We read the paper that forced Timnit Gebru out of Google • Hacker News Board

Re: We read the paper that forced Timnit Gebru out of Google

#531

Post by breatheoften » Sat, Dec 05, 2020, 5:11 PM UTC

free_rms wrote: post #468

qsort wrote: post #428

Earlier quoted context omitted.

Playing the devil's advocate here, I'm with you that half of AI ethics is obvious and the other half is wrong, but is't it the goal of the field to try and give meaning to things that aren't obviously well defined? To make an example that's en vogue right now, AI explainability. Nobody even has a definition of what it means for a model to be explainable (is a linear regression "more explainable" than ML? isn't Google…

Yes, a linear regression is explainable. You know what your features represent, and the regression puts a straight-up co-efficient on each of them. You can look at the weights you trained, see how strongly your input features contribute to the output prediction, and say things like "if all else is equal, 25-49 males are 0.08 likely to click on this ad". Deep nets aren't explainable in that way, even if you devoted a…

Even linear regression is not very easily explainable for real world systems with currently utilized technologies -- especially when you have a virtually unlimited range of mechanisms intentionally or unintentionally subset or modify the sampling process (or select for desirable results at "sampling design time") ...

Re: We read the paper that forced Timnit Gebru out of Google

#532

Post by kenjackson » Sat, Dec 05, 2020, 5:13 PM UTC

tinyhouse wrote: post #522

kenjackson wrote: post #510

Earlier quoted context omitted.

That’s the beauty of modern discrimination. You never come out and say you’re discriminating. Rather you just apply the rules selectively. An inconsistent process is the best best way to discriminate. My prediction is you will see some people leave Google due to this. Likely those who worked closest to her. Apple is already extending interest.

There is no evidence they applied rules selectively. What I meant by inconsistent process is that it's subjective. These process usually say something like "Two L6+ need to approve...". Then it's assigned to someone and that person can spend 2 minutes or 2 hours on it. Paper reviews are esp prone to that. How likely your paper being accepted is often more of a factor of who reviews it than how good the paper is. The…

Other Google researchers have stated that they’ve never had their papers internally reviewed based on the criteria she did (lit review for example). Just for things like corporate sensitive information. If no one else in Google history has had the same criteria used, that seems selective to me.

Re: We read the paper that forced Timnit Gebru out of Google

#533

Post by d_e_solomon » Sat, Dec 05, 2020, 5:13 PM UTC

thu2111 wrote: post #512

kevinventullo wrote: post #434

Earlier quoted context omitted.

If you train an AI on the largest dataset it's possible to obtain then you have, almost by definition, done the most you can to avoid bias of any sort All politics aside, this is not even true for toy ML problems. If I’m trying to do digit recognition and “all the data I can find” is a billion hand-written 0’s and a million hand-written 1’s through 9’s, naively training on that data will yield a model that’s pretty c…

We're talking about tech firms that have access to the entire internet and use it. Hypothetical examples involving imaginary datasets that nobody would use don't prove anything relevant. And note that my argument is not about whether you actually avoid bias, it's about whether you've done the most you can do to avoid it. If you used all the data you've got, then you've done the most you can, even if for some reason t…

> We're talking about tech firms that have access to the entire internet and use it.

I think you're trying to say that a large enough dataset will be free of bias. I don't see how that follows. If I train a model on home mortgage decisions, I will replicate the bias on that currently exists on that dataset - https://news.northwestern.edu/stories/2020/01/racial-discrim... - unless there are conscientious choices to reduce that bias.

Researchers in ethics in ML are specifically trying to enable tech companies to do a better job of not replicating bias and justifiably point out where that is occurring.

Third, I would argue that applying an ML model to do something faster if it replicates the bias of a previously human decision is even worse. The bias has taken the human element completely out and systematized the bias and made it possible with even less friction.

Re: We read the paper that forced Timnit Gebru out of Google

#534

Post by kenjackson » Sat, Dec 05, 2020, 5:14 PM UTC

mlthoughts2018 wrote: post #526

kenjackson wrote: post #500

Earlier quoted context omitted.

Did she say to stop doing your job? Or to stop writing about achieving diversity? Damore said more than just that there might be reasons for the lack of women in tech. He also said that we should stop trying to increase their representation. Among a bunch of other things.

She said to stop doing your job. Re: Damore, I agree. He was pushing a bad agenda that tried to draw more than is reasonable from some academic research on gender disparity. Very, very similar to what Gebru did as well. First by pushing agendas in a publication leading to its disapproval, then in her follow up email.

I can’t find where she said to stop doing your job. Can you quote it here?

Re: We read the paper that forced Timnit Gebru out of Google

#535

Post by labawi » Sat, Dec 05, 2020, 5:14 PM UTC

qsort wrote: post #408

rococode wrote: post #118

Earlier quoted context omitted.

