One data scientist on the hype around artificial intelligence (2017) | Hacker News Reader

One data scientist on the hype around artificial intelligence (2017)

builttoadapt.io

52 pointscontrarian_27 comments

This is a negative value article. It spends a lot of time dwelling on a low-value semantic argument that "when people in business say 'AI' they really mean machine learning". Sure, but these words have been used synonymously for a long time - my grad degree ~15 years ago concentrating in machine learning was called "AI"; who cares? Then the article goes on to claim:

>I anticipate that after this passes, we can start to do the right thing — focusing on using machine learning to build things that are meaningful and realistic.

UGHHH. Why did you hide this in a long article making the 180 degree contradictory point that AI (which really means machine learning, remember?) is dumb?

This article is a disservice to its reader because it downplays a huge shift in the world that any ML practitioner should understand very well: machine learning/AI will ultimately replace a lot of what humans do, and it's moving at a faster pace than ever. I've led several small-ish projects (1-2 people, 3-6 months) that could replace dozens or even 1000s of experts in their respective fields. There are thousands of people like me, and there are more every day. The buzz may wear off, in the same way Time Magazine stopped talking about how the internet was going to take over the world after the dot-com crash, but the effects and the efforts will continue unabated.

> I've led several small-ish projects (1-2 people, 3-6 months) that could replace dozens or even 1000s of experts in their respective fields.

Can you give some specifics on projects?

In one project at a previous job we were able to pinpoint the genre, mood, instrumentation, etc of any new song with (usually) better-than-human accuracy in milliseconds using deep neural networks. This is better in pretty much every way than the 1000s of music experts that are employed by competitors and vendors.

https://tech.iheart.com/mapping-the-world-of-music-using-mac...

Not to mention personalized recommendations, which basically aren't possible at scale without some level of ML:

https://tech.iheart.com/mapping-the-world-of-music-using-mac...

https://news.ycombinator.com/item?id=12269568

The thing to keep in mind is every machine learning practitioner who is worth their salary is doing stuff like this. A lot of our every day work isn't as sexy as teaching a computer Go, but it's game-changing to most industries.

Whenever there's a big controversy, especially when it's about whether a "this" is or isn't a "that", the first thing I assume is that people are trying to make a taxonomy out of a continuum. I believe AI is like that. Just as there's a continuum from replicating molecule to god-like alien, there's an analogous continuum from a NAND gate to strong AI.

Heck, why assume it's a continuum when it could be n-dimensional space. There's lots of stuff that can't be mapped to the real number line.

Totally agree. This article contains very little that's new.

It criticizes Watson, but Watson was roundly debunked last year as far behind IBM's marketing machine.

It trots out the "teenagers & sex" quote, which frankly is applied to every new technology. "Teenagers & sex" belongs to a very small family of cliches that everyone in tech has heard. The fact that it gets used indiscriminately makes it mean very little with each new application.

And finally, I'll like to point out the irony of someone trotting out tired thoughts to get attention while criticizing supposedly overhyped tech. To the right, we have people making exaggeratedly positive claims about tech to grab your eyes, and to the left, we have their mirror image.

> artificial intelligence (AI), a sub-branch of machine learning

I could have a failed mental model, but I'm under the impression that the relationship is the other way around. AI is a broad field encompassing various strategies to build intelligent machines. ML is one particular strategy where large volumes (think Big Data) of training data is used to teach by example. (Which makes Deep Learning a subset of ML, where "deep" neural networks are at play.)

It's actually neither way. AI and ML have some intersection. There're machine learning methods which have little to do with AI, for example, logistic regression. There're AI methods which have nothing to do with ML, for example, logical inference.

Dumb article. He talks about inflated expectations vs. real productivity, but does not give any evidence.

I was expecting some evidence and then I hit "I hope you liked this article."

