CS-6603 - AI, Ethics & Society
| AIES, AI-Ethics |
Toggle to Select Spcific Semesters
Reviews
If you’re planning to take this class you already know what you’re getting yourself into. Just like everyone else has said this class is easier than a high school class. The assignments are pointless and repetitive (Like copy/paste lots of numbers pointless). I did not watch a single minute of the lectures and got a 99.7% on the midterm and 100 on the final. I did each homework assignment the day it was due and had little problem. I’d recommend this class if you need to get an A. Just turn everything in and I promise you will get an A.
Fairly easy class overall. As other students have mentioned earlier, the assignments are very repetitive and time consuming at times. This is a great class to take if you have a lot going on in your life.
I very, very strongly disliked this class. I have bachelor’s degrees in philosophy and applied ethics. This class should not have ethics in its name. I was hoping for an AI ethics course. This is a data science course.
This class is extremely easy, so its great if you want to pair with a harder class or if have demanding work/home life already.
With that said, this has been my least liked class so far (out of 8 total). The entire class content could have been condensed into a 3 or 4 page paper. To be clear, the content is important - people making and using ML/AI need to account for potential bias in the training data - but this just shouldn’t be a semester long class, or a masters level class. The assignments are extremely repetitive/boring. They are easy, but you’ll come to dread doing them because or how pointless they are. You will spend more time reading the instructions than you will doing the assignments.
Like many reviews say, this course is easy if you have a data science background or you have a teammate with such experience. Working on it alone with no prior experience definitely poses some challenge. A lot of time gets spent on plotting graphs and figuring out how to do the requirements. A realistic week might be 10 hours/week while doing projects and having no prior experience.
I feel like this course got tweaked to increase the workload and the end result is you have a lot of pointless busywork in assignments and some assignments can be kind of challenging if you don’t have a data science background because it’s asking data science questions without teaching you how to answer them
The ethical discussion topics were interesting and can be banged out in like 20 minutes
Assignments are kind of hit or miss with some being a decent amount of work
Take this course if you have a data science background and are looking for a super easy course or if you don’t have a data science background and are looking for a semi-easy course
Rating this course as “disliked” because there is a lot of pointless busy work (e.g.: Make a graph for this 1 set of data and then repeat the process 10 more times to create a total of 11 graphs so you end up wasting hours of time without learning anything)
I regretted not pairing this course with a more challenging one. This course is very easy, although some assignments can be annoying because you are creating the same figure 20+ times. The lecture is interesting but does not go into too much depth in bias mitigation.
TAs are great and the grading is fair. My main suggestions are 1) pair it with a more difficult class, or save it for the semester that you barely have any time for class; I spent on average 2-3 hours a week; 2) the group project allows for individual submissions. Do it yourself instead of finding a group. The project is very straightforward and I was able to complete it in one day by myself and got 100/100. You will probably end up wasting more time communicating with group members.
This was my 4th class in OMSCS. I was in the middle of a job change, moving from one state to another and took this along with INTA-6450 ( Data analytics and Security). This was an easy pair except for the week i was traveling. I was able to work ahead in AIES and finish the assignments to get some time for my move.
The assignments were easy to finish but gave a lot of learning to me since i am new to pandas and Scikit-learn. It seemed like a lot of unnecessary work plotting 20 graphs with different combinations and writing reports. This class is a good introduction to the data analytics and provides exposure to jupyter, pandas, matplotlib, and makes you familiar with handling/manipulating data and displaying graphs. For some of the assignments, instructions were not very clear, so there were lots of piazza threads asking for clarity. Vijay from the TA team was very active and helpful in responding to all the queries.
The mid term exam was a mess for me because of honorlock issue that kept blocking me from opening any document on my laptop even though the exam was open book. I did score around the median and was able to get 90+ in overall.
The final exam was just about picking up an article related to AI ethics and write a report about it with the supporting data. The difficult part was finding a recent article/video published within last 6 months that had enough supporting data reference in it. Once you find the suitable article, it was pretty easy to write the paper.
Final Project was a group project (group of 4) where i struggled a bit because most of the group members were absent due to personal and professional reasons. Thanksgiving was also one of the reasons our group couldn’t meet. Most of the team members were more comfortable working with Excel and didn’t want to use python for the project unless it became absolutely necessary. I had to do most of the coding 1 day prior to submission for the steps where we needed to submit the code. Most frustrating part was someone in the group asking : why we used certain function/method at the last moment when we didn’t had time. So if you get a good group, the final project should be easy-peasy. The grading on final exam and project was very lenient.
Last week of the semester was little busy with final project, final exam , critique and discussion all pending around the same time.
Overall, an easy class with some learning for me.
This was my 8th class in the program, and I have experienced both higher workload classes (AI, recent CP to name a couple) and this class is around half the workload of those.
I went into this class a little skeptical, but I was pleased with the course. The lectures were well done. I can’t say that all the lecture material was needed for the projects and exams, but certainly make sure you pay close attention to protected classes and domains as those show up on nearly every assignment.
If you’re looking for an intro to the data analytics space, this is a good prep for that with exposure to jupyter, pandas, matplotlib, and gets you use to handling data and displaying some graphs.
