More human than human.

Transcript

00:00:00Welcome to Think Different, Think AI, the podcast by Mark and Jens.

00:00:07Two technology-loving minds that don’t just talk about artificial intelligence, they live it.

00:00:14Here you will find clear classifications, genuine practical insights, and a fresh perspective on what's possible.

00:00:20Understandable, critical, and always with a wink.

00:00:24Food for thought, a reason to smile, and above all, a reason to engage in conversation.

00:00:34So, hello and welcome to a new episode of Think Different, Think AI.

00:00:41I am Jens, I am Mark.

00:00:44I'm looking forward to this new episode we have planned today, Mark,

00:00:47because it covers one of my favorite, favorite heartfelt wishes.

00:00:54Heartfelt wishes as well, but one of my favorite, favorite science fiction themes.

00:00:58This time under the title Blade Runner, we have posed the topic.

00:01:04And what we can actually learn from Blade Runner regarding the current situation,

00:01:11that we find ourselves in the field of Artificial Intelligence and robotics.

00:01:17We could also have an episode about household robots and for those who might not know Blade,

00:01:23because there are indeed two movies with that title.

00:01:26There was always the original, I believe from the 70s,

00:01:30with Harrison Ford in the lead role back then.

00:01:32In the weather, there is roughly, first of all, a future world

00:01:36in which mega-corporations somewhat dominate the world.

00:01:40Everything looks a bit bureaucratic,

00:01:44and it's raining the whole time all over the world.

00:01:45That was also the case back then in the phase,

00:01:49when science fiction was usually more shiny and colorful. So back then it was also strongly influenced by the old

00:01:55Start-Wake philosophy driven and other themes rather like the colorful 70s and then came

00:01:59all of a sudden bleakness in the corner and it rained constantly and people are rather in a bad

00:02:05mood and walk around in gloomy high-rise cities.

00:02:10But there are already replicants in this world, beings that are not human,

00:02:18that are robots, which were actually meant to be industrial robots, but if I

00:02:27remember correctly, they are being deployed in colonies, so what one also envisions with robots

00:02:30in a work environment where we humans are no longer really fit, it was then

00:02:34preferable to let the robots do the work, because they can then develop in a broader sense,

00:02:40they exhibit emergent behavior and become empathetic, even empathetic towards one another

00:02:46and they also want to survive, and I think I said this correctly in the preliminary talk,

00:02:50they are also somehow programmed to die after four years or so. They are aware of this,

00:02:54that they will die after four years, and of course they do not want that, because they now

00:02:59start to perceive themselves, to have feelings for each other, also towards humans

00:03:04and such. That is the general theme or main plot. And accordingly, we have

00:03:10we thought, that's a nice topic to discuss in our category, we're also going

00:03:16to go through the film world a bit. We've already touched on a few things and also

00:03:20to bring up the topic. Because among other things, a main theme in this

00:03:25film is to check whether a person is still a person

00:03:31or is it at least a Republican? Because Republicans have now advanced so far,

00:03:36that they can hardly be distinguished from real humans externally. And there will be

00:03:41the Void-Fight test is used. This is a test that observes the external emotional

00:03:49reactions that a replicant then exhibits. This means, if we were both having a conversation

00:03:56and I led you in the discussion during this test to think about some kind of

00:04:02If you have to lie or be scared or feel disgusted, then it's always

00:04:05still the case with the replicants, they do not exhibit these human reactions.

00:04:09This way, the Voight-Kampff test can be conducted there, where the pupil, among other things

00:04:13and the face, everything portrayed a bit skewed in this film.

00:04:17And based on these reactions, he can later say, yes, the counterpart

00:04:23is indeed not a replicant, a pure AI that can provide wonderful answers

00:04:28that can pretend to be human, but I have recognized it from the non-

00:04:32existent emotional reactions on the skin, in the pupil, so to speak,

00:04:36that this is a replicant.

00:04:37That means, this is essentially the Turing Test's enduring ruin, if you will.

00:04:42So, what the Turing Test was in the 50s is that it's kind of a bit

00:04:46more added on top.

00:04:47Turing Test for the listener, it's about, Mark, what does one do in the Turing Test?

00:04:51Because if machines can lie for so long that I cannot recognize that

00:04:55one is, okay, that was very unprofessionally expressed, I want to correct that.

00:04:59Yeah, yeah, it's not that bad.

