Hello and welcome to episode 72 of Linux downtime. I'm Joe and it's just me for the intro.
This is part two of a chat with Molly White. So if you missed the first part, definitely
go back and listen to episode 71 first. Before we get started, just a quick thank you to
everyone who support sales with PayPal and Patreon. We really do appreciate that. You
can go to linnoxdowntime.com or slash support for details. And remember that if you support
us on Patreon, there's various tiers to get advert free RSS feeds. And if you want to
get in contact with us, you can email show at linnoxdowntime.com. So without further
ado, let's get back to Gary, Amlith and me chatting with Molly White. So the Fediverse
mastered on that sort of thing. That seems to be so far delivering on the promise of true
decentralization without any of this blockchain and crypto nonsense. I kind of am a little
bit worried that it might get infected with it. But I'm really hopeful that it warrant.
What do you think's going to happen with that?
Well, I do think that there have been attempts by the crypto community, the crypto industry
to try to sort of subsume a lot of decentralized web projects into the web three umbrella,
which really irritates me because I really love the decentralized web, the Fediverse,
all these different things. And it feels like they're sort of taking credit for something
that they didn't do, you know, like a lot of the stuff really predates a lot of the
crypto stuff in a lot of ways. But so far, I haven't seen a ton of indication that crypto
is going to try to take over the Fediverse or infect the Fediverse. There doesn't seem
to be that much interest in it, which I think is maybe kind of telling in some ways around
the crypto world that despite all the talk of decentralization, you know, if there's not
a token involved, a lot of people in the crypto world just don't actually care that much.
So, you know, I feel like there are far fewer crypto people unmasked it on than there are on
Twitter, for example, a lot of these federated web communities, decentralized web communities
don't actually have that many blockchain people.
It does depend on what circles you're in, though. I've seen some instances completely
dedicated to cryptocurrencies. I've seen some projects that try to so deeply integrate cryptocurrencies
that you can pay while you're profile. So no one else can see your posts until they pay you in
whatever cryptocurrency that may be. And for the instances that have spun up around cryptocurrencies,
they have been very quickly defetterated from is one of the things that I've seen.
That's really nice. That's the bit with the Fediverse, right? You can just totally block a whole
instance, like with the far right ones, we've been able to do that and just have nothing to do
with them. And that's kind of part of the charm and promise of decentralization, right?
Yeah, that's another thing I was going to say is that like the idea of Mastodon, for example,
becoming totally infected with the crypto ethos is somewhat unlikely, I think, just in the sense
that people who don't like it can just choose a different server that doesn't integrate crypto
in any way, that doesn't federate with servers that are full of crypto people. And so it seems
likely to me that there will always be some portion of the Fediverse that has no real crypto
entanglements. And so far, it seems that it's been the majority rather than the minority. But
I guess who's to say what the future will hold there.
Can I ask you about central bank digital currencies, CBDCs? Is this something that you've
followed at all? It is, yeah. To me, it's ringing alarm bells, not necessarily the same alarm bells,
but this is centralized digital currencies that are controlled by governments, but
it still seems like a bad idea to me. Yeah, that's largely my view as well. It's that it has alarm
bells that are somewhat different from the cryptocurrency alarm bells, but still loud, and in some cases
louder, I think. A lot of the CBDC stuff really worries me because as soon as people start talking
about programmable money, they're talking about financial surveillance. And a lot of the financial
surveillance that I hear people talking about is really, really frightening to think about.
And I've been to blockchain conferences and things like that, where there are people who are
actually involved in the policy side of things or involved in the law enforcement side of things.
Who are talking about CBDCs as this very exciting promise of blockchains. And I'm used to going to
blockchain conferences where people are really excited about blockchains. But when I hear some of
these policy government people get excited about blockchains because of CBDCs, it's really startling
stuff because a lot of the time they're talking about how they can use this technology and this
sort of new monetary format to really control people. And so it's huge clacks ons blaring in my head,
I hear that kind of stuff because the idea of basically preventing people from sending money
to somebody through the technology itself or having enormous access to how people are
transferring their money. To some extent, that's in place today where the government has a lot of
access to your financial information. But ideally they still have to subpina it or gain access to
it through legal means. Yeah, and you can go to an ATM, get cash out and spend it totally privately,
whereas CBDCs are surely a step along the road to a cashless society and everything that entails.
And people who don't have access to bank accounts or whatever, there's a lot of people who live
day to day with cash. And I personally haven't used cash for a long time because it's just much more
convenient to use contact risk. But there are people who really need cash, right?
