Past Performance is Not Indicative of Future Results (2020) • Hacker News Board

Re: Past Performance is Not Indicative of Future Results (2020)

#31

Post by MAXPOOL » Sat, Jul 31, 2021, 4:43 PM UTC

For a short and very non-technical article, this is well written.

The current approach to machine learning is not going to go towards general-purpose AI with steady steps and gradual innovations. Things like GPT-3 seem amazingly general at first. But even it will quickly plateau towards the point where you need a bigger and bigger model, more and more data, and training for smaller and smaller gain.

There need to be several breakthroughs similar to the original Deep Learning breakthrough away from statistical learning. I would say it's 4-7 Turing awards away at a minimum. Some expect less, some more.

Re: Past Performance is Not Indicative of Future Results (2020)

#32

Post by RandomLensman » Sat, Jul 31, 2021, 4:43 PM UTC

jon_richards wrote: post #18

iamnotwhoiam wrote: post #8

Are there any approaches to artificial intelligence that do involve qualitative data or don’t rely entirely on statistical inference?

Does qualitative data actually exist? Named colors are considered qualitative, but rbg and cmyk are quantitative. Does converting from one to the other switch whether it is qualitative or quantitative? Surely semantic meaning is qualitative, but look at word replacement in Google search. That’s entirely based on statistics, thesaurus graphs, and other ultimately quantitative data. The neat thing about neural nets is…

Kind of does exist even with colors: try to map "brown" into an RGB or CMYK data point.

I think the real difference is that in qualitative data the numerical representation does not mean anything. Sure, the names of the archangels can be represented digitally (quantitative) but that is just a change of representation - the bit strings' numerical value carries no theological meaning.

Re: Past Performance is Not Indicative of Future Results (2020)

#33

Post by shreyshnaccount » Sat, Jul 31, 2021, 4:43 PM UTC

jstx1 wrote: post #16

> I am an AI skeptic. I am baffled by anyone who isn’t. I don’t see any path from continuous improvements to the (admittedly impressive) ‘machine learning’ field that leads to a general AI - I share the skepticism towards any progress towards 'general AI' - I don't think that we're remotely close or even on the right path in any way. - That doesn't make me a skeptic towards the current state of machine learning thoug…

I'm in favor of changing the terminology from AI and ML to something along the lines of 'prediction model' so that the idea of machines 'thinking' is replaced with them 'predicting'. it's just easier for our mushy meat brains to think that AI and ML means that it'll lead to general AI or as I like to call it 'general purpose decision maker'. it's all about the language!

Re: Past Performance is Not Indicative of Future Results (2020)

#34

Post by marcinzm » Sat, Jul 31, 2021, 4:43 PM UTC

itisit wrote: post #27

version_five wrote: post #23

Earlier quoted context omitted.

You don't think a person has ever called the police because they hear a noise they thought was an intruder, or saw someone or something suspicious only in their mind? People make these kind of mistakes too.

Of course, but the consistency of the false positive is the issue. An able-minded person can readily reconcile their confusion.

Then that’s a question of training data.

Re: Past Performance is Not Indicative of Future Results (2020)

#35

Post by karaterobot » Sat, Jul 31, 2021, 4:50 PM UTC

version_five wrote: post #9

This article is mostly a straw man, while still containing some valid ML criticism. I am a ML s(c|k)eptic too, in that popular conceptions of what ML is currently overpromise, often don't even understand what ML actually is, and are often just some layperson's imagination about what "artificial intelligence" might do. This article is the opposite. He's treating ML as basically a simple supervised architecture that do…

> This article is mostly a straw man, while still containing some valid ML criticism.

I don't think this is an example of a straw man, given that his audience is readers of Locus, a science fiction magazine. While researchers and practitioners in ML understandably hold a more nuanced, informed view, the position he's arguing against is pretty common among the general public, and certainly common in science fiction.