The article mentions IBM trying to cash in on the AI hype with Watson, and towards the end wonders what the next overhyped trend will be. This ad might hold the answer: https://www.youtube.com/watch?v=2O2CLoCxAWA

This is not hype. See here: https://rajpurkar.github.io/SQuAD-explorer/

In this benchmark people or models are given a text, and later asked a number of questions. Questions are quite real. See for example here: https://rajpurkar.github.io/SQuAD-explorer/explore/1.1/dev/S...

Models already have performance which are as good as human's. This is real. This is not hype.

I agree! The question answering datasets, and some of the models built with them are getting good. I work on GAN and general classifiers at work, but on my own time my main interest is in sequential language models.

This time the hype is founded - Deep (Reinforcement) Learning truly pushed many AI domains out of uncanny valley. Image recognition/generation, speech recognition and synthesis, shallower language understanding - we truly have tech we have never seen before and only dreamed about. Of course, it won't solve everything, but the "solved level" got a massive upgrade with recent advancements. If we get GPUs that can compute DNNs 1000x faster, then we will see magic everywhere around us; so far the good models take very long to train, making them less adaptable to changing conditions.

The cute robot soccer game is a disingenuous argument. Put a machine gun on one of those Boston Dynamic robots with image recognition targeting and tell me it's not scary.

Well, yeah, 'cause you will never know when or where it's going to start spewing bullets. (I'll just note that a Tesla just ran into a parked fire truck.)

I stopped reading when it said AI is a sub-field of machine learning. It’s actually the other way around.

This is a really good article, I'm impressed. If you read only one deep learning article this month, make it this one.

I'm going to borrow that analogy of "teenagers perceptions of sex" - it's hilariously accurate for deep learning.

I fully agree that big data has essentially been renamed to AI. Which makes sense because AI today ~= ML/DL which require lots of data.

And, as someone who went through the data warehousing fad in the late 90s, there's a lot of naive belief in pouring in a lot of data and magic happening.

That said, there has been a lot of advance that, once it's happened, we just don't call it AI any longer. Route optimization (Google Maps), predictive analytics in some domains, image recognition. Yeah, a lot of it is just fuzzy pattern recognition but some of it is pretty good.

The more fundamental question IMO is how far DL can even take you. We've actually seen a lot of progress there but we also haven't seen a lot of forward motion in cognitive science for example. So do we just run out of steam in some of the areas, like autonomous vehicles, where we think we're doing pretty well today.

Big data is not AI. Big data is processing of a giant datasets with quite classic models, such as Logistic Regression in a distributed way. That's what frameworks such as Spark and Hadoop do. AI or deep learning is different. Usually, they don't distribute across many machines so well.

The teenage sex joke used to be a C++ joke: https://simplesassim.wordpress.com/2004/01/07/c-is-like-teen...

looks like the guy is trying to convince himself,

But machine learning is a sub-field of AI.

I really enjoyed this article. AI has dramatically overpromised every decade since 1960. Advances have always been made, but why do they have to exaggerate so much about its potential?

I’m too scared to comment or say anything else though. It would be the equivalent of saying something negative about bitcoin on r/Bitcoin.

Amara's law: "We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run."

Amara's law seems like a common human fallacy but it's also true that it's very hard to estimate correctly a growing trend. We have something that is obviously important and will have profound effect, but even small error in the estimates of technological proliferation can lead to 5-25 year time differences.

Another thing is the hype from outside the field. There are more people outside the field hyping it up than there are people inside hyping it. Investors, media and marketing are are powerful force when they jump in and they don't cool down easily.

> I’m too scared to comment or say anything else though. It would be the equivalent of saying something negative about bitcoin on r/Bitcoin.

I'm not sure where you got that impression but skepticism of AI hype gets a lot of airtime in HN threads.

Part of why we hype is because it’s how we get advancement. Yeah realistically the AI we all really want to see might be another three decades out, but if we want that to happen we have to get investors hyped about it now.

> Part of why we hype is because it’s how we get advancement

That's a partisan overstatement. It's how some other "we" gets funding, at best. Yes there's a correlation between getting funding and success, but to say the correlation between hype and success directly, seems disingenuous.

Reply on news.ycombinator.com