Artificial Intelligence Is Shattering Our Framework: A Talk With Aleksandra Przegalińska

aleksandra_przegalinska.jpg, default, Aleksandra Przegalińska, photo: Niemirowski Studio, center, #000000

In the future, will we read books written by robots? Aleksandra Przegalińska, a philosopher, futurologist, and researcher in the field of artificial intelligence, discusses the ways in which AI is used in culture and the development of machine-based creativity.

Aleksandra Przegalińska: Artificial intelligence is shattering the framework in which we’ve operated so far.

Patryk Zakrzewski: It’s often said that the decades-long history of research into artificial intelligence is divided into winters and summers: periods of increased interest in this topic and stagnation. Right now, we’re back in one of those intense periods, with reports of new developments coming in all the time. What are the groundbreaking developments that have happened in this field over the past year?

Aleksandra Przegalińska: This idea of winters and summers is a bit problematic because voices keep emerging thatwhich contest this sharp divide that suggests there have only been, so far, periods of decline and flourishment in the work on artificial intelligence. It has rather been a much more complicated and fluid story. Of course, the term ‘AI winter’ is still in circulation and still frightens people. On the other hand, there are opinions, for instance from the well-known robotics scientist Ben Goertzel, that there has never been such a period. There have been periods when there was less funding and less general interest in AI, but that doesn’t mean innovative projects didn’t emerge during those times. Some very good expert systems were created during one ‘winter’, and some groundbreaking robotics solutions during another. Others postulate that we should add ‘spring’ to the equation. MIT researchers, for example, argue that we’re currently experiencing the season before the full blossoming of AI capabilities.

In any case, this warm season in the development of artificial intelligence has lasted for a long time, and it’s not like the year 2020 was an important boundary line in this field. The most important, groundbreaking things happened in the 2000s – especially in 2009 and 2010. That’s when deep neural networks were developed. But we’re still discovering their capabilities: processing images, photos, and numerical data. Year after year, the number of new applications increases. This is the breakthrough we’re working on all the time: the emergence of brain-inspired neural networks and the emergence of methods such as learning by reinforcement, used in the most recent networks which learn by gaining experience in a manner similar to humans.

The past year has certainly been interesting because of the development of the medtech sector, which is the application of artificial intelligence to study vaccines, find pharmacological therapies, and track the development of pandemics. Then there are all the telemedical solutions, such as chatbots used in healthcare. I think an important part of this was that artificial intelligence entered a sphere that isn’t controversial. The capabilities of artificial

intelligence here are much more useful to people than when it’s used in the banking sector or marketing. This has led many people to understand that artificial intelligence can actually be helpful. This is evident from the fact that medtech start-ups have been enjoying success for some time now, whereas previously they tried knocking on investors’ doors to no avail.

If I had to name some breakthroughs from the past year, they would be two more projects that seem extremely interesting, although we don’t quite know the implications yet. One, of course, is GPT-3, a text generator that’s based on deep neural network solutions. It does excellent text processing and wrote a very interesting article for The Guardian. It has many interesting potential applications, and in 2021 we will surely learn about more of them. The second piece of recent news is about the creation of MuZero, a neural network created by DeepMind, which can play games of all kinds (including chess, Go, and Shogi) and can also transfer skills gained in one game to another.

Video embed

Do You Love Me?

PZ: The article you mentioned from The Guardian was titled: ‘A robot wrote this entire article. Are you scared yet, human?’ The feeling of unease triggered by contact with artificial intelligence has been called an ‘valley of strangenessuncanny valley’, although the term originally applied primarily to robots that were too similar to humans. What are we more afraid of in AI today: its appearance or its capabilities?

AP: People are afraid of various things in relation to AI, and it’s hard to unequivocally classify these fears. I don’t know if you’ve seen the Christmas video of dancing robots made by the firm Boston Dynamics? They look adorable, but a lot of people are afraid of humanoid robots like that. We have this evolutionary fear in us of something that can’t be classified: it’s impossible to tell if it’s alive or not, it resembles an animal, and it behaves strangely.

Humans are also afraid of the competencies demonstrated by artificial intelligence, although they desire them at the same time. When a system such as GPT-3 writes an article for The Guardian, on the one hand you hear people exclaiming, ‘Wow!’;, but on the other hand there are questions such as, ‘Okay, but what else can it do? Will it take my job?’ So, there are still such fears, not as deep and primal, but much more rational. Every innovation like this triggers anxiety – when a system is able to do something on a human level and we can’t figure out whom we’re dealing with.

The creators try to circumvent the first fear by constructing very few humanoid robots. On the other hand, neural networks are able to generate images of fictional people, and it turns out that we aren’t afraid when we don’t know we have a non-real person in front of us. So, it’ll be interesting to see in which direction this will develop – perhaps human-like robots will be developed in the virtual realm, while being abandoned in the physical realm.

In terms of the competence or expertise of artificial intelligence, we don’t seem to be going backwards. Every now and then, there’s news of improved systems. While people certainly fear them and they do bring about some change, when we observe their capabilities, the change doesn’t seem as frightening or dramatic as it did at first. However, these are narrowly specialized projects created to do specific things, and it’ is people who tell them what to do. These fears are then confronted with reality, which shows that there’s a long way to go before such a GPT-3 will be writing articles instead of a journalist.

