Are lay articles good enough to be the main source of information for PhD research?

Question

Doing PhD in new field that does not have enough resources in it. Most of the information I find are articles written in websites such as medium.com. I really feel awkward citing a website article, as I feel it is not good enough. It is not from university, it has just the author name, not much info about him/her, and the way it is written is for non-academic purpose.

Is it ok to have many of these articles as the source of my information in my PhD? If it is not ok, then what can I do? there is no reliable resources! The field is Artificial Intelligence.

To be more specific: It is not about AI in general, it is about a topic in AI that has not been explored thoroughly, the only place that has done that is Google and they keep the research they have done private and not accessible for public, i have contacted them but they rejected to help with their research

Edit : I know that there are millions of AI researches in the internet. but there isn't anything I was able to find on the area i am studying. google has a product and they are selling as a service and they are not publishing the research they have done on that area, i have contacted google and they did not allow me to look at their research.. there are other researches on other areas may touch that field but none i found on that field specifically.

Answer 1

It sounds to me that there's a confusion between two things:

• Motivations of the research: imho it's completely fine to use non-academic sources to justify why the research is being done, especially in the case of a new application/domain. This part can even require a thorough analysis of what it would be used for, by whom and what are existing solutions (for instance commercial solutions available). Such work might even qualify as a (probably small) contribution by itself, in the form of some preliminary work to introduce a new application for example.
• The actual research contribution, i.e. what is being done and how: it's quite unlikely that a non-academic source would provide all the necessary background (especially theoretical) on which the application is based. Even if it did, its scientific validity could be questioned. So for this part it's difficult to imagine a case where one doesn't rely mostly on regular academic sources, typically in the form of a literature review of existing related works. Note that related works don't have to address the exact same problem: as long as they share some similarities (e.g. similar method for a different task), they are relevant. In this part it's important to provide details about the similarities and differences between the existing work and the contribution.

Answer 2

I find the entire premise of the question quite odd. First of all, to answer your question: I’d say that in most cases I’ve encountered the answer would be no, you can’t use popular science articles as a primary source.

That said, I seriously doubt that they’re all that’s out there. Google works on this problem: they came up with it and no one else ever heard of it or studied it? How did the popular media hear about this amazing idea that has somehow eluded the entire AI research community except Google and your advisor? I feel like you’re either looking at the wrong sources, your advisor is not pointing you in the right direction, or something else is off.

Answer 3

From the way you describe, this sounds really strange especially since there is a lot of real research in AI. I cannot believe you could only use those articles. However, your supervisor is the right person to ask - most likely, they alone will decide whether or not your dissertation is enough.

Answer 4

Answering for anyone who comes across this question from social science etc. as it is a bit different. We can use popular media sources, but only for certain things.

As someone who uses popular press articles as a "source" for my research, and who advises students who do the same, there is an important distinction to be made and a significant amount of methodological explanation required before you can use much popular media.

The distinction: you can use popular media as a source of information: of facts, or, perhaps, as a source of contemporary public commentary around a subject. You cannot use popular media as a source of analysis. So, for example, you could use popular media articles from the time period, in conjunction with legislative records, to understand the details of and popular sentiment about a change in Australian tax law in 1980. You can't (only) use a post on medium as a sole source of analysis as to why that tax law change was academically significant.

The explanation: you have to indicate that you are fully aware of the drawbacks of using popular media as a source of information for analysis, that you have considered the issues with doing so. You then have to show why it is methodologically necessary to use the media source. I use a lot of local Latin American crime reporting in my research, which I cross reference with government records that I have access to. The local reporting adds context, flavour, and a bit more detail to the government records, sometimes revealing types of information that don't exist in official reports. Media sources also provide names of people that I can follow up with. I note this when I explain my research methods.

Answer 5

Let me provide an example that may be analagous to your current situation.

GPGPU (general purpose GPU) programming first started taking off in its current form when NVIDIA released the GTX 8800 graphics card family and the CUDA programming language back in 2006 through 2008.

This represented a step change from the previous GPGPU techniques in that it was the first true dedicated programming environment that didn't require a connection to a traditional graphics rendering pipeline.

The details of the GPU hardware architecture were not well understood and the only material available on the details were some marketing materials from NVIDIA.

Various research papers started cropping up at the time that attempted to work out the details of the underlying hardware architecture. Basically, they tried to take proprietary NVIDA R&D, understand the how it was done and explain it to the academic community. A good example of this is this paper:

Hong, Sunpyo, and Hyesoon Kim. "An analytical model for a GPU architecture with memory-level and thread-level parallelism awareness." In ACM SIGARCH Computer Architecture News, vol. 37, no. 3, pp. 152-163. ACM, 2009.

These early papers tended to reference two major sources of information:

1. Marketing and technical publications from NVIDIA (in your case, from the Google product / service)
2. General technical material on GPUs, hardware architectures and specialized programming languages (in your case, AI in general and the math or other techniques your specific AI applications build on)

Answer 6

There's nothing wrong with using popular articles as a source of information - for example if I were trying to get a sense of a new field, Wikipedia is one of the first resources I make use of. But using these as primary sources is very iffy. These popular-level articles are written by people who read the research works and then simplified them for laymen. If you're working at PhD level, surely you can read and understand the research works too.

The situation you describe is rather weird. If you're just looking for a reference for "Google has done this", then it'd work as a source. However if you want something more substantial, and if Google is holding some private information which they're not telling others, then the popular-level articles aren't going to help. They won't contain the relevant information either, and you won't be able to duplicate their results using only those. What are you hoping to get out of them then?

In any case your advisor is the best person to ask about this, because he/she will have a better idea about what research question you're hoping to answer, and how to go about answering it. It is plausible that your advisor wants you to reverse engineer Google's product based on the popular-level article, and if that's indeed your goal, then the answer to your question is "yes".

Answer 7

With all due respect, I believe you are mistaken. There is a lot of published research on AI, including the work done at Google. Of course, they don't publish every piece of work they've ever done, but there's a lot out there. For example, if you were interested in something related to AlphaGo, you should look into

Silver, David, et al. "A general reinforcement learning algorithm that masters chess, shogi, and Go through self-play." Science 362.6419 (2018): 1140-1144.

With that said, I see a lot of people, especially those new to the field, finding the papers hard to read, or hard to track down the paper that answers their question. If one wants to learn about how to use a particular implementation of a particular method, I've heard many times now that it's much easier to read blog posts than to read research papers.

If that's what you're doing, that's fine; technical papers aren't always the best place to learn "Hello World". In whatever paper you write, you generally don't cite the website that taught you how to use the package you used, but rather the paper associated with the package you used. For example, if you are using TensorFlow, you should cite

Abadi, Martín, et al. "Tensorflow: A system for large-scale machine learning." 12th {USENIX} Symposium on Operating Systems Design and Implementation ({OSDI} 16). 2016.

rathe than citing

https://www.tensorflow.org/tutorials

even if that's where you learned how to use the code.

Answer 8

I can suggest a case in which it might be appropriate, but in general, it would be risky. As you say, the sub field you are exploring is new and little if anything has been published.

Suppose you find a claim in an article that you can base your research on - either proving it or refuting it. The "idea" for the research comes from a reading of the article. Research that refutes a published claim might be stronger than supporting the claim, I think. This is because the person(s) making the claim may have done unpublished research and have some "proof" that it was correct. If you merely confirm it, you are just following, not leading.

But, you also don't say how far along you are in your research. If you are at the beginning, things may change and other publications may speak to the same issue before you finish. So, the question may be moot. But if you are near then end then you do what you can.

Certainly a "green field" topic will find little to cite other than the general literature on the larger domain.