible for struggling businesses to think beyond conversational chatbots and integrate an automatic, goal-oriented system of efficiently producing information in a format as expected by the end user.
2. Is it a chance or threat?
In my opinion is a chance, at the moment NLP is battling to detect nuances in language meaning, whether due to lack of context, spelling errors or dialectal differences. On March 2016 Microsoft launched Tay, an AI chatbot released on Twitter as an NLP experiment. The idea was that as more users conversed with Tay, the smarter it would get. The result was that, after 16 hours Tay had to be removed due to its racist and abusive comments. Microsoft learned from its own experience and some months later released Zo, its second-generation English-language chatbot that didn’t make the same mistakes as its predecessor.
3. What practical applications of this technology do you see?
What NLG is mainly capable of is its ability to create on organized structure of data from the information processed in previous stages of NLP and NLU. By placing this well-structured data in a carefully configured template, NLG can automate the output and supply documentable form of data such as analytics reports, product description, data-centric blog posts, etc.
1. What do you think about generating texts in natural language by AI?
ReplyDeleteI think that shows how fast the world is developing. Artificial Intelligence is very popular issue nowadays. Almost every device like TV, Phone etc has some AI functions. In my opinion it is one of possible applications for AI. If AI could generate text without mistakes I would not have to write answers for your question now! Generating text in natural language by AI can be applied in writing books. I cannot see anything wrong in condition that the generating text will be applied in proper way and will not misuse for example to generate the fake news.
2. Is it a chanse or threat?
As I wrote above, I think that generating text in natural language processing by AI is very good idea and one of applications of AI. I think that the scientists developing AI should take into account the worst scenarios. We cannot get into situation when we will not know what is truth and false. I think that there is golden rule that you can do everything but you have to know when you have to finish. In my opinion that there is the same in this situation.
3. What practical applications of this technology do you see?
I think the there is the wide filed where you can apply the generating text in natural language processing. One of the appliance is writing a book. It could be useful for parents who have to put the child asleep. I suppose that there are a lot of case when the child says “ Mam, Dad, tell me a tale, please” and the parents have no idea. They could use this generator. This technology also can help and improve process of creating movies.
1. What do you think about generating texts in natural language by AI?
ReplyDeleteThat's amazing! To be honest...I didn't expect nothing special from this language model but when I read what he generate that change my mind! This text are very good and I would have problem with recognize if the text was generated or write by human.
2. Is it a chanse or threat?
I think that chanse...I can't see any threat in this case. In the worst scenario the model will produce fake news :D
3. What practical applications of this technology do you see?
Chatbot is the first thing which I will produce using this model. The text is so good that if feed this model using text which we have from comunication between support assistant and client then this model could produce answers on his own.