This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.
Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.
We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:
-
Shaming.
-
Attempting to 'build consensus' or enforce ideological conformity.
-
Making sweeping generalizations to vilify a group you dislike.
-
Recruiting for a cause.
-
Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.
In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:
-
Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.
-
Be as precise and charitable as you can. Don't paraphrase unflatteringly.
-
Don't imply that someone said something they did not say, even if you think it follows from what they said.
-
Write like everyone is reading and you want them to be included in the discussion.
On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.
Jump in the discussion.
No email address required.
Notes -
Maybe Pre-Neural Net AI research. Again depends on where you draw the line on what is "AI", before I get into another conversation with the other AI researcher here. But Machine Learning is entirely considered under the "AI" umbrella and has for a long time, at least a decade and half pre-transformer. Charitably you are describing the Minsky-esque AI researchers who were big into Symbolic AI which was a thing in the 70s-80s, ie almost 3 decades before the transformer.
The 70s may have been when the Woo Era of AI began, but (I would contend) it lasted all the way until the rise of Transformers in ~2017. It's not that neural nets and other Transformer-precursors didn't exist, but they were just one tool among many, alongside symbolic AI, and any of these tools had a seemingly equal claim to be the secret-sauce for general intelligence. In this environment, it was still the galaxy-brain-types like Minsky who could claim the intellectual highground by saying, effectively, "Well yes, that's a cute trick, but it's already incorporated as just one small part of my grand overarching paradigm..."
Of course, the reason why the field attracted so much woo (including Epstein) is because the AI was seen as a "type" of the human mind itself, and so it was thought that by building an AI we would finally understand ourselves and thereby solve in one fell swoop all the age-old problems of philosophy. We saw this e.g. with early (pre-2017) LessWrong, where Bayesianism was put forth as the ideal form of reasoning not only for a hypothetical AI, but also for you, the rational human. And not coincidentally, this was also the time when Ben Goertzel (another close Epstein associate) was most active in that sphere.
Nowadays the woo has been sidelined, not just because we actually have stuff that works, but because there's no longer much hope that AI will help us understand the human condition any better. If we manage to build human-level AI under the Transformer paradigm, we will find its workings just as inscrutable as ourselves.
Does anyone know what, if anything, the connection is between the work that led to transformers and the rapid developments in corporate big data/machine learning (particularly recommendation algorithms and similar marketing strategies) going on at around the same time? The model of AI history in my head has the post-Woo era of AI beginning with the Netflix recommendation algorithm in 2006, but I don't know how accurate it is.
Thanks for this, I wasn't really aware of that history. Almost my entire interaction with neural nets was through the medium of computer vision object recognition and Caffe, plus Q networks when they came along; I attended lectures where people discussed how you could use RNNs to turn Shakespeare into Shakespearish but it never occurred to me how far it would scale or that it was really worth paying attention to.
As far as I can tell, that history is scattered in bits and pieces over the Internet, so it's a bit obscure. But it is one major milestone that's not widely known.
More options
Context Copy link
More options
Context Copy link
I would say only partial.
Recommendation algorithms were generally clustering approaches and support vector machines, I think. Neural networks were regarded as a potential (but not very good) way to do clustering for the most part, and one of the clusterings they were trying to do was vision (image class analysis). The breakthrough moment in my opinion was the publication of AlexNet in 2012, which demonstrated that neural networks weren't just a way to do clustering and image analysis, but that if done right they were a better way to do it and produced a step change in performance on a task usually considered impossible.
Transformers came out of google's work on translation, which has a fairly different heritage. That's less about clustering and more about sequence analysis using autoencoders, recurrent neural networks, etc. Data-driven neural networks were already pretty much considered SOTA in that field if I remember rightly (Google's lead researcher famously said, 'our translation performance goes up 10% for every linguist I fire') but benefited from the massive increase of interest in neural networks, hardware for driving neural networks, collecting big data pools, and general theoretical work on representations/efficiency/etc.
I don't know that AlexNet et al were truly breakthroughs in and of themselves so much as a logical extension of existing NN stuff enabled by the step-change in GPU hardware -- as I recall LeCun was doing similar work in the 1980s, and I myself did a project with a simple one-layer LSTM network in the late 90s in which I had to work at night because the server load involved was bringing down other (so-called) "important workloads" at my department (at a major university) during the day.
None of this was really feasible, even as research, until pretty near 2010 -- the hardware just wasn't there yet.
They weren't, really, I agree. "Deep learning" as in Massively-Multilayer Neural Networks was known to be a good thing ages ago, we needed the hardware to do it. I would say AlexNet was a breakthrough in the since of marking a step change from, 'hey guys if you gave us some really massive computers I bet this toy would do something useful' to 'holy shit guys, see what I can do'.
All credit to the neural network guys for sticking to their guns until the hardware caught up, of course.
More options
Context Copy link
More options
Context Copy link
RankBrain (2015) was pre-Transformer and was used in production search. Basically a learned embedding model for semantic relevancy.
AlexNet was the turning point, I agree.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
I have a few AI books from that era that I've been meaning to flip through again just to get a kick out of the huge vibe shift in the field. When I remember acquiring them 15-20 years ago, it seemed it was all attempts to create bespoke approaches to each aspect of the human experience. Hand-coded language models ("verb", "subject", "object"), visual edge detection, spectrograms for speech recognition, physical world models, and so forth. Nowadays, "lolz, just throw data and compute at it" seems the preferred approach.
To be fair, the limitations of the bespoke methods were known at the time: different words can serve different roles in different contexts. Sometimes we noun verbs and verb nouns. And so on for the other spaces.
Nouning verbs is dangerous. It uses a word that pointedly fails to describe itself, and if you as much as think the word "non-self-descriptive" then you will no longer be capable of coherent thought.
Verbing nouns is safe as long as you verb the noun "verb" first. Whether "verb" is a self-descriptive word or not is unclear, but not dangerous.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link