Grading was fair in my opinion as long as you “checked the boxes” of everything the assignments asked for. Most assignments had some vagueness, but found that a lot of it can be done at your own interpretation as long as you describe what you did.
The course presents very interesting topics and discussions. It also serves as a great introductory course for data science, pandas, numpy, etc.
As others mentioned, the course itself is easy when compared to other courses in the OMSCS program. However, easy is subjective. The course is easy for those already experienced in Python / Pandas, etc. but it may be challenging to those who lack experience with Jupyter Notebooks.
While I did have experience in Python, this was my first time deep diving into Python for Data Science. By starting assignments early and watching tutorials on the side, I found the homework to be very easy. However, it is time-consuming. My advice is to start early and don’t leave your homework for the last-minute.
My only complaint about the entire course is the way the instructors wrote the assignment instructions. I found that I often spent a lot more time trying to decipher what the instructions were asking for than actually working on the assignment itself. I found that classmates experienced the same issue and would post for clarification on Piazza. If a clarification in Piazza was made, the instructions were never updated to reflect that clarification. Because of this, I had to skim through many posts to ensure I’m completing the assignment correctly.
Anyways, the class is also composed of discussions which are very easy. It’s just a matter of answering the prompt and responding to two other classmates. The midterm was open notes and consisted of a mix of multiple choice and write-in questions. I felt that there were trick questions but received a passing grade either way. The final exam was converted into a take-home essay that had its challenges but it wasn’t difficult. I felt that the grading was more than fair. I received perfect scores in all assignments minus the midterm. I am in no way an expert in Pandas but this course is definitely do-able as a beginner.
I can see how this course would be a freebie to those whom are more experienced in the field of machine learning. I can see how someone who has taken courses such as ML or AI would find this class boring. However, the course is meant to be an introduction. I enjoyed it.
The class overall is very easy compared to other OMSCS classes. The content is interesting and somewhat engaging but it defintely won’t be nearly as technical as other classes. As a word of caution, though the class is easy some homeworks and projects are time consuming. The trick for these is to simply just put down coherent thoughts and as long as it makes sense you’ll be able to get High A’s. The midterm was suprisingly harder than i expected but nothing crazy. Final project is a good chunk of work however so try to find a group for that.
Yes it’s a “freebie” but it also does its goal of teaching you things without beating you over the head with it and boring you out. While the material and level of effort is more indicative of an introductory class, the discussions and lecture content are at least pertinent to current events and can be applied directly to them. I’ve taken other “freebie” classes that were just absolutely painful to get through (boring projects, boring reading, boring lectures, etc).
As other reviews mention, this mainly focuses on discerning bias and ethical harms towards protected classes (age, gender, etc). There’s some very rudimentary statistics and bias mitigation techniques taught through the lens of discussions and homework assignments where you apply them. If you want you can pretty much skip the lectures altogether and just do the assignments without too much difficulty.
Exams are very simple and straightforward, no gotchas, simply just reviewing concepts that will be (at that point) well-reviewed and discussed a lot. I can see (from ML point of view) how this class may be frustrating (it doesn’t really do nearly enough exercises in bias mitigation/examining ML implementations and strategies for ethical concerns) because it mainly focuses on the social side of this topic, but unless you signed up for the class without reading reviews then some part of you just wanted an easy class with low level of effort and this is exactly that.
Get a grip, y’all. Take a hard look at yourselves.
- You came to OMSCentral to read the horrible reviews.
- You are under no obligation to take this class. There are other ML electives available. RL, DL, etc.
- You basically chose this class anyway (1) as a freebie to fulfil your ML elective; (2) because the workload is low; and (3) you don’t need to spend on your non-CS/CSE token.
- You came back to OMSCentral to remark further that this class isn’t worth your money. It’s a freebie, duh.
See this vicious cycle?
My background is in data science. I took AIES because I am in another course this semester that requires over 20 hours/week, so I just needed something light to pair with it. For this purpose, AIES was satisfactory for me. The TA’s (especially Vijay on Piazza) are very responsive, timely with grading, and generally helpful. I think a lot of student questions were due to a bit of ambiguity in the assignment descriptions, so maybe they could look into improving those.
I gave this course a “Dislike” because the only thing we’ve covered all semester is “Protected Classes”. Protected Classes are features in a dataset like race, age, sex, etc. You don’t want to build an AI that incorporates historical biases due to an unbalanced distribution of such features. In the preceding sentences, I have covered what you’ll spend a whole semester on. It’s tedious to do assignment after assignment on the same thing. I do think it’s possible to expand the breadth of topics covered in the class while keeping it brief and engaging, but it would need to be overhauled.
The lectures are pretty good. Prof. Howard picked rather relevant examples of AI ethics issues. I thought they were well structured and clear.
There are only about 5 coding assignments. I think they took me about 5-6 hours each (but I have 3 hours/week since other weeks were just discussions on the class forum, which were ~30mins to do). The assignments are more tedious than difficult (little things like formatting tables, copying and pasting text, etc. take up the most time). They kinda felt like busy work. If you’re learning Python and Pandas/Numpy though, I could see them being a good warm-up to harder Python-based courses in OMSCS.