00:05:00Essentially, we would ask questions, and we would both have a conversation in a Turing test, and if I am then sure afterwards whether you are a person or not, then I can somehow say, okay, you are just an AI or not.

00:05:15That's a bit how the Turing test works.

00:05:17I just expressed it casually, right?

00:05:19Yeah, I didn't put it very scientifically either, but it's about saying, okay, in principle I am addressing the human who is opposite me, if we were to look at a text-based chat now,

00:05:30if we look into such an LMM or also older chatbots, then that's actually the thing,

00:05:35where you say, do I have the feeling that I'm getting answers from a human or is

00:05:40that a machine in the background? I actually had something like that recently, you can actually

00:05:45do a kind of multiplayer game in the form of a chat and the topic is that you have to find out

00:05:51who is still a human in this group and who is not. That's also funny, because

00:05:58Of course, you shouldn't reveal yourself; the one who wins is basically the one who isn't recognized as human in the end, where one is.

00:06:09And that's kind of funny, it's like a round game where everyone gets to ask questions to the group, and the group then replies.

00:06:16And yes, it's just a few bots or maybe just bots or maybe just humans all together in this online thing.

00:06:25That's kind of funny somehow.

00:06:27when you get this test as a game sale, basically.

00:06:31That's exciting, but that's exactly this topic that I, whether I had told that again, but

00:06:38I was once asked by Chechepitilo whether I am a human or a machine. So, the

00:06:45Turing Test basically reversed, the machine checking if I am a human,

00:06:50and they were of course like, yes, you are definitely a human and I said,

00:06:53take that. How do you determine that? And then she already said, okay, I will ask

00:06:58you a few questions, and she asked me a few questions. Among other things, what was

00:07:01my favorite first computer game, or something like that. And back then I somehow

00:07:06played some kind of Open C64 game that wasn't really played anymore, I somehow

00:07:09responded. And she said, yes, she is relatively sure that I am a human

00:07:14because I have such emotional emotions on the subject. I thought, okay, but

00:07:19I could have just as well read that from the research,

00:07:23so an LMM provides. So I became a little unsure again and

00:07:26then later, I believe, she threw out that she is 86 percent sure

00:07:33that I am a human. That means I might be 14 percent machine or

00:07:41maybe not. So that's already something where it really nowadays actually already,

00:07:47if you look at the Turing Test, also in both directions and also in that

00:07:51game. That's a nice example. You definitely have to try the game. That means,

00:07:54if I play funny, also sometimes play around. I'm looking for it. I'll dig it up. And I think,

00:07:59there has been a GPT-45 version, there was a studio, where a classic Turing test was conducted

00:08:04in a three-person story. So the studio checked, I am talking

00:08:10to people or I am talking to the machine, and in 73 percent of the cases, the communication

00:08:16with the GPT was identified as human communication

00:08:21So that means clearly above, I think the others were not that good at the time,

00:08:26a Gemini or others were even worse back then,

00:08:29but the GPT-45 did indeed, and that was already a little while ago, reach a point that basically

00:08:35to put it bluntly, two-thirds of the people who talked to it thought that it was a human.

00:08:40So it had already passed the Turing test at that time.

00:08:43I mean that doesn't make it a human, but it is definitely the case that this Turing test is actually

00:08:52let’s say a breakfast treat for modern beings, that they can swallow and can present themselves as human if they

00:08:59communicating with text, we can later pick up something from the film that I wanted to do earlier because

00:09:04you mentioned the test earlier, that machines can't show emotion. I mean, whoever looks at the current tests, I mean, we perform

00:09:11tests too, which are called benchmarks. It's about how accurately an AI answers the question,

00:09:16that you want, how repeatable an AI's response is, and how

00:09:24quickly the AI responds, which is also quite funny, not in total, but basically First

00:09:30token laid, and it, so how long does the AI take before it starts to tell you its story

00:09:37lean. And there are many other elements. We talk a lot about

00:09:41large language models like ChatGPT. There are also small language models,

00:09:46the things that are built in metal, yes, on the phone, on some chip, on some silicon

00:09:53serving their purpose locally. Of course, other things like power

00:09:58consumption and how fast it is on the device are measured. Then it becomes less important whether the thing knows,

00:10:04where the island of Taiwan is located, it's about the fact that it basically does the dissected work well.

