Looking forward to our brave new world of Artificial Intelligence…oh, wait

Oh dear. There’s a language processing module called GPT-3 which is really good at generating natural English text, and it was coupled to an experimental medical diagnostic program. You might be able to guess where this is going.

LeCun cites a recent experiment by the medical AI firm NABLA, which found that GPT-3 is woefully inadequate for use in a healthcare setting because writing coherent sentences isn’t the same as being able to reason or understand what it’s saying.

…

After testing it in a variety of medical scenarios, NABLA found that there’s a huge difference between GPT-3 being able to form coherent sentences and actually being useful.

For example…

Um, yikes?

Human language is really hard and messy, and medicine is also extremely complicated, and maybe AI isn’t quite ready for something with the multiplicative difficulty of trying to combine the two.

How about something much simpler? Like steering a robot car around a flat oval track? Sure, that sounds easy.

Now I’m scared of both robotic telemedicine and driverless cars.

Or maybe I should just combine and simplify and be terrified of software engineers.

Linux goal for the day: iMovie replacement!

I tried KDENLive. It was too much — so many quirks and clumsiness, perhaps because it was just trying to pour every single video editing option possible in willy-nilly.

Today, I’m experimenting with Shotcut, on the recommendation of a reader. It’s simpler, so not quite as overwhelming. I might be able to work with this.

A couple of common problems I run into everything Linux. With no universal interface guidelines, every program seems to want to do everything their own way. I appreciate how the programmers can find that liberating, but in the absence of constraints and standards, they always seem to make bad choices, and you have to just stare in wonder at how they’ve decided to arbitrarily fuck up their own work. It also means that using the thing is awkward and not at all fluid, at least not until the user develops their own novel workflow. It also means the user has the power to make the most godawful ugly videos ever — I was testing various things out and made this nightmarish thing with purple 3-D titles and funky video effects and random stuff appearing in spectacularly elaborate ways that will never be seen by the eyes of any other human. I’m about to take it ’round back, shoot it, throw it in a hole, and set it on fire. But I figure out how some things work while building that monstrosity!

Bottom line: Shotcut might be my replacement for iMovie, as ugly as it is. I might just have to become insensitive to non-Mac ugliness. I know, I shouldn’t complain, it’s free software…but iMovie is also free.

First try:

Also, I’ll be on this channel at 3pm tomorrow (9 June) for a free-for-all rant about Kent Hovind and other creationists.

Yes. I need more limbs.

When we were talking to my daughter the other evening, she was struggling to manage a phone in one hand and a busy baby in the other, and I told her she just needed a third. Surprise! Science provides with a wearable robotic “third arm” that can punch through a wall. The “punch through a wall” feature seems particularly useful in the context of child care. Except, I’m sorry, the video makes it less than useful.

Nice gadget, but it requires one person to wear it, and a second person with two arms to control it remotely, effectively requiring four arms to enable three-armed activity, in which one of the three arms is rather clumsy. It’s going to require a better control mechanism, something with a neurological link to the wearer. As long as we’re doing that, why stop at three? Why not…eight? I am ready for my robotic exoskeleton that will let me climb walls and punch through walls and destroy walls any day now.

Oh, and do more efficient childcare, I guess.

More serious analysis of online conferencing, please. It’s our future.

You’d think that after a long period of isolation I’d be looking forward to a return to normality and the opportunity to mingle with others at a conference again. Strangely, I am not. There are great things about real-life conferences, but also great awkwardnesses. I stopped attending the annual Society for Neuroscience conference many years ago as attendance soared past 20,000, which made the face-to-face appeal diminish as we became a faceless mob…and also as it became obvious that a subject as complex and diverse as neuroscience couldn’t be appropriately managed in a one-size-fits-all event.

So I thought this article about organizing scientific conferences online was somewhat informative. I’ve been involved in running a social justice conference online, once upon a time, so I’m familiar with some of the compromises, but it’s good to see some new ideas. Zoom has all kinds of potential, and they used Crowdcast, but I thought the way they applied it was a good mix of traditional and novel uses.

Even the traditional elements were improved.