I haven’t taken the final yet. The midterm was pretty easy, and I personally don’t think any preparation at all is needed to do well on it.
The final project is perhaps the most useful part of the class if you do it all on your own instead of in a group. I learned a decent amount with it regarding fairness metrics to gauge “unbalancedness” and how to reweight the data to improve fairness.
This was a waste of a class. I, like many others here, took the class as a freebie, but I can say that I sincerely learned nothing. It was extremely repetitive- I think the professor’s emphasis on protected classes was important, but every assignment and video focused on some iteration of that. We didn’t delve into different schools of ethics, and no part of the class required any critical thinking. My undergrad ethics class was far more challenging and thought-provoking than this one. It’s disappointing, because such an important topic deserves to be treated with respect. I agree with other reviewers who mentioned that Joyner should remake this class. It’d be far more engaging, challenging and thought provoking.
Other reviews already provide a synopsis of what to expect. I took it as I wanted a light course load with straightforward assignments. The only thing that is somewhat time consuming in this course is the homework assignments, but they can be knocked out in a half day or day if you’re somewhat familiar with Python using some of the data science libraries. Given the other reviews, I went in with low expectations and was actually a little surprised that some of the Python homeworks weren’t as straightforward as suggested. Not difficult but not as gimme grades as what I anticipated.
Other than that, everything else is coming up with reasonable answers for participation. The material is a little repetitive about what bias is and most of the information is common sense. The midterm was a little nitpicky but nothing too bad to ruin your grade. As others mentioned, take this course if you want a straightforward course or a lighter workload. Happy I took the course after reading the reviews as they painted an accurate picture about what to expect and motivations to take the course.
I agree with most of what is written already that this class had a lot of potential since the topics are of incredible importance and relevance in today’s society, but the execution is poor.
I feel like if Joyner took over this course, it could be an HCI-style class with weekly writing assignments to make you think deeply about how ethics play a huge part in everyday decisions and is something that computer scientists need to grapple with as ethics start to take center stage in many of today’s innovations and technologies.
I would not underestimate the assignments. They are tedious and took me quite some time to complete. They are not hard by any means but they require making a ton of graphs, data cleaning, and interpreting the instructions. The final project I chose to do alone and oh boy, it was rough (time wise). I got it completed but it was no fun.
This is a good class to pair or if you just want an extremely easy semester to rest up from harder courses. The ML track has a ton of time consuming courses and it’s not a bad thing to have a freebie to help you to gain your strength back.
This course is disappointing because it has the potential to be a really interesting topic and it completely falls flat. The lectures cover some interesting material, and some of the supplementary readings are insightful, but the graded work is so easy that a dedicated middle schooler could probably get an A in this course. The most complicated assignment in this course requires you to plot about fifteen graphs and calculate mean, median, and mode of some data.
There’s “exercise” (the class’s term for a discussion prompt) that requires you to use an online tool to attempt to mitigate bias in a dataset. The tool links to a python library that you can use on your own to attempt your own manipulations to datasets. In the project that focuses on dataset bias, the course staff recommends against using this library, as it is somewhat complicated to use.
This is a computer science masters program. Having a course that requires effectively zero computer science knowledge or ability devalues the entire program.
I just took the midterm for this class and it was the worst midterm I’ve ever taken, in my life. It was very easy: What is a mean, median, mode, etc. but that’s not my complaint, I was never expecting to learn a lot in this class based on the reviews here. One of the questions asks you to calculate the Mean of ~25 numbers that range from 10k-100k, from a PNG of the numbers, so you have to copy them by hand into the Honorlock calculator and just hope you don’t mistype any of them. What is the purpose of a question like that???
The assignments so far are ok. They could be worded more clearly, but they’re easy. This class could be vastly improved while still being a freebie. This class actually has a decent amount of potential, the topic is interesting and Python is a useful skill. It’s just executed poorly. Take it for an easy but tedious grade.
End of Semester Update: I think the final was way better than the mid-term. But just know what you’re getting into here. An easy A, but you won’t learn anything, and it’s somewhat tedious.
The TAs are pretty responsive, which is nice. If you want an easy A, take this class. My review is negative only because I’m disappointed. This subject has a lot of potential but this class really doesn’t get deep enough into anything for you to learn anything.
I found the subject matter important and interesting but unfortunately it just doesn’t dive in very deeply. In my opinion a graduate level course should not be wasting time explaining/testing on Stats 101 concepts (mean, median, mode…)
Aside from the lack of depth, my one big complaint was that the class was paced such that there was a ton of work due the last week of class - if you worked ahead it might not have been too bad, but if you stuck with the class schedule like I did you ended up with a week that was significantly busier than the rest of the class.
Basically, if you want an easy semester or need to quickly take care of a Machine Learning or Interactive Intelligence elective, this class is a good fit for you. If you want to learn a lot look somewhere else.