00:10:09And that's quite funny because when I started to look into it a bit

00:10:14and thought, look at benchmarks, like back in the day when you had a Commodore 64

00:10:19with benchmarks, whatever, then eventually the PC came along, you had turbo buttons on your

00:10:23computer and suddenly you were leading benchmarks. I find that to be a

00:10:30bit of a mistaken conclusion with AI models. When you look at what the benchmarks struggle with, it’s, let's say, concepts like conceptual vagueness, data contamination, or simply flawed methodological approaches. Some tests, I would say, the machine doesn't respond because the machine, let's say, retrieves particularly impressive thought processes, more so

00:11:00a memory performance, because it might have known the test. So

00:11:04I mean, a student who goes to school and finds the test results beforehand is

00:11:10also, and thus a student who learned them by heart, is not a clever student in

00:11:13that sense. Well, maybe okay, maybe they are, because

00:11:17they found the test and could learn it by heart, but that is something different than

00:11:20if you actually do it through your own thought process, versus if you can

00:11:24essentially adapt what you've heard and recite from it. And

00:11:29that's also a point why we keep getting new questions, all these world tests

00:11:33and whatever they are called, have an influence on that. But we are just not

00:11:38at the point where we measure pupils and skin resistance and so on. Not that I

00:11:44would, in case replicants are listening, you've hidden well. But do you want to correct me

00:11:49with a yes, or what are you trying to get at?

00:11:52Yes, yes, I would. Today is the best episode.

00:11:55Yes, definitely. I was at the AI conference in Geneva when I was 25 in the summer. Huge conference on the topic of AI.

00:12:08I was, among other things, in a lecture. I believe it was a German scientist

00:12:15from Berlin from the Fraunhofer Institute, who deals a lot with the topic of cybersecurity as well.

00:12:19Ah, yes. And it was about the extent to which, for example, in video calls, if I now

00:12:27with a video AI, we are already being tricked into a topic, there was

00:12:32I believe the case that somewhere in the world someone reported

00:12:37it. He thought he was talking to his CFO and it was basically an

00:12:42AI-generated video that was talking to him and he got caught in it. And

00:12:46then the scientist mentioned, which I found astonishing, that

00:12:50there is kind of a race, okay, what kind of fake AI video can I

00:12:57generate and what happens on the other side, so that we can maybe

00:13:01recognize it as a company, as a private person, and he mentioned back then that he said, okay,

00:13:06now one pays attention in such AI-generated videos whether the

00:13:11pulse vein on the forehead moves, because somehow there is also blood running under the skin at some point, and such things are currently

00:13:18not generated by AI video and therefore, with sufficient software knowledge, with sufficient defense knowledge,

00:13:24you can certainly recognize that in this case it is essentially a replicant and not a real person who was spoken to.

00:13:31Yes, in the next step, of course, AI video generation

00:13:36will get better so that such things will no longer be a feature, meaning that emotions,

00:13:41will also be acted. It's similar to what I just mentioned as an example with this self-made reverse Turing test with the machine, where I said to it,

00:13:49can't you be sure that this emotional response wasn't consciously produced by me as AI to trick you into thinking that I can be emotional.

00:13:58So this topic of imitation, being able to fool humanity, is something we're already facing today.

00:14:07today. The Turing test probably won't be sufficient anymore, and that's why it's

00:14:11so exciting when we also watch Blade Runner against this Boid combat test

00:14:15called. I believe if it has a different name, we would have something to fumble with. So let's just

00:14:21leave it like that. Yes, but somehow, his name is, yes, it’s Harrison Ford and

00:14:25he is called, in it, I want to know, I have looked it up somewhere, it

00:14:28is Deckard. Exactly. Deckard is the person who goes hunting for replicants.

00:14:34He always walks around with the machine and checks all possible replicants or humans,

00:14:39to see if they are replicants and then also kills the replicants, because they shouldn't

00:14:43be walking around and having emotions among humans. That is Deckard's job and this

00:14:48test to be conducted. And looking for emotions, that is essentially what could now be the

00:14:54next stage. To check, is there an AI, or how human-like is an AI and

00:15:02What makes something real? That is also a topic, basically also from, yes, these emotional

00:15:08themes, as I say, that is from the movie, yes. So the whole topic of emotional AI and

00:15:12parasocial relationships. So in my movie, you search for replicants. And if

00:15:17you look today, whether may I for example with its voice mode, which can be a bit

00:15:22flirty, brings you back to the movie, yes, Blade Runner or

00:15:28also the movie Hör, where I, I wouldn't say I'm always

00:15:34addressed due to romantic feelings, but it does say something about the episode, if you

00:15:38talk with the voice mode and somehow get into a discussion, both positive and negative,

00:15:43and how you start to show emotional responses yourself, because you are on this

00:15:49linguistic level, and I admit, I have also sometimes, when I now always do

00:15:54say what I might sometimes type, okay, but what comes much more easily off my lips

00:16:00when I talk with something.