Right, yeah, exactly. And I remember I was at one of these conferences and there was someone who
was from a regulator, I had previously worked at a regulator, I think, who said something like,
oh, well, if cash was invented today, we would have never allowed it. I'm like, yes, but I feel
like that says something about our system today and not something about cash. But yeah, I think
that we really do need cash or some sort of digital analog to cash, I think would be amazing.
But I don't think that that is what people envision when they're talking about CBDCs,
they're not talking about digital cash, they're talking about a digital currency with a paper
trail that the government can access and potentially a very strong level of control around what you
can do with that money and who you can send it to and how much and how frequently and so forth.
Is that not the one slightly legitimate argument for cryptocurrencies though,
that they could be digital cash? I don't think they can. I think that true digital cash,
if it were to be possible and achieved, would have no real record behind it as far as where
transactions were coming and going from. And I think ideally it would have some of the qualities
of cash that allow it to be used today in ways that don't necessarily enable wide-scale money
laundering or criminal activity. Just the fact that it is physically challenging to move huge
amounts of cash from point A to point B is actually really useful. And that doesn't exist with
cryptocurrencies, which is why they've become so popular. Among the reasons why they've become
so popular for ransomware and various other forms of cybercrime. I think a digital analog
where you could actually have some cash-like quality of not necessarily surveilling people,
but having some reasonable limit on how much money could go from point A to point B would be
really exciting. There is a proposed policy to research and try to develop something like that
in the US called the Ecash Act, which I've been following with some interest. I have some questions
around the actual technical implementation of that, which I think is challenging, but that's
what I would hope to see out of any digital currency. I know that's much different than what
people view CBDCs to be. I still don't understand how you can even try to make a cash-like system
out of a cryptocurrency, right? Because a cryptocurrency has to always be auditable, right? You have to
account for every transaction all along the way, whereas cash is like you take, like Joe said earlier,
you take the money out of the ATM, you go spend it however you want, and at some point it makes
it back to a bank, right? So then there's your two records and cryptocurrency, you have a record
every step of the way. And so it's like, I mean, maybe somebody's got a novel way to solve that,
but I don't know, I don't see it. What about crypto tumblers? Is that the way to around that?
That's basically just a remailers back in the day with PGP, right? It sounds great on paper or
even tour, right? It sounds great on paper, but once you can, like over time, you're going to find
ways to break that and figure out where stuff goes. And like people can do that with Bitcoin to a
certain extent too. It's not anonymous. It's just there's no names tied to the transactions.
It's pseudonymous. Yeah, exactly. Yeah. And I also think that just, you know, as far as usability
requiring everyone to basically launder their own money just to have like a base level of privacy
is just not really feasible. Ideally, you can have something where the average Joe, you know,
could perform a transaction and enjoy some level of privacy. Whereas today, generally speaking,
the laypeople who are trying to use crypto are subjecting themselves to this enormous amount of
scrutiny because they aren't aware or they aren't capable of hiding their transactions.
Whereas, you know, people who do have more resources or more knowledge are able to
obfuscate things to some extent. And I think that's kind of a really unpleasant system, you know,
where it's sort of only the people who are very capable or very wealthy are able to enjoy even
very basic levels of privacy at this point. I'm curious. Do you all know what offline cash is?
Have you heard of it? Is that the paper Bitcoin thing? Yes.
Offline cash. They come full circle. Yes. I was hoping maybe it was actually like an interesting
project that someone to come up with. But no, I'm hoping it's a joke because if it's not,
well, that's the danger with these things. They start out as a joke like Dorridge. And then suddenly
it's actually worth, you know, at one point it was worth 50 plus cents or whatever, nearly a dollar,
I think. I may have mined a lot of a joke coin at one point in time myself.
Specifically, nigh and coin, if anybody's curious. And then the server just fell down and,
you know, it was fun, but never made any money off of it. But that's fine.
OK, this episode is sponsored by Servimania. Go to Servimania.com slash LDT to get 15%
off dedicated servers recurring for life. Servimania has over two decades of experience
building high performance infrastructure hosting platforms for businesses globally.
They offer a wide range of fully customizable dedicated cloud,
co-location and IP transit services and free initial consultations.
Servimania offers a 100% uptime SLA with some of the best bandwidth pricing in North America.
On network speeds of up to 20 gigabits per second in nine locations worldwide.
With Servimania, every customer receives a dedicated account manager,
free 24 seven live chat and support with one of the quickest response times in the industry.
So go to Servimania.com slash LDT to find out why my friend Alan Jude has been a
Servimania customer for over five years. Use the promo code Linux downtime to get 15% off
dedicated servers recurring for life. That's Servimania.com slash LDT and promo code Linux downtime.