Re: Past Performance is Not Indicative of Future Results (2020)

#36

Post by jon_richards » Sat, Jul 31, 2021, 4:51 PM UTC

RandomLensman wrote: post #32

jon_richards wrote: post #18

Earlier quoted context omitted.

Does qualitative data actually exist? Named colors are considered qualitative, but rbg and cmyk are quantitative. Does converting from one to the other switch whether it is qualitative or quantitative? Surely semantic meaning is qualitative, but look at word replacement in Google search. That’s entirely based on statistics, thesaurus graphs, and other ultimately quantitative data. The neat thing about neural nets is…

Kind of does exist even with colors: try to map "brown" into an RGB or CMYK data point. I think the real difference is that in qualitative data the numerical representation does not mean anything. Sure, the names of the archangels can be represented digitally (quantitative) but that is just a change of representation - the bit strings' numerical value carries no theological meaning.

Brown is (165,42,42). You can argue about false precision, but the term “brown” has false precision as well. The likely variation in interpretations can be described by error bars. Your understanding of someone saying “brown” is informed entirely by statistical inference of your past experience with “brown”.

Changing the representation of the names doesn’t matter, but attempting to understand the meaning behind the names is ultimately quantitative. The numbers are run in the giant black box that is your brain and then your consciousness receives other qualitative answers.

Asking for an AI without statistical inference or quantitative data is asking for consciousness without a brain.

Re: Past Performance is Not Indicative of Future Results (2020)

#37

Post by okareaman » Sat, Jul 31, 2021, 4:51 PM UTC

> But the idea that if we just get better at statistical inference, consciousness will fall out of it is wishful thinking. It’s a premise for an SF novel, not a plan for the future.

My impression of Silicon Valley types like Ray Kurzweil in "The Age of Spiritual Machines" that if we wire up enough transistors somehow consciousness will somehow arise out of the material world. The somehow is not explained. Materialism is a dead end in my opinion. I am more interested in theories about consciousness as a field and our brains as receivers.

Re: Past Performance is Not Indicative of Future Results (2020)

#38

Post by darkwater » Sat, Jul 31, 2021, 4:52 PM UTC

jstx1 wrote: post #16

> I am an AI skeptic. I am baffled by anyone who isn’t. I don’t see any path from continuous improvements to the (admittedly impressive) ‘machine learning’ field that leads to a general AI - I share the skepticism towards any progress towards 'general AI' - I don't think that we're remotely close or even on the right path in any way. - That doesn't make me a skeptic towards the current state of machine learning thoug…

> I think it's important to make this distinction and for some reason it's left implicit or it's purposefully omitted from the article

I beg to disagree. They clearly state your opinion at the end of the piece, using the metal-beat analogy. Great things were done by blacksmiths beating metal, but not an ICE

Re: Past Performance is Not Indicative of Future Results (2020)

#39

Post by mirekrusin » Sat, Jul 31, 2021, 4:53 PM UTC

shannifin wrote: post #30

> I don’t see any path from continuous improvements to the (admittedly impressive) ‘machine learning’ field that leads to a general AI any more than I can see a path from continuous improvements in horse-breeding that leads to an internal combustion engine. While I also don't expect that AGI will emerge solely through optimizing statistical inference models, I also don't think "improvements to the machine learning fi…

It gets philosophical quickly, is "consciousness" repeatedly modifying cloud of random floats?

Re: Past Performance is Not Indicative of Future Results (2020)

#40

Post by version_five » Sat, Jul 31, 2021, 4:53 PM UTC

itisit wrote: post #27

version_five wrote: post #23

Earlier quoted context omitted.

You don't think a person has ever called the police because they hear a noise they thought was an intruder, or saw someone or something suspicious only in their mind? People make these kind of mistakes too.

Of course, but the consistency of the false positive is the issue. An able-minded person can readily reconcile their confusion.

An ML system generally can reconcile (and also avoid) this kind of confusion, with present technology. The example is more a question of responsible implementation than of a gap in the state of the art.