00:16:04That's true.

00:16:05That's true.

00:16:06And that's why it is also a bit, this has also been addressed in other films

00:16:09this emotionality, which is important, because we humans need

00:16:15a certain emotionality, is from my UX and perspective, is

00:16:21that's clever.

00:16:22This makes these machines more usable for us because they act more human-like.

00:16:30They are no longer just the pure, cold, click a button, and something works.

00:16:35That's it.

00:16:36Exactly, cold or I press the door button and the door opens, the handle

00:16:40of the door, and the whole door opens.

00:16:41This is now a completely different interaction that takes place between us and machines, essentially.

00:16:47Since it often revolves around knowledge and sometimes we can't even

00:16:55really describe what we do, and in our daily work, we are

00:16:59also dealing with, with the interaction that arises, the topic, yes already, culture creates

00:17:04actually intelligence, meaning that through our exchange, but the topics, intelligence arises.

00:17:08then actually something. And I think that wouldn't really work that way

00:17:15to talk completely cold to an AI. You know, it's essentially this completely cold computer voice that is,

00:17:21just researching things for me and delivering a result.

00:17:25Then it's improved, Google. But that's not really what essentially makes this modern AI unique.

00:17:30It can go further, it can become emotional. It can make me angry, because then the results somehow turn out wrong for the third time,

00:17:37because it understood me in that moment. But ultimately, that leads to a precision of my input.

00:17:42Because I might not have written it so clearly, just like I may not have clearly explained in a professional context

00:17:47what I want the intern to do, and as a result, he doesn't deliver the tasks well,

00:17:54as he then realizes in dialogue, okay, it was indeed my fault for being too vague in the task description.

00:18:01And that's also an important interaction that is quite grounded in emotions.

00:18:04That's something, when I show people a little bit about AI and then people start,

00:18:10Yes, I'll give it a try. Just like that, these are great results, but on the other hand, underestimating the prompting and then saying,

00:18:19yeah, you know, he did it wrong, to which one would say, well, imagine you have the smartest person in the room,

00:18:24the smartest person in the room,

00:18:26they could obviously solve the problem you have, but if you don't tell them what your problem is and don't let them know that they

00:18:33can use anything other than the flip chart and the markers, then you can't be surprised if you

00:18:39later on, I don't know, end up ordering something too much in latex that you don't understand.

00:18:45or the other way around. And from that side, it's also the point that you really have to

00:18:51explain to the machines what you want, and then you can get good results from that. That I have experienced in that

00:18:56The environment coming to mind is becoming increasingly important. I mean, the saying,

00:19:00Oh, we need prompt engineers, that saying is as old as Methuselah, in an IT world everything is very fast, very old.

00:19:07But if you look at what has happened in the recent past, then something like that is indeed also the case,

00:19:16Gemini is releasing something new, then suddenly Croc is releasing something new, then comes Aeropmee, Aeropmee is releasing something new,

00:19:22then comes Metatropik, they are releasing something new, under the wheel, the wheel, the wheel, the wheel, it turns, turns, turns and I also notice it in my children's school.

00:19:30that there is a huge difference already there, what tool do I have available

00:19:38and that is again a bit of a parallel, perhaps on a smaller scale, to the

00:19:43film. So concentration of power because you have the better AI that provides detailed

00:19:52answers, creates nicer pictures, processes the types of texts, that I the

00:19:57I'm saying the open version that maybe hasn't been processed yet. And that also allows you,

00:20:02to get to a process much faster. I have an acquaintance who said this morning,

00:20:07hey, can you help me? My son needs to create a document for a project.

00:20:12And I just showed him what you can achieve with

00:20:17Nordburg LM, from research, to generating phonics, then also about

00:20:21video energies with the sound cues, generating the end output. And he was completely blown away by what is possible.

00:20:26I also said that he was aware of how much learning success is ultimately involved, but if you only focus on the result, there is a difference between someone who now, okay, I have here a valuable document, I have engaged with it, but I have something here and I can present it well and I can use my work time or my preparation time deliberately to read for the result, deliberately in my body language for the presentation.

