I have found Ethical Institute to be useful to think about ethics principles in a practical way and use their frameworks/tools to keep a check
Regulating complex things tends to create a false sense of security.
Airplanes and cars are complex, but move slow enough in development to where I guess it works… but look at the SEC and Wall Street for an example of where fast moving, less tangible things get to be less easy to regulate.
How do we penalise AI models to make them ethically correct?
It’s chapter 3.
1 Like
Some companies are starting to explain their approach on AI ethics and share publicly: i.e. https://www.microsoft.com/en-us/ai/responsible-ai , https://ai.google/principles/
4 Likes
It’s interesting to talk about ways metrics can invite manipulation and gaming. I wonder what the alternative is though - how good would hospitals be in absence of regulations around patient wait times? I.e. there’s harm in overreliance of metrics, but there could be harm without them as well.
1 Like
80% of viewing at YouTube came from the recommendation algorithms. I am such a sucker for those recommendations. That algorithm gets me every time.
1 Like
How many of these ethical issues could simply be solved by actually TESTING the models in trials with real world data? Or is even that not enough?
I’m a bit tired of the false sense of security argument at the moment…
Being badly regulated is another topic. You can’t say a thing should not be regulate just because it could be badly regulated.
9 Likes
Though as important a topic is, given it directly mostly addresses the negatives - damages done, these to me appears a very first world problem given we are doing lots of damages both in terms of wars (Middle East), poverty (Africa, Asia, S. America & even within N.America), environmental exploitation (global warming), wildlife damages (extinction events) which are larger damages.
Knowing AI/Algorithms are gonna touch us in every aspect, AI ethics is definitely very important - not downplaying it. So there is a dichotomy between the thought above (relevance) vs importance. Do others also carry any such?
1 Like
Thanks!
Is there any downside with Federated learning? Especially Homomorphic encryption.
1 Like
Have people investigated biases that emerge from models other than deep learning/optimization models like agent based models?
Not my argument. I’d like regulation of AI models. Question is more along the lines of how to do so accurately, and without regulatory capture like the SEC.
The FAA has a pretty great record, recent Boeing issue not withstanding.
1 Like
I think it is very important for other parts of the world, especially in areas with high cell phone usage. Eg. many countries in Africa have high cell penetration, and people get their news from facebook and whatsapp and youtube. Though useful, it has been the source of many problems
4 Likes
@rachel When we talk about ethics, how much of this is intentional unethical behavior? I see a lot of the examples as more of incomptent behavior / bad modeling where the product/models are rushed without sufficient testing/thought around bias etc. but not necessarily mal-intent.
6 Likes
I believe this strategy is going to change after COVID. I think we are gonna be more likely to create and support products that make us more resilient rather than optimizing for better, faster, cheaper.
Do we think that Deep Learning, with its ability to recognize patterns, will be able to recognize ethical behavior? Will it ever be able to do a better job than humans?
3 Likes