I read the reviews prior to taking the class and took this one because I needed an easy summer course. I got what I wanted, most weeks I put in 2-3 hours of effort. Not so during the final week, where 1/3 of the grade was due all at once. That week I put in over 30 hours, which was my fault because I decided to do the final project alone. In the end, was it a good class or a bad class? Not as bad as I thought, but not very useful either. My biggest lesson was that people who complain about bias are often unhappy about unequal outcomes. Data can be bad, and bias does exist, but if you take a test and the other guy scores better, you’re not doing something right, bruh. How do you make this class better? I’d teach it as an intro to ML using Python, using ethics cases for the homeworks. Right now it’s 80% wokeness 20% Python, the right ratio is 20/80. And that’s it, I’m sure I pissed you off. Now go eat your French cries.
LOL
This class is pretty useless… Most of the projects are really just busy work and you don’t really learn much of anything. Most annoying thing other people have mentioned are the graphs and tables you have to generate to their specifications. Things like make a table of different means and medians.
Took this class for an easy A but straight up didn’t do some assignments if they looked too annoying, but an A was definitely very attainable.
Class can be summarized as:
- Sometimes there is bias in datasets since people are biased.
- Getting rid of bias is hard.
- Bias is bad.
- Make a graph on why bias is bad.
I have taken back to back, very demanding classes (CP and AI), for the last two semesters so I really wanted a major mental break without delaying my graduation. I knew that this was one of the easiest classes in the program, and that is 100% the reason why I took the class.
I agree with other reviewers that it’s extremely easy (probably high school difficulty), but honestly I think most students who take this are counting on it to be a gimme class to either take a break or get a guaranteed A. I never spent more than like 3-4 hours a week on this class, and could complete each of the projects in one sitting. The projects are python coding assignments, and there are a lot of case studies and written critiques that need to be completed as well. Some of these written assignments are just busy work, and you definitely need to be on top of when things are due. There are frequently projects and a few case studies due on the same exact day. Also, I spent more time trying to figure out how to graph things in Python than actually completing the stat/AI part of each assignment. I get graphs are important but I felt this was a big waste of time.
I wouldn’t recommend this class to anyone who actually wants to gain a deep understanding on AI ethics, which is a shame. This is actually a topic that I’m very interested in and is becoming increasingly relevant in everyday life. I will say that the instructor for this class changed this semester somewhat at the last minute, so I could see this class change and evolve in future semesters.
While this class is very easy, its also like watching paint dry. It you want an easy A be prepared to make 50 of the same graph, to watch boring videos, and to work with un-inspiring data.
I think its an important topic, but this class is borderline high school level in difficulty.
This is a great intro to python, statistics, and machine learning course, but not great for data science and machine learning practitioners.
Listen. I know why you want to take this course and you know why I took it. And you’re here to find out whether it actually is what you think it is. I work in ML with Python/Pandas/NumPy everyday, so my review is from that perspective.
You will learn almost nothing, put in almost no effort, and wind up with an A. So if you’re someone like me who wanted a breezy semester between tougher classes or someone who’s doing the SDP/SAD/CN/DBS/etcetera ultra easy degree path (you know who you are) and you just want to check off another class on your way, then take this.
So why do I strongly dislike this class even though it was exactly what I expected? Because it doesn’t change the fact that the course is stupendously easy and belongs nowhere near any master’s degree.
Do enjoy doing busy work similar to your 5th grade homework assignments?
Do you like completing assignments where the greatest challenge is unpacking the terrible instructions to try to understand the task at hand?
Do you like making the same graph 50+ times?
If so, this is the class for you!
This entire course could be reduced to an hour long lecture.
Good class if you’re just looking for the path of least resistance to earning this degree, but a complete waste of time.
This class is incredibly frustrating. I read the previous reviews that it was boring and easy, but thought it could still be interesting (you get out what you put in, right?) and I wanted an easy class for summer. But I regret taking this class. Let me start by saying I’ve been a statistician for 20 years, and I already have one master’s degree in stats. This isn’t even stats 101, this is who-knows-what-garbage. A lot of the stats in here are just plain wrong. Mode is not a valid type of average. Mean is average, mode is most, and median is middle (that’s stats 101!). There are whole lectures on misleading graphs, but then in the homework projects, we are forced to create nonsense graphs that don’t mean anything and don’t show anything useful. Why??? And why are we calculating correlation on categorical variables that we’ve have to make up some nonsensical ordering for? It doesn’t give any useful information, and there are plenty of statistical test for association between categorical variables. And when the instructions are not micro-managing you into producing garbage, they are just plain incoherent, and everyone is confused all the time. This class has been out long enough now that everything should have been worked out. But don’t waste your time. It’s not interesting, it’s not useful, and much of it’s not even correct!
The other reviews are coming from people who might have already completed other hard ML/AI courses, and to be fair they are playing spoilsport for this course. It is like someone who has already completed GA is complaining that SDP is not as hard as GA. Different courses have different goals and difficulty is not necessarily a goal that profesors look out for. And this course is not an easy course if you are new to AI and python in that case I guess this is a very good course, There are 6 time-consuming projects, around 6 writeups and two exams and also debates.
