Welcome to the KuppingerCole Analyst Chat. I'm your host. My name is Matthias Reinwarth, I'm the Director of the Practice Identity and Access Management here at KuppingerCole Analysts. My guest today is Alexei Balaganksi. He's a Lead Analyst with KuppingerCole, mainly in the areas of cybersecurity, but much more. Hi Alexei.
Hello, Matthias. It's great to be here again.
Great to have you. We want to cover a topic where we want to talk about actually opinions. We are talking about a topic that is really in all media. Be it consumer media, be it tech media, be it in blogs, on Twitter. You can find it anywhere. And we will cover that topic also during our upcoming event in Berlin in May, there will be the European Identity and Cloud Conference. There will be quite some buzz around the topic of machine learning, about machine learning models and how they are now available to almost everybody. Want to talk about the generation, or the automated generation of pictures, but much more importantly of texts. You already guess it, it's the discussion about ChatGPT and how it will change the way of how people are doing business, of how children and pupils are learning and how analysts are doing their work. And I think you have a really strong opinion about what ChatGPT means for analysts or for journalists or for everybody writing texts, right, Alexei?
Well, yeah, you’re absolutely right, Matthias. This is very bubbling, a very hot, widely discussed topic nowadays. And of course one has to understand that the AI in general, the machine learning in general is definitely not new. I mean, we've been using AI / ML tools for decades, really. What people are now talking about is this new generation of tools which to many people will just look like magic, basically. I mean, wow, now a machine learning tool, basically a robot can draw a picture and that will look much better than, for example, any picture I could draw. Or wow, a chat bot can actually not just reply to your ticket order, for example, as it could do 20 years ago. It can actually write an essay for you. Or even, I don't know, there have been experiments where students have actually successfully submitted a diploma written entirely by ChatGPT and it was accepted. It was not a really good mark in the end, but it was good enough for human academics to think that it was a real diploma. So yeah, stuff's happening. Really exciting stuff. But as always exciting things sometimes cause overreaction. And in that sense I am definitely a skeptic. Maybe I am slightly too old for that. Or maybe it's just my background since I actually have a degree in applied mathematics. So statistics and all those methods which are actually used in machine learning are my specialty, or at least were 20 something years ago. So I just cannot look at it as a magic trick. I'm looking at all these developments with some degree of skepticism and trying to understand is it really worth it? I mean, what other potential problems we can face while we’re relying on these tools and whether it really makes sense for me, for you or for KuppingerCole, the analyst house and the entirety of, for example, the cybersecurity industry to rely on those tools blindly.
Right. And I think it's really of importance to understand what these tools can deliver, where they might even add value and where not. So I think we both of you, both of us, of course, are relying on the work that we are doing as analysts. So we create texts, we do research, we try to generate new information because that's the job an analyst has to do most of the time. So really understanding new developments. But don't you think that there could be support in that process also by ChatGPT or any other of these now upcoming new types of assistance that can help you?
Well, first of all, again, this is not really anything new. For example, a few years ago already, Oracle has come up with their, as they call it, autonomous database, a database service which basically runs without any human intervention. It's tuned, it's powered by a set of machine learning models. To me, it actually sounds like a much more sophisticated and much more important achievement in the field of AI than a general purpose chat bot. And of course back then, a few years ago there was also a really heated discussion. Does it mean that all the DBAs, the database administrators are going to be out of work? So there were some tragedies predicted, but in the end, nothing really happened. I mean, yes, so the database is running fine. People are using it. The DBAs now have much more free time to do something more interesting, research related, whatever, and not just this menial daily job of basically tweaking the dials and stuff like that. The difference now is basically that the same technological jump is now just much more visible for the general public because nobody understands a database, but everybody understands writing an essay because everybody tried to do it back in school. So yeah, a chat bot like ChatGPT released by OpenAI recently or similar developments like Google and Microsoft, they can really now be convincing enough to not just persuade you that you are talking to a human. So they're passing Turing tests with flying colors. Not a problem at all, but they can actually replace some of the jobs that traditionally have been done by humans. Like, for example, ChatGPT can probably write a convincing article for a newspaper, and nobody will even think that it was written by a robot. And of course, the question is now, does it mean that all journalists will be out of their job soon? It's a difficult question to answer, because, yes, there will definitely be some journalists who will be out of their jobs because they are just really bad journalists that cannot write something good enough to sound better than the stuff generated by robots. So yeah, those people might be in trouble, but do they really represent the majority? Probably not.