Carbon emissions arguments tend to ignore the value of what's being done as well. BERT and other transformers were meaningful experiments that were valuable in furthering a major research direction and enabling more effective consumer and business applications. In that sense, it's like any other company doing R&D - of course energy will be used and of course there will be some inefficiencies. I think it's quite misle…

This is exactly the problem I have with naive environmentalism. The most recent data I could find for the United States total carbon footprint was 5 billion (metric) tons in 2016. Total energy consumption of all computers, mobile phones, datacenters, servers etc, combined , isn't even a percentage point of that. Yes, CO2 emissions are a problem. You are not going to solve that problem by targeting sectors which entir…

I think the main was sidetracked. If we collectively emit as much CO₂ as 0.1 or 100 persons lifetime for a single model, every year .. who cares? I don't.

But, what if Amazon wants it's own model with its own curation? Maybe we need different languages, maybe countries would like to have their own model with a different world-view. Why shouldn't a researcher train their own model, maybe experiment with different versions? Why should consumers be relegated to pre-trained model with inscrutable preconceptions?

Re: We read the paper that forced Timnit Gebru out of Google

#536

Post by eecks » Sat, Dec 05, 2020, 5:14 PM UTC

derangedHorse wrote: post #454

eecks wrote: post #410

AI ethics research is funny. It's obviously important but it's also kind of ... obvious. I am surprised they get paid so much. * Lots of processing uses more energy.. * Large amounts of data might contain bad data (garbage in, garbage out) * Wealthy countries/communities have the most 'content' online so less wealthy countries/communities will be unrepresented. here's some more: * AI being forced to chose between two…

It's always obvious in hindsight, much like what can be said of something like Dijkstra's algorithm, but the fact of the matter is not everyone can spend the energy to both understand the context of the problem and direct their attention towards the evaluation of ethics within that context. Some people even find it hard to understand the situations of others enough to identify where their technologies can be used for…

> wouldn't be touted as a research point in a respectable publication of AI ethics

Check out https://www.moralmachine.net

> Maybe the following revised statement would be closer to what we can call a contextualized "truth"

Your rewriting of my comment and critique demonstrates that I got my point across. I put perfect in quotes on purpose. The whole example could be full of the same.

Re: We read the paper that forced Timnit Gebru out of Google

#537

Post by free_rms » Sat, Dec 05, 2020, 5:17 PM UTC

the-dude wrote: post #427

danielheath wrote: post #303

Earlier quoted context omitted.

She offered “to work on an end date”. Whether that counts as a resignation or not is likely to be a matter for the NLRB.

Suppose your girlfriend announced she was working on an end date , how would you feel?

I'd fire her in response, it's still a firing.

This technicality matters for unemployment and COBRA, and it has a concrete definition.

Re: We read the paper that forced Timnit Gebru out of Google

#538

Post by melling » Sat, Dec 05, 2020, 5:17 PM UTC

marcinzm wrote: post #425

melling wrote: post #9

Earlier quoted context omitted.

Google claims they neutralized their legacy emissions over the entire history of the company. Now about the billion cars on the road and the 40% of the world’s electricity being generated from coal.

The existence of other problems does not make a particular problem go away and humanity can in fact focus on more than one thing at a time.

Google is carbon neutral.

How much more of a problem are a billion cars and 40% of our electricity being generated from coal?

We’ve squandered decades ignoring the big problems and now people want to run into the weeds with a thousand little problems, that individually don’t amount to much.

Re: We read the paper that forced Timnit Gebru out of Google

#539

Post by the-dude » Sat, Dec 05, 2020, 5:19 PM UTC

free_rms wrote: post #537

the-dude wrote: post #427

Earlier quoted context omitted.

Suppose your girlfriend announced she was working on an end date , how would you feel?

I'd fire her in response, it's still a firing. This technicality matters for unemployment and COBRA, and it has a concrete definition.

This is not a discussion about the technicalities, I am sure you are aware.

Re: We read the paper that forced Timnit Gebru out of Google

#540

Post by YeGoblynQueenne » Sat, Dec 05, 2020, 5:19 PM UTC

thu2111 wrote: post #376

hezag wrote: post #49

I did not read the paper (just like most people here), but by the title — “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?” — it does not look like the CO2 emissions thing is the main topic of this research. BTW, "Stochastic Parrots" is a very descriptive name for the problem > Moreover, because the training datasets are so large, it’s hard to audit them to check for these embedded biases. “A me…

Why? It sounds like ideologically driven circular reasoning. If you train an AI on the largest dataset it's possible to obtain then you have, almost by definition, done the most you can to avoid bias of any sort: the model will learn the most accurate representation of reality it can given the data available. Gebru is the type of person who defines "bias" as anything that isn't sufficiently positive towards people wh…

>> Gebru is the type of person who defines "bias" as anything that isn't sufficiently positive towards people who look like herself, not the usual definition of a deviation from reality as exists.

I haven't seen this "usual definition" of bias as a "deviation from reality" that you give anywhere before. Can you say where it is coming from?