The hype around blockchain and crypto and NFTs seems to have died down now and seems to have
been replaced with hype for something else. AI, machine learning and there just seem to be so
many parallels to me. That can't have escaped your attention, Molly.
Definitely not. It's pretty startling to see the exact same thing happen immediately after we
just got through this crypto hype and now we're sort of bombs away with the next one.
But yeah, I mean, I think it's something that we just see a lot of in the tech industry where a lot
of the innovation that happens in tech is bogged down with this crazy sort of marketing and
myth making around it that is really harmful, I think ultimately to any good that the technology
could potentially do because you have to end up separating the signal from the noise in a lot
of cases. And a lot of the time the noise really drowns out the signal, I think, which we're seeing
a lot of in AI right now. With crypto, there wasn't that much signal, so it was kind of easy to
differentiate. But you do think there is more signal with AI then? You think there are useful
implementations of it? Well, absolutely. I mean, we use it today already. We have been for years.
It's not like AI and machine learning just came to be this year. And it's a pretty low bar to say
that there's more use case there than crypto. But yeah, I mean, I do think there is some use case
there. But I think that a lot of what we're seeing now is how do we balance the positive potential
of something like AI or machine learning with the negative externalities of it? And with crypto
it's pretty easy to take a stance on that because there wasn't a whole lot in the positive potential.
And there was a lot in the negative externalities, whereas in AI and with machine learning, it's
variable definitely depending on what you're looking at. But there could be very strong
positives for it. But I think the negatives are also potentially
concerning. And I also think that the negatives that we should be worrying about are very different
from the ones that people tend to talk about because they're not the sort of Terminator
fantasies that a lot of people end up falling into. Yeah, well, that's the kind of easy story
to tell in the media. These things are going to take over. But if you have even the most basic
understanding of how they work, that's very unlikely. The idea of a general artificial
intelligence just seems to be just way, way off. But yeah, there are, I mean, already we're seeing
dangerous stuff with deep fakes. And so, you know, this is a red headline, this is going to be the
first deep fake presidential election, essentially. And what does it mean for democracy? And those
are the questions that people should be asking not when Skynet going to happen.
Right, I agree. And I think also questions just around labor when it comes to AI, you know,
I don't think there are necessarily as many concerns as a lot of the companies would like you to have
around OAI is going to take all of our jobs. But I do think there are concerns about the level to
which AI could sort of enable corporate surveillance or mistreatment of workers, make it difficult
for workers to obtain the same, you know, pay and rights that they should be getting. And so,
you know, I think there are a lot of those questions too that really need to be considered as we see
more employers start to integrate this stuff into their companies because the idea that
every job is going to be replaced by an AI is not particularly reasonable. But the idea that,
you know, your work life could become a lot less pleasant. I think this is pretty likely,
honestly. Now that you mentioned that, I'm realizing too how you could use machine learning to
basically turn your workplace into a surveillance state too, right? So like, you could just have
microphones monitoring keyboards and listening for a specific tone and you can figure out what
somebody's typing on their keyboard. The repercussions of that are just insane. But that seems to be
the path we're going down for better or worse. I think corporate surveillance has really become
an issue, you know, fairly recently where people are being surveilled by these, you know, so-called
productivity monitoring tools, their industries out there that are basically entirely surveilled,
like trucking, for example, is, you know, there's an enormous amount happening there. And it's all
in the name of efficiency and, you know, making sure that your workers aren't slacking off and
stuff like that. But it ends up really affecting the quality of life for these workers a lot of the
time. And it's difficult to fight back against a lot of that stuff, I think.
Well, yeah, the whole work from home change that happened with COVID that was supposed to
improve our lives. Or, you know, one of the benefits of it was the lack of community and
everything. But as you say, with these surveillance tools that a lot of companies are forcing people
to install on their computers, there would be better off being in the office probably than being
at home and being surveilled 24 seven. Yeah, and I think that, I mean, although some of these
surveillance tools don't necessarily require AI to be effective, the widespread rollout of them,
to some extent, does because, you know, if you don't have the sort of AI systems in place,
then you have to have someone watching all of this surveillance data to some extent. Whereas a lot
of the work that's happening recently, I think in machine learning and AI is around, you know,
okay, now we can broadly roll out the surveillance technology and surveil these employees, you know,
at a very large scale without having the same number of managers all watching, you know, every
little piece of their workday because the computers can just do it now. That's what really worries
me because, you know, surveillance technology isn't necessarily new, but the ways that that data can
be sort of munch and, you know, rolled out at scale is becoming more and more accessible.
