2
$\begingroup$

I hope this question is ok here, but since I have found a tag which deals with these issues (profession), I'll ask away. I also hope this may be useful to other people with similar doubts, since I am failing to find valuable information on this topic online.


I am interested in the theoretical side of CS, such as computability, logic, complexity theory and formal methods. At the same time, I am deeply fascinated by Artificial Intelligence and the questions it poses to our understanding of the notion of intelligence and what does it mean to be a human being.

In general, is AI a more "applied"/engineeristic field, or are there theoretical aspects to research in?

In short: If I prefer formal/theoretical compsci, is AI a bad career choice?

(note: I am asking this because I am a CS undergrad considering getting into a AI MSc).

$\endgroup$
  • $\begingroup$ One big problem is AI is buzzword-of-the-year, and everything that can be (and much that should not) is being labelled "AI". As a result, some AI courses may be theory-heavy, some will not be, and some might be data science with a new trendy name. Do you have a synopsis of your proposed MSc course? $\endgroup$ – Neil Slater May 7 at 13:44
  • 1
    $\begingroup$ Yes. uva.nl/en/programmes/masters/artificial-intelligence/… $\endgroup$ – olinarr May 7 at 13:45
  • $\begingroup$ There are other MScs which are even more "theoretical" (such as Utrecht's course) but they seem too keen on psychology, cognitive science and philosophy for my interests. I have chosen UvA because it looks like the most compsci-heavy course out there. But I am afraid this comes at the cost of a too much "applied" inclination $\endgroup$ – olinarr May 7 at 13:47
  • $\begingroup$ For context, do you have any idea of any course - on any subject - that does offer your preferred mix of theory vs applied work? I know that might be hard to assess for you, but it is also hard for someone answering to try and predict what counts as enough theory or too much applied work for you (and not themselves). $\endgroup$ – Neil Slater May 7 at 13:55
  • 1
    $\begingroup$ Maybe it is a tag that can be used for clarifications regarding the AI field and AI jobs, without mentioning a career path or asking for a recommendation. There are some tags on this website that shouldn't exist or are a little ambiguous. @DukeZhou, you apparently created this tag. Can you clarify this? $\endgroup$ – nbro May 7 at 15:26
3
$\begingroup$

There are certainly results in theoretical computer science / pure math with deep implications for AI. But to my knowledge these results typically aren't labeled as results of artificial intelligence, but as something more congruent in that particular field (For example in CS, we might say "agent with unbounded computational power"; in math might say some statement is "decidable/undecidable" with respect to some system). Of course they still matter in the field of AI, but you need to know what you are looking for.

See my question What are some implications of Gödel's theorems on AI research? for some examples. Or you can look up MIRI's research guide for a better idea of what existing work is out there that links formal math / CS to AI research.

Another point to raise is that there is no good definition of AI in fields outside of normal discourse (or even within, perhaps), so its difficult to decide what discussions pertains to the study of AI. Questions like whether ZFC with/without choice is expressive enough might not be on the mind of most AI researchers, but could still have some implications.

So to answer you question more directly, there is certainly a field of study regarding theoretical AI. Regarding whether or not its a good choice is something for you to decide, but it is (in my humble and not-very-well-educated opinion) very difficult field that isn't very popular, and has not seen major progress in many years.

$\endgroup$
  • $\begingroup$ AI has not seen major progress and isn't very popular? $\endgroup$ – olinarr May 8 at 4:05
  • 2
    $\begingroup$ Sorry I should be more clear: The specific brand of "pure AI" construed as unbounded agents / decidability results / incompleteness theorems is largely unpopular relative to the engineering aspects of AI (neural nets, probabilistic inference) $\endgroup$ – k.c. sayz 'k.c sayz' May 8 at 4:11
  • 1
    $\begingroup$ But in another sense, yes: AI (more specifically AGI) has not seen major conceptual breakthroughs since the days of Simon and Newell. (imo) Recent progress in NNs and probabilistic methods are very powerful but also very task specific. But very popular indeed. $\endgroup$ – k.c. sayz 'k.c sayz' May 8 at 4:14
2
$\begingroup$

Any "serious" AI program is full of theoretical and mathematical foundations (you will study plenty of statistics and optimisation methods) anyway, but I would say that much useful AI today is an applied or engineering area. Anyhow, you will need to be comfortable with a lot of mathematical details (especially, linear algebra and calculus). If you're more interested in statistics, optimisation or robotics, you should go for AI.

If you study "pure" computer science, you should also have one or two courses related to AI (at least, one ML course). If you are more interested in traditional CS algorithms, data structures, software engineering, operating systems, compilers, theory of computation, computer networking, programming languages and/or databases, then you should go for CS.

However, before enrolling in a master's program, you should really have a look at the details of the courses they offer. Furthermore, you might also take into account that, during your studies, you might change idea regarding one subject.

$\endgroup$
  • $\begingroup$ This answer contains some guidelines to choose among AI or CS solely based on the reader's interests. It completely ignores future job opportunities. $\endgroup$ – nbro May 7 at 14:22

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.