We largely retained the legacy conference format of a single track for invited talks (30 minutes plus 15 minutes for questions) from established scientists, and contributed talks (18 minutes plus 4 minutes for questions) selected from the submitted abstracts to highlight work from up-and-coming researchers. However, the online platform used – Crowdcast – allowed for some significant innovations. First, everyone was able to see the speaker more clearly than in a lecture theatre. Second, Crowdcast allows anyone to submit a question to ask the speaker at the end, and viewers can vote on those questions. This led to a question and answer session that was considerably more lively and democratic than in a typical legacy conference, where participants often note that the same established professors are asking the same questions at every talk. As in the case of the short talks, it may be better to extend the questions even more to capitalize on the quality of the questions asked in the safer and more democratic online format. The third innovation is the chat window that appears alongside the talk. We did not anticipate how significant this would be. Students and others were able to ask basic questions about definitions or ask for links to papers while the talk was going on. Other participants could answer them in real-time without disrupting the presentation, thereby allowing a deeper level of engagement by the audience than is possible in legacy conferences. Moreover, since recordings of these talk were available immediately after the session, it would be possible to go back and revisit portions of the talk that may have been missed or were presented too quickly.

Right, you’ve got to keep the talks limited to familiar blocks of time. We do a lot of training and practice to maximize information in small specified chunks of time. They didn’t do one hour talks, though? I’m used to conferences with plenary sessions with hour-long time slots for bigwigs in the field…they usually don’t live up to their billing, though. An hour is a long time to fill.

The other thing you need for a conference is the schmoozing. They had a way of doing that that seemed to me to be trying too hard. Typical nerds.

One feature of a legacy conference that would appear to be impossible to replicate online is the social aspect: chance encounters during the coffee breaks, social events or banquets. In place of this aspect, neuromatch algorithmically matched attendees to other like-minded scientists for individual 15-minute chats. We use a combination of topic modeling techniques and linear programming to solve the matching problem based on a sample of their research abstracts (Achakulvisut et al., 2018). The matching part was based on a highly popular experiment carried out at the Conference on Cognitive Computational Neuroscience, but it is particularly well-suited to an online format. There remains considerable scope for further innovations in replicating or improving on the social experience of legacy conferences, especially as the online format may be less socially intimidating.

Wouldn’t it be easier to just randomly put people into small break-out groups? This sounds too planned — one of the benefits ought to be serendipitous encounters using a simple algorithm that assumes every participant is equally interesting with unique attributes that anyone might find productive. One useful parameter I wanted to know is what is the optimal group size for these chats. Was it one-on-one? Half a dozen in a group? Small classroom size with 30 participants?

I suspect that one good thing that will emerge from this pandemic is more online conferences. It vastly reduces the expense, gets rid of the bother of air travel, and helps participants manage their time better. I currently try to attend one conference a year because it’s such a huge investment of time and effort — and this year the one I’d planned on got cancelled, of course. But if they were online, I’d be able to schedule that arachnology conference, the Society for Developmental Biology annual meeting, and SICB every year without killing myself with constant travel to the airport, while still learning new things and engaging with new people. Make it so, scientific societies!

Also, if you’re interested, Skepticon is going online this year. I’ve long wanted to attend Dragon*Con, but it’s been impossible because it always falls during the first few weeks of classes, when I can’t possibly just take off. Of course, if it went online, what would happen to all the cosplay events? And now I want to attend even less, because it’s held in a state where Brian Kemp is the Republican governor who is mismanaging the pandemic, and we’ll probably see another wave about the time Dragon*Con opens for business as a big bustling petri dish. Many Shubs and Zulls will know what it is to be roasted in the depths of a Sloar that day, I can tell you!

I wonder if they’re even considering alternatives — I get the idea that our regional SF con, Convergence, is going ahead with the idea that they’ll be doing business as usual in late August, while developing contingency plans. I’m not so confident. Getting crammed into a single building with thousands of other attendees, many of whom need to be reminded about the basics of hygiene, seems to regularly lead to icky cases of con-crud. Only this con-crud can kill you!

P.S. I’ve only just noticed that searching for “online conferences” produces strange images of people sitting around a conference table staring at a screen on the wall, or the always-popular image of two people shaking hands through a pair of computer screens. I really don’t think they get it. That’s not how it’s going to work, or can work. And shaking hands? Is anyone else feeling repulsed at the idea of physical contact with some stranger’s filthy hands?

News from the tech world

No drama here. We all know that science and technology is the domain of cool heads, objective thought, and rational personalities, so this post is going to be a bit boring.