Right. And I think especially journalists or analysts, maybe a good example because we cannot just rely on a training material that is already there. And this is what these chat bots do. They they are trained with more or less “the Internet”. So they have all the training material at hand. But that does not mean that they understand new trends, new technologies that we deal with all the time as analysts when we are talking to vendors, when we try to identify new trends. On the other hand, a journalist sometimes has to leave the building and take the camera or the mobile phone and just go there where news happens. This would be also some kind of work that just cannot be taken over by just a trained bot. Right?
All right. And this is a really important point you just mentioned. First of all, AI, at least like quote unquote AI, those chat bots, which we are talking about today, they cannot, they still cannot really think of the way we traditionally understand people are thinking, or even like dogs, crows and stuff, animals who can use tools or whatever. It's not that kind of thinking yet. It's basically just using lots of computing power and massive amounts of data to do mathematics, to recombine those data, to find patterns, to detect anomalies, to apply some algorithm and to produce some really interesting outcomes. And of course, those outcomes can be generated much faster than a human would do. So yeah, we are already there when writing an essay can be done much faster than by a human. But again, it says nothing about the quality of the output. As you just mentioned, AIs machine learning models are trained on data and they're still trained by humans. That data is never current enough. It's never complete enough. It always has inherent biases. We all know those stories from the past where an image recognition model cannot reliably detect black people, for example, because just nobody bothered to train it on black people. Or if you're going back to cybersecurity, I mean, new stuff happens not just daily, but every second, literally. And unless you ensure that your model knows about that stuff, that it's updated and retrained, if not every second, but at least like every hour, that will never be current enough to keep up with the malicious actors. But in the end, again, it's just kind of a machine learning model and AI in the modern understanding.
It just doesn't have its own voice, it doesn't have its own opinion. It's all basically a rehash and combining existing data. So would you want to read a journalist who has no opinion? Would you want to hear an analyst who has absolutely nothing to say on the subject other than repeating you the same stories you’ve probably read somewhere else? Probably not. So it's, to me, it's much less a technology issue, it's a purely reputational issue. Like I will just proclaim Zero Trust for AI. Fine, ChatGPT has its uses DALL-E or Midjourney, or whatever, the other image generation tools have their uses, but not for me, not for me as an analyst. As an analyst, my most valuable contribution to any discussion or research is my opinion after all, and my expertise, and both just cannot be replaced by an AI, at least yet. At least I hope so. So I would just say Zero Trust for AI, I will never use ChatGPT for anything regarding my work because why should I? If I am looking for some existing information? Wikipedia is still pretty useful. If I am using for hands-on information, just as you mentioned, like a journalist would take a camera and actually travel to like a war region, for example, and find something, I would rather do my own research than relying on someone else's rehashing of data. And in the end, it's my reputation at stake. It's my opinion that counts. And of course, I would still value a human opinion much higher than one from a chat bot.
Do you fear being exploited by models that are trained on the data that you produce on your intellectual property? Do you expect that to be an issue in the future as it is trained with everything that's around? It might be trained with what you and I say.
Well, I can totally understand the frustration of people like artists, because for example, those were already affected the first time those image generation networks appeared, especially there were artists just like, look this picture, I mean, it has like the face of my older work or I don't know, the legs from some other artist, just a combination of existing parts. So yeah, in a way I can understand their frustration for me as an analyst, as I am mostly writing texts, I am mostly kind of dealing with less artistic and more factual data. It's probably less relevant because again, kind of you cannot claim ownership on facts. So I would rather say, no, I am totally fine if a chat bot reuses my research provided of course it cites me as a source or cites KuppingerCole as a source, actually it would help us to spread the awareness of our research. But yeah again, it's a subject which has very little to do with technology and more with the society and the laws and the regulations, just keeping up with all the developments.
Right. This is really a podcast episode that contains opinions and expectations. One thing that I really, I'm not quite sure really what that means when it comes to the far future. If we think of more and more content on the Internet being generated by those AIs, these machine learning processes, and we should expect that because that is happening just right now. And if we have other of the same machine learning models being trained with data that has been generated in that way, does that also, or could that mean that we are narrowing down our intellectual scope because it will be again and again something that is reproduced and rearranged, but it's just more of the same?
