2

I'm a computer science and software engineer that worked in industry for some years with deep applied knowledge in ML and AI, i commonly read research papers and this year started a graduate scientific research program , in this first weeks we are studying basic concepts like what is research, and common descriptions like :"research seeks to generate new knowledge" , "you observe a problem or a phenomenon , create hypothesis about that which then you seek to prove experimentally and collecting data" , my question is in computer science and AI, what are the problem and phenomenon? and what are the hypothesis in AI research? Commonly in papers researchers propose new algorithms or new models but what were the common "observed problem" and the hypothesis? it's not clear to me how CS and AI research fits in this classic research definitions.

If possible can anyone provide examples on specific AI research papers ,how they fit in the classic definitions, what are the problems under study,hypothesis and all scientific method parts in the paper?

  • Hypothesis: "A cellular automaton can match more than one pattern simultaneously, and so can test for multiple patterns in time O(n) where n is the number of cases to be checked." – Bob Brown Mar 13 at 1:46
  • 1
    I'm really struggling to understand the question, can you edit? (I've nonetheless done my best to answer.) – user2768 Mar 13 at 8:25
  • Sorry @user2768 english is not my main language so maybe i did not a good job explaining, basically what i'm trying to say is that i'm taking a scientific research graduate program but is a general one, not a CS or AI specific research program, but i'm trying to find how classic and general definitions of research teached in the program apply to CS and AI research,and how the steps of the scientific method fit in CS and AI research. – Luis Leal Mar 13 at 18:48
  • I have by now many questions but let me do an specific example: in one of the first classes they mention a classic science research example: 1) You observe a problem (i said phenomenon but not sure if it's the correct translation , but let says something of interest which is maybe unknown) and you formulate a question about it. 2) preliminary research with background and existing related work. 3) formulate hypothesis 4) do necessary experimental work and collect data 5) analyze data 6) conclusions (Question in the following comment) – Luis Leal Mar 13 at 18:57
  • But for example in AI research, how that description fits when researchers publish a new paper proposing a new algorithm or a new model? how the research started? how they performed observations to a problem or situation if what they are proposing is something that did not existed before? how they came with the idea if the model or algorithm was nonexistent? How they formulate their hypothesis if they were not able to do observations? – Luis Leal Mar 13 at 18:59
3

AI is a very applied domain, so from a big picture point of view the problem is often defined by a real world application. For example, machine translation fills a need for low cost translation (so this particular need is the problem). In this general setting, an hypothesis is for instance: "the machine can produce good quality translation", and researchers try to prove it.

However it's more common for a paper to motivate the work at a more fine-grained level using the existing body of literature. For example many approaches have been studied for machine translation, with varying levels of performance and flaws. A researcher in this field might want to study a specific question, e.g. "how to improve translation of multiword expressions from language A to language B" and formulate an hypothesis such as: "recent approach X can improve translation of multiword expressions from language A to language B". Here the problem is formulated as the need to improve over the current state of the art, implicitly assuming that the underlying problem (machine translation in general) is sufficiently established.

  • completely makes sense. Thanks – Luis Leal Mar 13 at 21:18
2

I'm guessing that the descriptions you have seen are from fields that work very differently from math and cs, including AI.

In the social sciences and most of the physical sciences, an hypothesis is a statement that might be true or false. It is stated in a certain way for technical reasons, but it is normally stated at the beginning of a study before there is any evidence whether the hypothesis is true or false. Depending on the field, some methodology is used to gather evidence about whether the hypothesis is true (accept) or false (reject).

The reason for proceeding this way is that the gathering of evidence, say with experiments or questionnaires, is an expensive and time consuming endeavor that is actually guided by the statement of the hypothesis. For example, if the research is guided by questionnaires to people, you don't ask the people a lot of random questions and try then to figure out what it means. You ask them questions related to the hypothesis so that certain answers (determined in advance) support the hypothesis and the opposite answers work to refute it.

But math and much of cs works differently. We have an idea for a theorem, or a way to improve garbage collection (GC) in a programming language. We work to prove that theorem or build a program to test out the idea. If we prove the theorem we have the basis for a (part of a) paper. In the GC case, we run the program to see if it is an improvement over other known approaches or not. This gives the basis of a paper, perhaps.

But, we don't usually call the (possible) theorem or the idea for GC the hypothesis, though in some sense it is. But in both math and CS you often just try something to see how it works out. Note that the trying out is actually the gathering of information and it may come before the "hypothesis" (theorem or gc idea) is ever stated. The time scale is often reversed.

Another way to say it is that we often start with an informal idea (rather than a formal statement), work on the idea and if we learn something, only then make a formal statement. The social and other similar sciences normally don't work that way, or at least don't present their work as if they did.

In fact, though, a social scientist will him/herself need to have that bright idea first about what might be worth studying, just as a mathematician does. But they will still state a formal hypothesis to guide their actual experimentation.

I'm guessing that you are puzzled by this since the formal hypothesis idea seems a bit foreign to what you see in papers. You are likely to see first the "idea" that drove the research, then the proof of concept experiment (a program, perhaps), and finally the conclusions. Not hypothesis is stated, and it isn't expected, though it could be formulated that way. But since the work was done before the hypothesis was formulated, there is no real notion of accepting or rejecting something stated in advance.

The difference is actually driven by the different kinds of evidence required in order to find truth that advances a particular field.

0

what are the problem

Problems are numerous, e.g., unsolved problems in computer science. (I'm unsure what is meant by phenomenon.)

what are the hypothesis

A hypothesis is a starting point. E.g., P is not equal to NP.

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.