First up: the owner and founder of 8chan are locked in a bitter battle. The founder, Frederick Brennan, called the owner, Jim Watkins, some mean names, and he then replied in the manner of a true defender of Free Speech by siccing the government of the Philippines on him with an arrest warrant, threatening to put Brennan in prison for 10 years.

Watkins is pursuing Brennan under the Philippines Cybercrime Prevention Act. The law was enacted in 2012 despite outcry from rights groups and the United Nations, which warned that it was not in line with international norms. Amnesty International said that it “rolls back protection for free speech” and would have a “chilling effect” on speech in the country. It has most notably been used against journalist Maria Ressa in what the Committee to Protect Journalists labeled a “campaign of harassment.”

Yay Free Speech!

In other news…Twitter. That bastion of reasonable debate and discussion has gradually evolved to become an important arm of the government. Like Alexander Hamilton had the Federalist Papers, our current president has been calmly making arguments for a new way of communicating with the people via Twitter, and rich people have noticed. A billionaire, Paul Singer, is buying a sizable stake in the Twitter corporation, so that he can use his wealth to push Jack Dorsey around. The fact that he is a billionaire clearly informs us that he is competent and wise, so this can only improve the site. Money gives you the right to rule any aspect of the world you want to change.

Singer has even taken on whole countries: in 2016, after a relentless campaign, he secured a partial repayment of debts by Argentina, arising from its financial collapse in the early 2000s.

Maybe he’s going to give Twitter to The Donald as a Christmas present?

Speaking of Twitter, you know it has a policy of supporting anonymity, for good or ill, and banks away a lot of confidential information which you’re only supposed to see if you’re a wealthy capitalist who wants to advertise on the site. It turns out there’s another way: worm spies into the organization who will compromise security and sell information directly to their clients, like Saudi Arabia.

A week after returning to Twitter’s San Francisco headquarters, Abouammo logged into the system he used to verify users, according to the complaint. That system, sources who’ve accessed it told BuzzFeed News, stores information including email addresses, telephone numbers, and last log-in time — sufficient personal data to track down a user in real life.

Accessing two Saudi dissidents’ information — one a prominent critic with more than 1 million followers, the other an impersonator of a Saudi Royal family member — Abouammo allegedly passed the information to al-Asaker. Twitter had long been a godsend to dissidents: Unlike Facebook, it had no policy requiring people to use their real names, allowing critics of repressive government to speak more freely. The allegations threw its value as a tool of anonymous dissent into question.

Hey! No fair! You’re supposed to pay Twitter for those privacy breeches, not some random employee! You know I think Saudi Arabia could afford to pay through appropriate corporate channels and grease the right palms here.

There. I’ve had enough of the great minds behind our technological enterprises, I think I’ll sit here and just drink coffee for a while.

Are you boycotting YouTube this week?

YouTube has threatened to drop any channels that are not “commercially viable” in the future, which is grossly capitalist and of concern to everyone, as capitalism destroys another medium. I guess I should worry; I have a YouTube channel, and last I looked it earns me about $15 a month, which doesn’t sound particularly viable to me (I’m not in it for the loot, fortunately). Meanwhile, the channels that earn big bucks and make their owners thousands and millions of dollars are total garbage. Do we really want to see the PewDiePieification of YouTube?

Anyway, there’s a call to walkout, you can read about it at Great American Satan and the Bolingbrook Babbler and stderr and Impossible Me and Intransitive. I’m pessimistic. I don’t think it can generate enough press or enough pressure to make any difference.

But we need to take a stand somewhere. This is an announcement of a policy to stomp down small content producers, killing niches which we already know will be selected against if they have the faintest whiff of sexuality, while the racist channels will be tolerated. They want to propagate the least desirable aspects of American culture and resist anything that might change it. I’m in. I haven’t even glanced at YouTube since the day before yesterday.

I want you to know if this experiment actually worked, three days of silence would cost me a whole $1.50. So I’m participating in this effort at great personal sacrifice.

Is Apple trying to drive me away?

My latest aggravation with the company: I upgraded to their newest version of the OS, Catalina, recently, and it looks and performs just fine, except for a couple of gigantic problems. Both Audacity, the great audio processing program, and OBS Studio, the most popular broadcasting software around, crash and die if you try to run them. It’s a conflict with how the OS manages permissions to access things like microphones and cameras. There a couple of ugly workarounds floating around that involve launching the programs from the Unix command line, but still, this is appalling.