You know, if anyone of our listeners is really interested in these philosophical and conceptual ideas, I could only recommend reading a really famous book by a Polish author, Stanisław Lem. He actually, I think it was at least 50 or 60 years ago, he has written a major study of this. I think the book is called “The Sum of Technology”. And basically by then he predicted a lot of the modern developments, the AI, the virtual reality, the kind of self thinking military weapons and stuff like that. And yes, with the amount of data the society, the humankind now produces. And I remember hearing somewhere that like the 90% of all data that we have collectively produced through the entire history have been generated in the last two years alone. So the amount of data grows exponentially. We just don't have enough humans to make sense of all this data. So here we absolutely need AI, we absolutely need machine learning models to deal with that information and not just regurgitate, but actually process and hopefully skillfully making some conclusions from that data and basically bringing those conclusions back to us because it will never work otherwise. But we have to understand the quality of these conclusions depends heavily on the quality of the inputs and the inputs is still us, basically. It's still our data, it's still our biases, it's still our garbage in, garbage out in the end. And we have to work as hard as possible to prevent that. There are some really interesting developments happening with regards to explainability of AI models. With the prevention of biases, whether it's like racial or national or whatever, any other type of bias. Some are conscious, some unconscious, but it has to be somehow measured and understood. And yes, there are tools for that as well already, just like there are tools for actually detecting the traces of ChatGPT in your essay. And again, it goes back to the reputation issue, like if you are a student, you can absolutely submit your next literally essay to your school professor using ChatGPT. You will probably save 2 hours of your life doing that, but the professor will use an existing tool to detect you're cheating and it will take him one minute. Do you think the risk is worth it? I would say probably not. And the risk is the same for me as an analyst. I could absolutely probably write my next product review using ChatGPT, but I will be detected even before it's published. And my reputation would be, if not ruined, but at least tarnished. Like, do I want to do that? No, I definitely don't. And this is again, this is my firm opinion. AI is still a tool, is still an extremely useful tool. Those models will grow in applications and importance in the future. But at least I really hope that in my lifetime, no AI will replace me and I will do my best of expressing opinions and providing points for discussions and so on to make sure that at least I am more interesting than ChatGPT.
I fully agree on that. But nevertheless, the genie is out of the bottle. The tools are there and you rightfully said it's a tool, it's a set of tools that can assist us, that can maybe make life easier in some parts, but it cannot and it will not replace everything that is around creativity and really innovativeness. The tool itself. They are innovative, There are great technologies in there and but it depends on how we use that. We will continue that discussion at EIC with other standpoints, with maybe more optimistic, more forward looking standpoints as well. So this is a great discussion. If you agree with Alexei, if you disagree with Alexei, leave your comments on YouTube in the comments section below this video, if you want to continue that discussion, please join us at EIC because there will be several tracks around that topic. I'm invited. I will be a speaker on that panel. Really looking forward to that as well. And while we are not yet at EIC, please be careful for what and how you use these tools. Have fun with them, but take it with a pinch of salt, as Alexei said. Additional thoughts from your side.?
Yes, absolutely, Matthias. We can encourage everyone to make fun with those tools or try to understand where it can actually be useful for a specific business and cybersecurity and maybe even identity related applications, but just be aware of all the potential risks. And those are not just the inherent risks of any AI model, as we discussed earlier, the biases and explainability and stuff. There are also risks in the specific implementations. Like we, for example, heard the stories about the chat bot by Google which said something wrong and knocked off $100 billion of Google's market capitalization. So probably that kind of risk is not really relevant for a smaller company like KuppingerCole. But again, if you blindly rely on every response from a chat bot to be true, then you should not. Even the likes of Google make source mistakes or just at least try to learn on their mistakes and not repeat all of those by yourself. So again, great tools, great opportunities, lots of new risks. Let's discuss those, understand those, and let's kind of try to find the balance between those benefits and risks.
Absolutely. And in the end, these are language models. These are not knowledge models. They don't claim to be genius. They don't claim to know anything. They are just great in creating text. So that is what they are about and this is what they are really doing well. But maybe as an exercise for the audience after listening to that: try to make ChatGPT lying, it’s possible and you just need to ask the right questions and it will lie to you. But it will do that in very nice words with a great language. Thank you, Alexei, for being my guest today. And if you have good lies that ChatGPT tells you, leave them in the comments section as well. And for the time being, I'm looking forward to the EIC. I'm looking forward to having you Alexei, back soon with another topic and maybe also again with an opinion piece. Until then, thank you very much and good bye.
Thanks, bye bye.