Ok, so this is definitely not a rigorous class like AI/ML/RL/etc. It’s more of a general survey of the many ethical considerations around building AI Systems, designed to get you thinking about these issues. You will do lots of data-cleaning and basic statistical analysis on a number of datasets. I really enjoyed the last half of the class where you focus on bias mitigation techniques and also how to measure fairness. The projects can be a bit tedious, but are still fairly easy if you are comfortable with Python. Midterm was open notes and very fair (no trick questions). The final exam was a take-home essay that took ~ 3 hours to complete.
Worst Course in whole OMSCS curriculum. This is my second last course. Apart from not learning anything this course expects you to have a good python/ML experience. Assignments were extremely boring and ambiguous. The TAs looks like nut jobs, writing ambiguous answers to ambiguous questions. I regret taking this course.
This is my last semester and I took this course along with GA. This is the worst of all the 10 classes that I took. I learned very little. Homework assignments are super boring and tedious and teaches nothing. A waste of time. Stay away unless you are desperate of getting an A.
There is benefit if you do not know anything about python/pandas and other related modules and if you re super rusty in statistics. If that is the case this class will be a valuable force to help you level up.
If you already know ML/DL and take this class, it will feel like a shore. There is extensive data cleansing/normalization and manipulation that is required and that will encompass a great deal of work.
The class overall is ok, and to this point I have not received my grade yet. If what everything they have been talking about this class being easy, it might be because of the lenient grading. This leniency in grading seems to be true so far. However, if lenient grading were not the case, I would have to say this class is not as easy as it has been pointed out to be. There is substantial work and certainly some several hours that you need to put in to get the assignments finished. Also, the assignments are at times not explained well which needs one to ask in piazza a lot to get clarification.
Not a graduate level course. A high school student can make it without putting much effort. The quality is extremely poor even for an elective course. I got nothing from the course. If you are interested in the AI ethnics, there are plenty of online resources much more useful than the course material. I wouldn’t recommend it for anyone except you don’t care about learning and just want an easy A.
Case Studies
Read a linked article and answer a few discussion questions, then respond to two other students. Should take 15-30 minutes each, and overall seems like 5th grade busy work
Homework Projects
The AI/ML assignments were large datasets that involved some mindless drivel of re-organizing data, filtering, and creating graphs/tables. I’ve never used python for this sort of thing, so I just knocked it out with excel. ~8 hrs each
Final Project
Not yet completed, will update after
Midterm exam
Open notes and proctored. Since it’s open notes, no studying is required…most of the answers are simply intuitive
Final exam
Not yet completed, will update after
Overall
Simple course; you may often wonder how this could be considered graduate level. This is an elective, so it checks a box. The content is interesting, but overall seems like a waste.
This is by far one of the most pleasant and well-run courses in the program. It is an excellent start to the program and the course material is interesting. There is a light pace to the course, great for pairing it with another class, or just taking an easier semester. The professor for this course is incredibly accomplished, and the lectures are strongly relevant to the material covered in the midterm. Course material, assignments, and exams all tie in very nicely and this makes for a great student experience.
I took this class and paired it with AI for Robotics, with the expectation that this would be the easy, non-technical class, and I was right!
The class is certainly very easy, and as others have mentioned, I wish that it actually delved deeper into either how to apply AI/ML techniques, or how the bias mitigation techniques work. I think that a lot of the concepts covered could have been condensed. The midterm and final exam project were both very easy, and this class did change my perspective and make me very cognizant of how AI can negatively impact certain groups. The projects were tedious and didn’t necessarily equate to a lot of learning, but still, the workload for this class was extremely light (which I wanted).
You don’t put too much into this class, but you don’t get too much out of the class. If you are looking for an easy class to pair with another class (plus I heard it might be an ML specialization elective now?), you will be satisfied, but you don’t learn that much, so I am a bit neutral overall.
I have a more in-depth video review, where I walkthrough the projects and the downsides of the class here: https://youtu.be/pe4EnivoRHk
Took AIES alongside GIOS as my 6th and 7th classes in OMSCS. I had previously taken AI, RL, CV, ML and HDDA.
This was without a doubt the easiest class I’ve taken by far. I feel that the first half of the material could have been compressed a lot more so as to leave more time to deep dive into the various bias mitigation strategies. As it is, after the class I still have only an extremely surface-level understanding of the different ways to mitigate bias, which is much less than I expected out of a graduate-level class from Georgia Tech. For example, there was an exercise where we discussed bias in word embeddings, but didn’t learn about how exactly to reduce bias in these word embeddings.
I would have liked this class much better if there were lectures that explained the inner workings of bias mitigation algorithms and assignments that made us implement some of these algorithms from scratch, or at least tested our understanding of how they work.
I think you’re going to get out of this class what you’re going to put in. This is a pretty easy course. There’s a book for the course to read called Weapons of Math Destruction, which is a fast read and you can front load. You’ll interact with it rather minimally in the course, but it’s a good primer on what’s to come.
You’ll probably want to come into this class with some proficiency with Python/Jupyter Notebook. You can get away with just using Excel for the first two assignments, but over time you’re going to have to be doing a lot of stuff that will need Sci-Kit, Numpy, and Pandas. The last two assignments also require you to reverse engineer/understand some Python libraries from IBM and Google for helping combat bias in ML algorithms.
