Hello!
We are a non-profit association developing free and open source machine learning software based on Julia. The software will allow people to use machine learning tools without the need for programming, but with high control over these tools and with Julia integration. Screenshots of creation, training and validation of a neural network can be seen here for those who may be interested. We plan to also release a Julia package, so that the GUI can be used from within Julia.
We would like to know whether donations may be enough for us to work full time on that software. If you found software such as ours useful, would you consider donating monthly?
As a hypothetical member of your target audience, I would probably pass this over since said code doesn’t seem to exist anywhere on the public internet. I think you would have a much better reception if you actually materialized this open source software. A couple of screenshots on Google drive tell me nothing useful about it, let alone how it’s differentiated from DrWatson/MLFlow/Trains/Tensorboard/Sacred/Aim/Pachyderm/Kubeflow/the other 100 or so “ML tools” projects (of which more are springing up every day).
I see you have a GitHub page. Why not publish your code first, get some eyes on it, drum up some community rapport and then poll about funding? As-is you’re unlikely to get any useful feedback.
Thank you for your feedback! We fully agree with what you said here and editted our post accordingly to avoid any further confusion. It is currently more of a hypothetical question. We would like to understand the general attitude towards donations in case of software like ours since there are no estimates that we could find.
It’s really quite impossible to answer this hypothetical without knowing more about your organization and tools. How is it different than existing efforts? Why is it distinct and separate? How big of a benefit will it provide? Why not apply to become a NumFocus sponsored project?
In fact NumFocus is a great reference org for you in general.
Sure, we can shortly describe ourselves. Currently we have 3 people in our organization, where two of them are software developers. We implemented the organization in such a way so that it is democratically governed by users and, therefore, decisions such as what features get implemented and where the money goes are decided by voting. The software that we are developing will be a standalone software written in Julia and QML. It will have machine learning methods and data visualisation tools. The main goal is to make the user interface in such a way so that as much complexity as possible is initially hidden so that people who are not familiar with machine learning can get the desired result without getting overwhelmed, but also to provide a way for specialists to modify any parameters they want or to implement custom code. None of the existing software was able to check all our boxes which prompted us to develop it ourselves.
NumFocus is surely a great organization and a way for projects to get funded. Currently, however, we consider it to be better for us to be financially independent. In the future it may change depending on what the users consider to be the best for the software and our organisation.
Great to see you all trying to make this a full time gig. In my personal opinion, the best way to do this is OSS everything. If you provide enough value to users, people will happily sponsor the project on GitHub. Feel free to ping me if I can help at all!
Since I don’t consider myself to be part of your target audience (ie I know how to program, and prefer doing it), I would not donate. I imagine that this applies to a lot of people here, so maybe this forum is not the best place for this kind of question.
While I think that ML is a wonderful technology, it requires some level of understanding to use sensibly. Although point-and-click ML is probably inevitable, my expectation is that it will be used by a lot of people who won’t be able to use it sensibly, generating dubious outcomes. This is happening already to a large extent in science — certain fields have not learned to interpret OLS and p-values after decades of use, so I shudder to think what they will do with point-and-click ML.
We agree that it is a serious concern and we plan to try to mitigate it as much as we can so that people do get sensible outcome. However, even in those cases where people fail and do something nonsensical and, therefore, get nonsensical output, it is better than if they did not try at all. Based on our experience the barrier to entry is very high in ML for people not familiar with coding. If such point-and-click software does not exist, then many people who can benefit from using machine learning may never even try it.
Not when they lack the background to see if the results are nonsensical. Then I think they are worse off. Of course to a large extent this is their problem, but there are always social spillovers.
That said, I am pretty sure that point-and-click ML will happen and there isn’t a way of stopping it. But since it is a product/service, I imagine it can generate revenue, like point-and-click stats software does.
Policy makers usually have scientific advisors who know how to use machine learning and will do it for those policy makers if required. This should be the case for most people in position of power.
Some harm is inevitable, that is true, but ML is surrely misussed currently as well, but it does not mean that we should make the barrier to entry even higher to avoid that.
Arguably, coding proficiency is not necessary to understand and use ML (or other predictive methods, Data Science basics, probabilities,…), although it often goes hand in hand these days. But coding-less ML may allow people to specialize on the data science aspect of things in connection with domain knowledge more.
I’m thinking of medical professionals - but it may apply to any domain - the intersection of people with medical expertise, ML knowledge and knowledge of proper data acquisition and processing is really limited at the moment, and I hope that this development may allow more people with expert knowledge to be able to learn basic data handling and ML application and statistics without requiring them to learn basics of programming and a language, too.
Totally agree, there’s always two sides of the medal.