One of the things that really scares me about the whole thing is that these large language models
cannot understand the way humans do. They don't understand what they're doing, they don't understand
the inputs or the outputs or anything like that. In January, an AI model called Carago
was beaten by an amateur. Go is, for those that don't know, Go is one of the oldest continuously
played board games in the world. It was like back in ancient China or something. And a few years ago,
an AI was able to beat the best human player. In January, the best AI model was beaten by an
amateur researcher. What they were doing was just like poking around at vulnerabilities in this model,
and they found that the model doesn't understand the concept of a group, which is one of the core
concepts of the game. So they just exploited that it can't understand that concept, and they
trivially beat it. I think it was greater than 90% win-lose rate. So they don't understand anything,
they can't be trusted to make decisions that a human should.
Yeah, and I think a lot of the risk also is that people ascribe a lot of responsibility and
sort of thinking to the models. And as soon as there's a mismatch between what something does and
what someone believes it does, I think you run into a lot of these problems where people start
expecting things of it that are really not reasonable. They expect them to produce truthful
statements, for example, not just statistically likely statements.
One of my concerns is creative work's being sucked up into these models,
and then being back out again. Not necessarily verbatim, but certainly in a way that could put
people out of work in the creative industry using work that they've done in the first place.
And that just doesn't seem fair to me. Yeah, I mean, I think the question around copyright and
creative rights when it comes to AI and a lot of this generative AI stuff that is just taking in
these huge sort of corpus of data and then spitting out sort of versions of them in some way,
I think there's a lot of unanswered questions there. And I think that the outcomes for creatives
could be really scary. And just the sort of impact that it has on people's rights as far as what
they post online, as someone who cares a lot about the open access movement and open knowledge,
it doesn't feel great that so much of my labor on Wikipedia, for example, is being used to power
all of these AI models. It's not like when I signed up to write Wikipedia articles, I was like,
yes, please train your models off my data. But because it's all published with an open license
and that type of thing, then they're able to do that. And to some extent, I think it doesn't
really matter if it's published with an open license or not, that a lot of these companies
are just taking all the data that they can, regardless of whether or not anyone's consented to it.
So I think as a society, there's sort of a reckoning that needs to happen around the data
that's being used, whether or not people have consented to its use, and what ownership means
when it comes to the actual products that are being spit out of these models.
You raised an interesting point there about consent for them being used. So like GitHub
co-pilot, for example, my code's not on GitHub for my own personal reasons, but people have put it
there. And Microsoft used the existence of code on GitHub to be that person giving consent for it
to be used in GitHub co-pilot. The open licenses allow that to be exploited. So like the companies
themselves have to do a better job of figuring out what that consent actually is, because what
they're doing right now is easily worked around. Well, the companies are not going to do it themselves,
right? We need some sort of legislation surely, of course.
I'm very conflicted about co-pilot, because I think the way co-pilot was trained is unethical
as a baseline. I don't think that can be argued. And I think with like GPL and AGPL, I think
copy left licenses as a whole, I think it likely violates those, but we won't know until that's
litigated in a court somewhere. That said, I use it every day. It's extremely useful and extremely
helpful. I hate to sound like one of those stupid productivity people who are always talking about
boosts and productivity, but it has genuinely helped. That's a hard conflict to get by. I disagree
with how this thing was created, but holy hell, the thing is useful.
Yeah, I mean, I think that's going to be a conflict that comes up a lot when it comes to this AI
stuff, just in general, is that a lot of AI tools are really useful. If you look at some of the machine
translation stuff that happens, for example, that is just extremely useful. A lot of it is a lot
better than the types of translations that were available before. But if the data is collected
unethically, if there is labor being put into this that is very unethically sourced and being
maintained unethically, then I think there still needs to be some solution to that.
I don't think that necessarily the means or the ends should justify the means in a lot of cases.
I also think that it is possible to ethically create useful AI tools. The solution should be
ensure that companies are doing so and providing enforcement and consequences when they aren't.
Well, it's been absolutely brilliant talking to you, Molly. If people want to find out more about
what you're doing, where should they go and find you? You can find me at mollywhite.net,
everything else links from there. That's the source of truth.
Okay, cool. Well, I'll put a link to that in the show notes then. I might put your
master done on there as well. You're pretty active on there seemingly.
Yeah, yeah, I love master done.
Yeah, cool. Right. Well, we'd better get out of here then. We'll be back in a couple of weeks,
but until then, I've been Joe.
Been Gary. I've been on the list. And I've been Molly. Say nice.
Yeah.
Yeah.
Yeah.
♪♪♪♪