In any case, I think you can probably do the bare minimum and spend a very small amount of time in the class and still do well in the course. However, if you pour a lot of time into doing things well in the course, it’s rather rewarding. I came into the class with some knowledge of Pandas/Numpy and some Machine Learning algorithms, I emerged out of it feeling much more confident in implementing things with Scikit.
I liked this course a lot. I went with expectation of getting an easy A ( got it as well). This course is sitter A course. The only way not to get an A is to not write the assignment and exam. But grades aside, after taking this class I realized that this course made me think on important aspect of software engineering and ML. In each of the dataset I used for assignment I was able to find the bias. This was an eye opener course for sure. The course does not have any specific reading , but the course material is thoughts/ideas of your peers. The discussion is the real material. 50% of the class is introduction to ML and python. Since, I have background in both so I did not bother to spend a lot of time in those areas and passed the tests and assignment without going through the lecture. So people with similar background might find this extremely easy too. remaining 50% was not difficult at all. Its just a thought process which the professor wants to develop in all of us. No fancy algo, just ways to find evidences of discrimination based on data, learning and reacting to others views. Conservatives across all the regions in the world might dislike the course.
Do not take it if you want to learn things. I did not learn anything useful in the course. The course materials are poor and the assignments are time wasting. The course tries to teach you the bias of AI/ML with no “series” implementation of any algorithms/programming assignments. Most of time you only need to “talk”. I think the course is a good introductory one for undergraduate students and cannot be a graduate-level CS course.
This class has a lot of potential, but fails in its execution. It is an important topic, and the lectures are outstanding. However, the material is not graduate-level (readings are websites or sources such as the NY Post); there are few academic/peer-reviewed articles used, and the main textbook (though an entertaining read) is a mass market paperback.
The biggest issue with this class is lack of clarity in the assignments; you will spend the majority of your time trying to decipher what exactly the instructor wants you to do. This could be fixed easily if materials were proof-read prior to being released. The mid-term suffered from the same issues: questions were poorly written and unclear.
Overall, the quality of this course is poor, which is a shame because the subject is great.
[Updated after completing the final project and final exam.]
TL;DR The difficulty of this class is on par with a high school course, and falls short of meaningful discussions into biases of AI/ML.
I really wanted to like this class. It has so much potential with its content being incredibly relevant, modern, and of huge impact. Unfortunately it doesn’t deliver and will leave you simply saying “There is a problem with bias in AI (hopefully not new information to you), but I don’t know what is being done in industry to address it.” The final project very lightly touches on bias mitigation techniques, but you’ll mostly just call library functions without understanding how the bias is mitigated. In its current standing this course feels like it should be offered in a MBA program, not a graduate-level CS program due to its shallowness.
Pros
- Minimal time required Of the 8 classes I’ve completed, this course has by far and away required the least amount of time. I have been able to knock out the lectures and weekly deliverables on Monday night. I’ve been able to complete each project in a single session on a weekend afternoon.
- Active instructional team The TAs do an excellent job of answering every Piazza question. Professor Howard participates as well, answering questions and occasionally posting articles to generate discussion.
- Easy If you’re looking for an easy course, I don’t think OMSCS gets easier than this. I have a 100 so far, including a midterm exam. The work required for this course is incredibly easy to complete, as evidenced by the very high assignment averages.
Cons
- Pace The course drags its feet on content that could be covered in a fraction of the time. Time is spent covering middle school-level descriptive statistics like mean, median, mode. You will complete multiple projects that simply process data and report descriptive and inferential statistics. To me this felt like Week 1 - 2 type material, but instead it took us all the way through the midterm. I wish this course spent time understanding the math and theory behind defining fairness and mitigating bias, but it gets brushed over in the last two weeks of the course in the final project.
- Lack of Meaningful Discussion Most weeks we are required to write a response to a case study or writing prompt covering some ethical concern. Part of the assignment usually requires you to respond to 2 classmates. The problem is that there is no incentive to engage in thoughtful conversation. Most responses are along the lines of “that’s a really good point” or “i agree with what you said about X.” Additionally, there is one discussion board for the entire class, so it gets extremely cluttered and infeasible to keep up with.
- Easy This can be a pro if you’re looking to double up courses in a semester, but if you’re wanting to be challenged in understanding bias and bias mitigation techniques in AI or to be pushed into evaluating ethical concerns at a deep level, you’ll be left unsatisfied. I remember the exact assignment where I stopped trying for this course. We were discussing bias in NLP, and were provided a research paper from Microsoft about addressing the biases. After reading it, studying it, and posting my thoughts in my weekly response, the only responses I received were “yeah ok kewl research but the devs need to watch out for their own biases.” face palm Based on my observations of Piazza, this course attracts the weakest OMSCS students who post questions like, “When the instructions say X, I just want to confirm that I should do X?” So it’s best to just ignore Piazza unless you have a specific question.
- Projects The projects are mostly just following a step-by-step series of instructions. I just completed the current project covering word vectors while watching a football game. It’s “do this” followed by “do that.” There isn’t much thinking involved. The projects don’t push you to understand why things work the way they do.
