The emergence of artificial intelligence that can talk and answer questions, even in Hungarian, became a world sensation. OpenAI’s ChatGPT, open to all, passed the medical exam in minutes, and a fifth of Australian undergraduates now use it to write their essays, and so it was banned in some educational institutions for this reason. We talked about the intelligent chatbot and the impact of artificial intelligence (AI) with Árpád Rab, senior researcher at the Institute for Information Society Research at the NKE.
Now we are in a traditional situation: as a journalist, I am asking you, as a researcher, about a single topic, and the purpose of our meeting is to produce an informative article. How long will it be necessary for us to talk like this? In other words, whose job will AI take away first?
This is not easy to answer because we shape the process ourselves. If we don’t pay attention, and we are not very mindful, we can become products, so it is not up to us to decide this issue. But let’s be cheerful! AI also helps us a lot at the moment: it monitors our correspondence, improves the sound quality of the recording of this conversation and helped us to arrange this meeting. That is why it would be more accurate to use the term artificial intelligences. It’s also important to see that artificial intelligence just pretends to be smart. Wow, he knew that I was jogging; how cool that he could record and analyse my movements. But in reality I’m just giving it a series of numbers, if I start from here and get there, if I’m moving at that speed and my body is moving in rhythm, that means jogging. So AI is both intelligent and stupid at the same time. To answer the original question: whatever our craft is today, let’s think about what we love about it. Whatever don’t like, I immediately look at how AI can do it. However, what I like, I have to be very good at. A lot of what we used to define as workplace values have been rendered meaningless by artificial intelligence. One example is tolerance of monotony. Artificial intelligence never fails in monotonous tasks, because it is too stupid to do so. If I look at my own work then yes, it reads a million and a half documents and writes a page on it, which for me would take a week. But the main question is what is going on inside me during the week while I’m researching. Because when I read, I also think about what I read, shape the text, and become curious about something or not.
What would a socially useful artificial intelligence look like?
Research into AI is as much a social science issue as a technological one. I will highlight two very important attributes. One is how to make AI transparent. So let’s understand why it says: Árpád, now turn left. If we don’t ask why, we become a product. The second important goal is to have lots of tiny artificial intelligences. At the moment there are huge public companies developing AI, their aim is more to create one big AI, because then it’s one platform and I become dependent on it. But society has no interest in that, so it’s better to have lots of little ones: one that tells me how to turn left on my bike, one that translates something, and so on. If there are a lot of little ones, then I can control it, then I can say: look, you’re a language translator AI. Why are you wondering about my weight? Don’t interfere with my studies! Just translate it! Do that correctly!
Can this be achieved by legal means?
I think it is not enough to say that the regulators will sort it out. The law lags behind a lot; the law needs court cases first, it has to process them, it has to see if anyone suffered any harm. But it is very difficult to put into words whether you were harmed by the fact that the route planner took you to a certain route. Because maybe that’s why I didn’t see a café I could love, instead I saw one that someone paid for me to see. But I will not know about this shortage itself. Law, of course, has a huge role in this. Business has an equally huge role to play in responsible behaviour. But what I believe is that we, users have the biggest role in using machines as smartly as possible. If we can do that, it will be easier to regulate and it will fall into its place. I personally think it is also very important to develop Hungarian artificial intelligences and to demystify the phenomenon. Let’s face it, irrespective of its complexity, it is still just a machine.
Why is this important? Can you give an example?
There is a lot of talk about social media intercepting us. Everybody has a story about how they talked to someone about herbs and, an hour later, herbs popped up as an advertisement. Even if it may seem so, we are not being intercepted. Technically, it would be very demanding: it would take a hundred and fifty megabyte audio files a day, uploaded, and in thirty days it’s already a gigabyte; the user would notice. On the other side, there would be 20 petabytes of material, daily. This is not the way forward, but instead that AI learns about us based on our behaviour and the combination of this data. AI can see that a friend of mine searches for herbs daily. It can see from the geolocation data that we met. So it throws in ten ads from my friend’s search history. I won’t consciously notice nine, but the one with herbs hits. My hand stops there: at that moment, the machine already knows that I’m interested in it for some reason. So it throws in another herb to see if it works. Of course, this also requires someone to pay to pop up herbs.
As you mentioned, we’re surrounded by a very diverse AI. But the real star with the press now is ChatGPT. What makes it different, or why does it have a different reputation from our hitherto almost unnoticed old helpers?
The development of artificial intelligence had already got this far, but society hasn’t yet encountered it. Here at the Information Society Research Institute of the NKE, we have been talking for a year or two that this situation will arrive. At Google, people have probably been talking about it and preparing for it for five years. The reason that so many people noticed it was up to the service built around the technology. It is also very important to see that there are many different ways of developing artificial intelligence, and all of them are used at the same time. One element is good programming and data, and another is that we collect behaviours, so the machine has to be used. The way I develop the self-driving car is to program it and run it on a test track, and, thirdly, I send it out on the road: it’ll do 200,000 kilometres and it will be smart in this way. The development of AI reached the stage last year to make it available in open test mode.