PZ: In the 1970s, the Polish sci-fi writer Stanisław Lem wrote about literature created by robots in his book ‘Apokryfy’ (Apocrypha). He called it ‘bit mimesis’: ‘[…] Mimesis has led to unemployment – but only among those who produce trivial literature (sci-fi, porn, thrillers, etc.); there, in fact, it has forced people out of the supply of intellectual commodities; this phenomenon, however, shouldn’t cause an upright humanist to despair too intensely.’. How do you assess this prediction from the perspective of what we know today? I think we’ll have to wait a while for artificial intelligence to function independently on the publishing market, but I can imagine a large number of AI ghostwriters in the future, with human authors playing mainly promotional roles.

AP: As always, Lem is on the mark. Besides, the GPT-3 Fiction project already exists. The creator plays with this system by generating portions of prose and poetry in the style of Edgar Allan Poe or Sylvia Plath. These writing samples aren’t bad, but there’s a certain craftsmanship. I think artificial intelligence will be used to generate advertising jingles, simple and functional compositions, desktop graphics and advertising leads. Of course, it will have to be ‘fed’ the right content, and someone will have to supervise it.

On my Coś osobliwego podcast Coś osobliwego, I recently hosted Janek Simon, a visual artist who uses new technologies in his work, and we talked about whether neural networks can function successfully as visual artists. As someone who also knows programming, he said it’s a great tool, but that it hasn’t demonstrated anything very interesting so far. It’s always the glitchy, processed images that we’re familiar with. They impress us for five minutes, but it has nothing to do with deep creative work, and I don’t think it will anytime soon. It lacks agency, authorship, and intention – all those things we’re unable to recreate artificially.

I think Lem was right in saying that artificial intelligence will be useful for simple forms of creative work, which of course raises a huge question from the perspective of the labour market. I can’t imagine AIs exhibiting their work in an art gallery, or writing an interesting novel that we won’t read merely as a curiosity: ‘Oh, how funny, it was written by a robot.’

PZ: Bleak predictions even mention the emergence of a class of ‘expendable people’ in the labour market, who will be replaced by robots. Or perhaps, looking at the situation more optimistically, people will simply be engaged in more creative pursuits, and artificial intelligence will be responsible for tedious work.

AP: Exactly, maybe humans will do more interesting things with these tools. I wouldn’t rule that out. Collaboration between a human and a machine, if only in the realm of visual arts, might produce a new quality or give the artist a break. The artisanal part will be left to the AI, which will, for example, fill spaces with colour at the human’s command.

I’ve always wondered what happens to these unleashed resources… So far, people have shown that they’re able to use them effectively. I’m currently reading a children’s book with my daughter called A History of Women and there’s a nice description in it of the emergence of household appliances – the iron, the washing machine, and all the rest – from the perspective of freeing up housewives’ time. The book describes what happened afterwards: how women’s aspirations and ideas about their lives changed because of it, and how it affected their occupational mobility. And it’s extremely interesting that people are benefiting from these innovations not by sitting on the couch and doing nothing – at least this is my thesis, perhaps idealistic, though I think it’s supported by facts – but by advancing to the next stage of this development process.

PZ: A GPT-3 sometimes creates its own neologisms. There was also the famous case of Facebook bots that developed their own language, incomprehensible to humans, and started communicating with each other in it. Do you think phrases created by machines could pass into our language over time? And, more broadly, how can interacting with artificial intelligence affect our behaviour?

AP: A famous study done in the USA comes to mind here, which found that children who had Alexa, Amazon’s electronic assistant, at home used its specific language to interact with it. The amazing thing, however, was that they started addressing other children and their mothers in a similar way. I read on Twitter that a little boy who was a year and a half old wanted to hear his favourite song, and so he whispered to his mother, ‘Alexa, play Christmas song,’ because he’d learned that she requested music this way.

So, there’s something interesting going on here, but I don’t know to what extent it will change our language, other than perhaps providing us with anecdotes or inspiring us to humorously talk to each other like AI systems. I think an AI would have to create something amazing, some sort of Ulysses of its own, for its words to enter everyday language.

PZ: And looking at it from the perspective of more traditional human categories of authorship… Let’s assume that language patterns will work much better in the future than they do now, and will also be able to handle Polish. Someone suggests ‘Trilogy’ to an AI, and it produces an excellent sequel, closely resembling Sienkiewicz’s prose. Who can we consider the work’s author: the creators of the system, the AI itself, or Henryk Sienkiewicz, who developed the language and style and invented the characters? Or perhaps the person who came up with the idea of putting this book into the machine?

AP: Maybe we need to start thinking about new concepts of copyright: collaborative, decentralized, and distributed among many people. Maybe this is an idea that’s difficult to adopt now in Western culture, one that would allow us to think in a more collective way about authorship. This topic also comes up in another context related to AI – namely, the self-driving car. The responsibility for its potential errors and the decisions it makes lie simultaneously with the software developer, the car company, and the driver from whom such a car would hypothetically learn to drive.

This is still futurology, but discussions about such collaborative authorship are starting to pop up here and there. Well, the idea of legal protection of authorship won’t last forever. People used to have different ways of thinking about it: first there was an anonymous creator, then people began to sign their works for posterity, and only later did the Franco-German concepts of copyright protection emerge.

Nothing stays the same forever, and it seems to me that the advent of such solutions (including those that will make decisions according to algorithms) will impose changes that we’re probably not ready for today, such as those where authorship will be assigned by percentage or shared differently. It’s hard for me to imagine it at this point, but I’d agree that what’s beginning to happen now is shattering the framework in which we’ve operated so far.

Interview originally conducted in Polish, translated by SG

Language English

stanisłam lem

Lem & the Future

Wywiady Culture.pl