Apple should have known about these conflicts long before the OS was released, and if they couldn’t fix them on their side, they should have flown a couple of engineers straight to those companies and helped them fix their code. It’s not as if Apple has a shortage of cash. I can understand how they might find it necessary to move forward, especially on issues that involve security, and break backwards compatibility…but then they are responsible for making the information available to enable apps to conform to the new standards.

Oh, yeah, and the camera software for my microscope is also broken. That’s a very narrow niche so I don’t expect a huge effort by Apple to make it work, but still, I’m sitting here feeling like my computer has been crippled in some of the functions I use most.

Well, that bad sci-fi vehicle might have a weakness

The bad news: in the future, rich people are dreaming of driving around in fugly brutalist armored trucks from Tesla, lording it over the masses in their slabs of cold gray metal.

The good news: they are easily cracked open to reach the soft, pampered sweet-meats inside.

Bad demo, Elon, bad. That’s not going to light up the eyes of the wealthy people who can afford to buy your dystopian style.

Although, to be fair, the 500 mile range is a very appealing feature. Our standard long drive is about 150 miles to the Twin Cities, and most electric cars would get us there, but not back.

What if Snow Crash was actually a documentary?

The novel Snow Crash analogized human minds to computer operating systems and suggested that they could be just as susceptible to bad code, like a mind virus. There’s a lot to like about the idea, but the book takes it very literally and has people’s brains being wiped and taken over by a mere brief exposure to a potent meme…which is ridiculous, isn’t it?

Maybe it would take repeated exposures to do that.

We’re doing the experiment right now. Facebook has these “content moderators”, a job farmed out offsite to groups of people who are required to view hours of atrocious content on a tightly regimented schedule built on the call center model. They don’t get to escape. Someone posts a video of someone being murdered, or of a naked breast, and they have to watch it and make a call on whether it is acceptable or not, no breaks allowed. No, that’s not quite right: they get 9 minutes of “wellness” time — they have to clock in and clock out — in which they can go vomit in a trash can or weep. It sounds like a terrible job for $29,000/year. And it’s having lasting effects: PTSD and weird psychological shifts.

The moderators told me it’s a place where the conspiracy videos and memes that they see each day gradually lead them to embrace fringe views. One auditor walks the floor promoting the idea that the Earth is flat. A former employee told me he has begun to question certain aspects of the Holocaust. Another former employee, who told me he has mapped every escape route out of his house and sleeps with a gun at his side, said: “I no longer believe 9/11 was a terrorist attack.”

Maybe Clockwork Orange was also a documentary.

It’s not all horrifying. Most of the work involves petty and mundane complaints from people who just don’t like what other people are saying, a domain where the principle of free speech applies. The company, other than the routine fact that it’s run by micromanaging assholes, is above average in how it treats its workers (which tells you what kinds of horrors are thriving under capitalism everywhere else, of course).

Everyone I meet at the site expresses great care for the employees, and appears to be doing their best for them, within the context of the system they have all been plugged into. Facebook takes pride in the fact that it pays contractors at least 20 percent above minimum wage at all of its content review sites, provides full healthcare benefits, and offers mental health resources that far exceed that of the larger call center industry.

And yet the more moderators I spoke with, the more I came to doubt the use of the call center model for content moderation. This model has long been standard across big tech companies — it’s also used by Twitter and Google, and therefore YouTube. Beyond cost savings, the benefit of outsourcing is that it allows tech companies to rapidly expand their services into new markets and languages. But it also entrusts essential questions of speech and safety to people who are paid as if they were handling customer service calls for Best Buy.

I think part of the problem is that we treat every incident as just another trivial conversational transaction, yet that is the least worrisome aspect of social media. There are obsessives who engage in constant harassment, and this approach just looks at it instance by instance, which means the obsessive simply has to escalate to try and get through. It ignores the possible of planned maliciousness, where organizations use the tools of propaganda and psychological manipulation to spread damaging ideas. You check one of their memes, they simply reroute around that one and try other probes with exactly the same intent. No one can stop and say, “Hey, this is coming from a bot farm, shut it down at the source” or “This guy is getting increasingly vicious toward this girl — kill his account, and make sure he doesn’t get another one”. It’s all about popping zits rather than treating the condition.

As long as Facebook and Twitter and Google persist on pretending this is a superficial symptom rather than a serious intrinsic problem with their model of “community”, this is a problem that will not go away.