- Supporting Materials Like I mentioned earlier, sometimes papers will be listed as an optional reference, but you won’t actually need to read them for this course. There is a great text listed on the syllabus, Weapons of Math Destruction, but it’s never referenced in the course which is a missed opportunity.
Overall good course. Not too time consuming with busy work. Assignments in python. There are written critiques and assignments. Two exams (midterm and final). Not too much writing (around 2 pages), no right or wrong answer but put forth your argument. For this semester, the final is like a mini-project (open everything except human interaction). Good course for elective in ML specialization. Code in python (Jupyter notebook). Probably need to spend some time finding a dataset for your project. Active Piazza interaction between students/TA.
The course (Summer 2020) consists of an open-book midterm, a take-home final exam, 5 homework assignments, a final project (individual or group project options), and a dozen or so graded discussion questions. This course requires a beginner’s knowledge of python and basic statistics; both the statistics and python portions are covered in lecture.
Overall, this course is pretty easy. The workload is low in the beginning and increases in the last ~4 weeks of the semester. The ideas presented in lecture are easy to grasp and not too technical. The lectures touch on general questions about ethics for AI, AI-related scandals that have been in the news over the last few years, and focuses heavily on bias in algorithms. The more substantial assignments are all in python and focus on mitigating bias in datasets and algorithms, with a major focus on protected classes (race, sex, etc.) The projects are interesting from a data science perspective, and I would say that this course would be good to have for ethical purposes for students who plan to work with AI/ML. However, I would say that this course does not gain you any deep understanding about AI, ML, or algorithms. If you are looking for a prep course for AI or ML, this is not it. Overall, the lecture style was soothing, though not particularly information-dense. The information presented in lecture was related to the homework, so that was a plus.
I would suggest taking this course alongside a harder course.
This is a really interesting class. It’s not a hard class from a coding perspective. If you are somewhat familiar with python, pandas, and mathplotlib, the coding assignments would be very simple to accomplish. Additionally, you also get to explore little bit of scikit-learn (linear classifier) as well. I also enjoyed exploring the AIF360 APIs to detect bias via various metrics (Confusion Matrix, Disparate Impact, Statistical Parity, etc) and mitigation techniques (Adversial debasing, Reweighing, Disparate impact remover, etc).
Additionally, the class provided a good intro to descriptive (frequency, standard_dev, mean, mode, variance, correlation, five number summary, etc) and inferential statistics (randomized sampling techniques, Simpson’s paradox, etc) concepts. We explored bias and fairness topics such as China’s social credit system, or the bias in word embeddings (using word2vec python module), or the bias in the underlying facial recognition libraries.
The recommended text book is Weapons of Math Destruction. Although the text book was not directly used in the lectures or exams, it’s a very good book to read to get more insights. The instructor, Professor Howard is very engaged and a really awesome person to interact with. The class lectures are of high quality. The TA team especially Vijay, Jeanette, Bryan were all very helpful. Vijay answered all piazza responses in a record time and was very patient in his responses. The most challenging part for me was the lack of clarity in assignment directions. I think that’s because we are more used to having “correct/incorrect” answers in assignments instead of exploring questions that are more abstract where there is no clear right or wrong answer. I highly recommend this class. This class along with ML4T should provide a good beginner level intro to ML.
Overview/Caveat
AI Ethics was brand-new this semester. While I haven’t been actively reviewing things in the past, I felt it’s important that folks be given an impression of this class. A lot of people have been asking, so I’ll try and make this somewhat detailed and go into specifics where possible. I’ve completed 8 courses in OMSCS as of this writing as well as my undergrad at GT, for reference.
Take this review with a grain of salt; throughout the semester, the teaching team made adjustments to the course based on feedback from students and also implemented some changes given the COVID-19 situation. This flexibility, in general, is a good thing, and it’s encouraging to see the teaching staff being responsive and making some needed changes. However, it does mean that my review may not be fully accurate to the course as it continues to adapt; I can only speak to my experience.
I’ll start off with good/bad highlights and get into more specifics below.
The Good
-
Content was topically relevant. Students were encouraged to read material from ongoing ethics issues, recent news stories, and thought-experiments relevant to today. The material is definitely not abstract/theoretical.
-
Course doesn’t assume a strong programming / AI background and could therefore serve as a gentle introduction to these topics for folks who are newer.
-
Students get exposure to some real examples of algorithmic bias and can see how marginalized groups can be negatively impacted, even in the absence of malicious intent.
The Bad
-
Theoretical depth is lacking. Initial assignments are just reporting basic metrics using python/pandas and the later ones rely on third-party libraries which aren’t explored from an implementation perspective. The student is on their own to dig into these libraries and see how the underlying algorithms work.
-
The exam was poor. Because the first portion of the course covered material which is mostly review for people, the midterm exam was comprised of mostly trick questions and ‘gotchas’.
-
The participation assignments felt largely superficial. They were the sort of thing which could work well on-campus but hadn’t been ironed out into a good online format yet.