So we’re all testers?
Actually, yes, that’s why OpenAI published the open source artificial intelligence ChatGPT. It’s a website that is available to anyone who registers and its use is free. The user can see a chat window, which they may have already encountered on other platforms. At first, everyone writes a word or two, a sentence, just out of curiosity. Then someone realises that they can get the answer to any question. I tell it: write about a red mobile phone in Shakespeare’s style. And it writes a sonnet. If you need it, it translates it into Hungarian. In fact, I talk to it in Hungarian to begin with; it translates it into English in the background and then translates it back. It answers in a very complicated manner and often well. When I ask it what e-learning is, it gives me an answer in bullet points. This is a big challenge for Google.
How did Google get to this point, what is the underlying principle?
Google created a very innovative search algorithm a long time ago, the full code of which no one knows, but we do know that it manages the information that people valued. It looks that this page is a visited one; there are a lot of links pointing to it, so it gets data from that page. We humans validated the searches ourselves, and it became better over time. But with Google, I type the question into the search engine, the results page loads, I click on it, and there’s the text. It is possible that for my question, “What is e-learning?” the article written by a really good researcher is in tenth place only, with the first nine being advertisements or even superficial. But with ChatGPT, there is only one step: I type in the question and the answer is right in front of me. No need to bother with sources and links. If I trust the developers, I’d say that’s really what e-learning is.
To have confidence, it might not be a bad idea to understand how it works. What can we know about this?
Two types of generative artificial intelligence are now on the market. Generative AI means that I give it commands and it performs them. One is the text version and the other group is where I type in words and it generates an image for me by using them. We call this prompt programming or prompt engineering. We’ll be working on this a lot in the future. Behind the service there is a huge database that both searches the internet and can compile the results into text. For this, it needs to understand what you want and give you the answer you want. Understanding what you want is the prompt part. So I’m typing “write a Shakespeare sonnet from a red mobile phone”. It is intelligently stupid, as we know: it doesn’t understand any of the words, but looks them up in the database. From the billions of web pages it sees, it searches for Shakespeare’s sonnets, and describes algorithmically what they look like, what words they use. It comes back to the red mobile phone, looks at all the mobile ads, looks at the mobile pictures, looks at the colour red, looks at a bunch of poems, looks at the context in which red is used, all in a matter of tenths of a second. All I see is that I typed in the question, I hit enter, and almost immediately it returns the words to me in Shakespearean fashion. And I say “wow, how clever it is.” Of course, it has no idea what it wrote. I, the human, am the one who says it is a poem. And then I start to think, “yes, it’s a miracle, but somehow it doesn’t make sense.” It got this far today.
There is always something strange about these machine answers, or, to paraphrase Attila József, “The nothingness in it floats / as if it were / the dust of something…
Yes, that’s partly because it is not personalised yet. On the other hand, this is a completely open test, where the developers have set very tough conditions to ensure that no one is offended or hurt by the answer. If I ask it whether red or blue is better, it gives me a basic neutral answer that it is subjective, a matter of taste and so on. But if everyone has a different database and I, as a developer, for some reason want others to dislike the colour blue, then I say the answer to that question should be that red is better and blue is worse. In the future, we will increasingly expect AI not to be so neutral.
For example?
Let’s say we want to have machine doctors. We don’t want to take the work away from doctors; we want to help them do their job and replace them in situations where a human presence isn’t needed. However, in this situation, the machine can’t remain neutral. It can’t say that it thinks you could lose weight, but that’s your choice. A doctor needs to say, “look, Árpád, you’re overweight. You need to lose that much weight. That’s how much time you have.” And that isn’t a thing to say nicely. It is also very important not to use these tools without a purpose. I think what separates us from the products is that when we sit down to use a device, whether it’s a motorbike, a smartphone, whatever, we have questions in our minds. Do I have a case that only I want? Or am I just waiting for someone to do something for me. If I’m just scrolling social media because I’m bored, I’ll become a product because I’ll be distracted. But if I want to know what happened to my best friend, I already have a case. Sometimes, of course, we can be comfortable, say in ten or eight situations a day. I start streaming music, and ask to select music to suit my mood. I don’t care if someone paid to get their song on the list, it’s just playing. The first step to consciousness is to find out when to be conscious. When does it matter that what the online translator has translated, I also have to review? It doesn’t count in nine times out of ten letters, talking about my work, because they are operational letters. “Hi, do this, do that”, it translates into English, that’s fine. But if the tenth, let’s say, goes to a new client, or an invitation to a conference where’ a single misspelling of a letter can be problematic, I don’t rely on machine translation completely.
User consciousness matters a lot.
Besides trust, I also wonder about what happens to those who provide the input. Google search is still transparent in this respect, because there are actually links as results, meaning. I can look at the human-generated content that is the answer to my question. In the case of ChatGPT, I have no idea who created the response.
