1
$\begingroup$

What are the steps involved to create an AI agent which can do the following

  1. can learn from text in digital format from a book
  2. Gain knowledge from the digital text input
  3. Can answer question from the book fed to it
$\endgroup$
  • $\begingroup$ A system which comes close to the requirements is the Platypus chatbot. It has a demo website and was described in the paper “Tanon Pellissier: Demoing Platypus–A Multilingual Question Answering Platform for Wikidata, European Semantic Web Conference, Springer, Cham, 2018.” Natural language is converted into the RDF syntax and the system is able to answer simple questions related to that knowledge. $\endgroup$ – Manuel Rodriguez Jan 3 at 15:45
  • $\begingroup$ Pretty sure the Watson had this basic capability, scraping the internet to guess the correct answer. The main problem is how abstract is the question. If the question is "What color is Clifford?" (Red) it's pretty easy, but if the question "What is the meaning of the the story?" it's a problem of a whole other order of difficulty. $\endgroup$ – DukeZhou Feb 6 at 2:57
2
$\begingroup$

Although I gave an up vote because the question, in principle, is a good one, there is considerable technology ground to cover before such a project would be feasible. A century or two might not be enough time to cover the ground, although no one knows. All past guesses have covered the range from a decade (which already passed, so that was grossly overoptimistic) to never (which dismisses possibility without a scientific basis for dismissal).

Where does one begin? Here's what is missing.

  • The term learning when discussing a student covers a wide array of mental abilities whereas in artificial learning the array of abilities is orders of magnitude more limited.
  • Providing answers requires comprehension and extrapolation from it, which machines are imagined by Minsky and others to be potentially capable of, but no demonstrations of comprehension have yet been made public.

In the 1970s and 1980s it was believed that rules based systems could cross these landmarks, but what might be termed artificial students are yet to emerge from that work. In this century, deep learning is expected by some to produce this kind of cognitive skill, but there is no real evidence that the current type of converging systems and continuous learning control systems will reach the level of cognitive skill to produce artificial students either.

There are six possibilities.

  • Production (rule based) systems will come back and achieve cognitive skills that include those needed for text book knowledge acquisition.
  • Artificial nets will achieve it.
  • Some new thing will.
  • None will.
  • All three will.
  • Some will.

As of this writing, no definitive answer can be given. If a researcher is young enough and is willing to work on understanding exactly what mental abilities are required to acquire knowledge from books and assimilated it sufficiently so that the knowledge can be leveraged to answer end-of-chapter questions and pass final exams, it is possible that the researcher may live to develop such an AI system. It is also possible that someone will invent something that can do it in 2019 or that our great grandchildren will not see such results.

$\endgroup$
1
$\begingroup$

Obviously you will have to cover a lot of technological ground before you start thinking about actually starting work. I may be wrong but here is what i think could be one of the solutions what areas you can focus on to begin work. At least theoretically.

Step 1: Teach the bot to interpret text,it should be able to distinguish between words, sentences.

Step 2: Teach the bot grammar. It should know tenses and understand basic concepts of grammar of that language

Step 3: This is the most tough part. Give it a understanding of context. When we read 'I ate an apple' we use previously acquired data of touch taste and sight to know what ate and apple mean so that gives us context.

To understand a book enough to answer questions even humans use additional information from our day to day lives. A bot who does not know what eating actually means would not understand what eating an apple will mean. this bot will be able to answer general direct questions most probably, by making basic assumptions but it wont be able to answer twisted or indirect questions.

Step 4: Store understood data , usually could be in a tree form

Step 5: Understand the question articulate it in tree searching format and find the answer.

I'm open to suggestions and criticism :D

$\endgroup$
  • $\begingroup$ For steps 1&2, take a look at the Natural Language Toolkit nltk.org $\endgroup$ – DrMcCleod Feb 3 at 12:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.