Lectures
- The lectures are professionally done and the professor does a good job ensuring material is engaging and fresh. I definitely skimmed through certain parts, however, as they should be review for most people taking this class in this program. The following are a few examples I feel were unnecessary which were covered at some length:
- mean, median, mode
- correlation vs. causation
- empirical rule
- That’s not to say all the material was irrelevant. In many cases there was a good balance between current events/real-world examples and concepts from the course. I’d just suggest a bit more technical depth.
Assignments
-
Case Studies/ Exercises: These are short exercises designed around a controversial topic, current event, or ethical issue. They seem tailored to facilitate an in-person discussion on a topic where students can debate back-and-forth. Online, however, this fell a little flat. The assignments usually mandated that a student “respond to two others” who had answered the question. Unfortunately most students just did this minimum bar and never replied when someone responded to their post. I hope this would be reworked in the future.
-
Written Critiques: These are more-or-less short essays. The idea is that a student should dive in on a topic or issue and write about it in a bit of length. My biggest issue with these was a lack of clear guidance as to how we’d be graded, but there were some thought-provoking questions
-
Projects: Several concepts in the course are taught in a more “hands-on” approach, using the projects to showcase concepts. Unfortunately these either are so basic as to be uninteresting to someone with a background in data manipulation in python, or very reliant on specific third-party libraries. This resulted in me not learning too much through the projects, which was a shame. It’s cool to see some of the bias mitigation libraries work, but algorithmically I couldn’t explain to you how they work in anything more than superficial detail.
The final project is (optionally) a group project. Since grouping was optional I didn’t see this as an issue. We were also probably helped out by the fact that most students in the course had been through several OMS courses already and so were not totally incompetent.
Exams
The course is slated to have one midterm exam and one final exam which is not cumulative. This semester, the final exam was dropped and a short mini-project replaced it. The midterm was a bit of a trainwreck; the exam mostly covered some of the basic concepts listed from the lectures above, but attempted to trick students doing silly things like switching units or using subtle language tricks to be misleading. Maybe these concepts are emphasized more heavily on campus, but many students in the online section missed these questions. Thankfully, the teaching team allowed for corrections to be made to recover points.
Grading
The grading for this course was quite lenient. Many of the smaller assignments (case studies etc.) seem to be predominantly graded on completion. My written critiques were also marked with 100%, though I was disappointed I didn’t receive any feedback on my responses. Output for the projects is mostly reports with tables of statistics and numeric metrics with some small written portions; these were also graded generously and I never had any of my numbers questioned. Grading for the course was a bit all over the place but the corrections seemed to make up for that. Expect the exam format to evolve.
Last Thoughts
This course has a lot of potential but one could tell it was a first iteration in need of a good polish. The teaching team’s willingness to adapt would suggest to me that the course may improve over subsequent semesters. If you’re on-the-fence, maybe wait a few semesters and see how things progress, but overall this course is a light-workload introduction to a few technical concepts with a solid grounding in real-world examples.
A new course which was offered in thi semester. It is a machine learning specialization elective. Good introductory course to ML. Kinda like ML4T. First part of the course is mostly about understanding data, statistics basics, predictive algorithms. I found the second part more challenging involving new concepts concerning fairness and bias. The assignments are easy. But since there are many exercises, assignments, written critiques do not underestimate the commitment this course requires. Exams are open book, so we can ease there. The professor and TAs are very understanding and help the students with any doubts you have. Overall a good course which will give you a new perspective of seeing AI.
This semester is the first time this class was offered online and while the class isn’t over yet, I know students are registering now so I wanted to leave my impression.
–Overall– This is a great class. The course content is relevant and curated well. Dr. Howard did a great job giving interesting projects that demonstrated, both in code and in real life, how algorithms can be used to create unfair outcomes and unscientific conclusions. She was also active on Piazza and in the Slack. TAs were responsive and turned around grades promptly.
–Details– Class consists of weekly discussions, several written critiques, homework projects, a final project and 2 exams. Weekly discussions are focused on ambiguous real world ethical situations related to AI and students are encouraged to provide their opinions on the topic and respond to one another. There’s no grade for being “right” or “wrong” but more around how thoughtful the answer is. Written critiques are just that, you have to write about an ethical dilemma in existing AI technology, but in more depth. Homework projects were probably my favorite. The course starts out much easier and slowly eases you into further analysis. The data analysis can be done any way you like but some of the data sets are so large I’d recommend using Python anyways to make it easier on yourself.
–Areas for Improvement– If I have any complaints, it’s only that this course could be made a bit more challenging. Not in the way of busywork; the pacing was perfect. But some early assignments are a bit too simple, and I’d like to see more coursework developed around building bias mitigation algorithms, automatically evaluating bias in an algorithm’s output and writing assignments based on current research in the field (perhaps by replicating an existing paper). This type of material was made available for additional reading but wasn’t required in assignments.
–Should you take this as your first class?– That being said, while the course felt easier than other OMSCS courses, I’d hesitate to recommend it as a first class. The course does assume you know a bit about ML already, and it assumes you are a proficient programmer who can figure out how to write your own analysis code. If you’re not from a CS background and are looking for a gentler introduction, take ML4T.
Overall, I think for a first class on this topic, this was excellent. I look forward to how the course will be improved over time.