This is a very serious issue. We can’t use source criticism, in other words. to decide whether the answer the machine gives is really the right one. It is a nightmare from a copyright point of view. Of course, it could be solved: for example, by adding its sources in a separate file each time it replies. It’s a downloadable file, so I don’t necessarily need to read it, but if it’s important then I’ll press the button. The other big question is why should I create something on the Internet when I don’t know who will use it. For example, I write a poem and ChatGPT might learn from it. It won’t give users my exact poem, but if my style appeals to them, they can use it in their own response. So they did steal something from me, but not my specific poem, so I can’t even sue them. I think the answer to this will be the fragmentation I mentioned earlier, when I say that this AI can only be used for this.
As an educator, how can you deal with this phenomenon; say, in the case of thesis writers?
Indeed, a lot of thesis writers are already using ChatGPT to write their thesis. Students are very fond of technological innovation, they think it is new to their teacher, so they’ll get away with it. I might try to guess what question the student asked. I type it into the computer, and if it gives the same outcome, I can tell the student that it is a fraud. But, eventually, we’d find ourselves fighting each other through digital tools instead of having an hour of conversation with the student. So the solution is not for a machine to recognise text written by other machines, but perhaps not to ask for 40-page theses. It might be enough to write five pages, or even just an oral presentation to show the student’s own ideas.
After all, is it good or not good that we can talk to machines?
Personally, I see at least seven benefits of artificial intelligence, broken down into, say, ten factors, but there are also three that represent very large risks. And that’s what we’re working on here at the institute, to reduce these three to two or one. User consciousness matters a lot. For example, if I want to know when Michael Jackson was born, I don’t Google it anymore, I ask ChatGPT. But, for important decisions, I can’t rely on that tool alone, I go back to where I can see the sources and the person behind the content. If we behave like this then the developers will also say, “well, we’ll invest in the sources. If we don’t behave like that, they’ll never put the resources into it. I don’t think artificial intelligence will take our jobs and leaving us with no tomorrow. Rather, it will take them away in a way that makes us feel that everything is eroding. I don’t get paid as much for my work anymore because part of it is done by a robot. I can’t make money any longer with my creative mind, my writing, my photos, because it’s all on the Internet.
Is it legal to do so or can it be restricted? How can we protect our intellectual property?
One answer to this might be, for example, a return to the physical world. We’re back to not painting digitally and hoping that someone will pay for the picture, of which there is really only a single one. If I’m a journalist, I accept that my article won’t be seen by the whole world, but only in a paper-based newspaper, but there they will know that I wrote it and the people I intended it for will read it. So we build up a bit of a wall around ourselves. Of course, let’s not forget that right now we only get a free test service, which otherwise has large costs. In fact, you can charge whatever you want for its use; at today’s prices it could be 20-30-40, or even 200 thousand forints. We get it for free now to get us used to it, but this won’t be the case in the future.
Imagine the economic part of it, that there will be a paid search in addition to the free one, which is, say, $20 a month, to buy the AI add-on?
This is already the case. Let me give you an example: online translators have evolved a lot because we correspond and chat a lot. Today, Google Translator, which is free, translates quite reasonably well. But there are also paid services on the market that translate even better. Not much, a little bit better, and sometimes it is worth the price. It depends on who you’re corresponding with. But in the same way, the free-for-all Google journey planner can distribute cars around the city. But if I’m a VIP customer, he’ll say “I’ll distribute you first, so you’ll always get there two minutes early”. That doesn’t seem like much to me, who drives once a day, but, say, a taxi company drives tens of thousands of times. For them, at 1000 cars, the benefit can mean days of work, for which the company says “okay, I’ll pay for the service”. I think there will always be free artificial intelligence. But whether it checks out, let’s say, the Scopus social science journal database for what e-learning is, or just the blogs on the internet, that becomes a pricing question. I’ll say that I’m opting for a full service. It says fine, but that will be 600 forints. For me it’s worth it, because I say I’m going to spend it on my thesis. And then the machine will look through these messages if it can. A few hundred forints of the price will be what the service provider pays to use the database, and in an ideal case the owners of the database will pay the person who created the content. This is not the case in academia at the moment, but it could be.
I feel that this whole phenomenon goes far beyond the issue of pricing…
Of course! Think back to, say, five or ten years ago, how we used to do things, how we used to search for information or plan our holidays. If we think about how we do these things today, it turns out that we are indeed living in the future. Artificial intelligence is here; it isn’t just an economic issue, it’s not just a curiosity and it isn’t the future. It’s today. It’s good that it exists, because with its help the human race might survive. But no-one should feel that this is something that happens to other people. Let’s not think that this must be a problem for lawyers only, or for students, accountants, journalists. No, it affects everyone. Today, six hundred thousand jobs in Hungary could be directly affected by artificial intelligence. This means that these jobs could be filled by a machine right now. This is a lot. The work of the future will be more about asking the right questions, phrasing them well, with a more accurate, precision brain, rather than mechanical repetition. So I urge everyone to take AI seriously, to think about it, to improve themselves and to experiment and play with ChatGPT.