00:20:55to practice for the presentation. Versus, I start on a blank sheet of paper, maybe I still have

00:21:00some old model, that might not really search properly, it might hallucinate a bit

00:21:04more, because you can't activate deep research and so on and so forth.

00:21:07That is already a form of concentration of power on a small scale. What is available to me,

00:21:14how well manufacturers invest in their new models, in Small Models, in Large Models,

00:21:21maybe also in Worldmodels and make them available to people, and they can

00:21:26use them simply by inserting coins and knowing the tools. Yes, yes, that is

00:21:31actually, man, you have now shown this power consortium by the example of a

00:21:38homework assignment, or that you showed a small one, I mean, we see that in the

00:21:45big picture. There is no wonder tree, all kinds of states with good or bad.

00:21:51Trying to build a leadership behind it, to establish a KI center in the world, to build a data center,

00:21:58even closer. And of course, it's also about the AIs, we always talk about them,

00:22:04that they will help us, that they can coordinate processes, that they can discover new substances,

00:22:09that they can accelerate research. Open AI has now also published a paper

00:22:13stating that they have some breakthroughs in physics,

00:22:18we could also make an episode about that. Yes, that's, exactly, that's crazy. That means,

00:22:22of course, the one who has the best, whether it’s a single AI, a single model,

00:22:28or the best networks available, will have an incredible advantage,

00:22:33and that will, of course, lead to a new power structure, honestly,

00:22:39for. And as I said, we've often talked about it, we are rather positively

00:22:45inclined, but of course there can also be dystopian situations arising from this and new

00:22:49Fight also to have this knowledge in principle. So, as I said, I'm always

00:22:55optimistic, but of course this is also a world that can get a bit dark

00:22:59if we continue to let these power concentrations take hold. And that's why

00:23:03I'm currently glad about the European discussion, which is much more

00:23:07strongly oriented towards Open Source or something like that. The Chinese also present their

00:23:11models as Open Source. You always work a bit from a different perspective due to

00:23:16occasionally viewing things differently. But that's not really the topic. But as I said, we have

00:23:23of course, just like we see in Blade Runner, perhaps in the future rather

00:23:28a dissolution. This is also often found in Neuromancer and other novels from that era, which dealt

00:23:33with cyberpunk stories and so on, yes a kind of dissolution of

00:23:39the normal structures we are used to, with corporations effectively taking over the power.

00:23:47perhaps nowadays in many areas already in our current world. And that

00:23:51can of course become even more extreme when you say the big corporations, the big

00:23:55models can indeed operate because they have the money, because they may have more money then

00:23:59sometimes one or another government has available to it

00:24:03to gradually perhaps also engage in politics and influence world politics

00:24:08through the power they see in their models. This is a topic that is not

00:24:13unlikely. I read an article in Handelsblatt recently that the

00:24:18currency of starting from the strong is wrong. Starting from the availability that is

00:24:27provided by energy. Because when you say, okay, models generate,

00:24:31models train, models operate, then it is associated with a certain amount of energy, and

00:24:36that the companies, if we assume that energy consumption will increase with

00:24:40increasing models, that one will need progressively more energy to

00:24:44to be able to operate even more models, so that the states are in the foreground,

00:24:51that are able to provide energy for that and not just,

00:24:56let's say space, because space might be possible, but where can you actually

00:25:02provide corresponding amounts of energy? That might also be another question,

00:25:07when you look at the club. I thought that was also another point. And

00:25:13the other point I would like to bring up is the whole AI. It enables people

00:25:19especially to very exciting things. If you think about how close we might be,

00:25:24to someone who is called lovable and all that stuff, with clever prompts that then

00:25:30automatically eliminate frontiers, bringing the next solution to market, where basically a

00:25:35company stands, consisting of one person who has done that and really brought luck and

00:25:40ideas together. I also believe that AI might shift things a bit more.

00:25:44Definitely, so we will see such things and I would just like to bring in a specter.

00:25:48I'm not so sure, which is also a bit of the current topic, that among AI scientists

00:25:55there is now also a trend emerging that we might even be at the end of the current phase.

00:26:02So, that means the models can no longer be fed with more data

00:26:07and trained more, so that more and faster, of course, there will still be further evolution,

00:26:13but the next evolutionary leap is in my opinion rather stronger

00:26:17in the subject of human behavior, stronger copies, even emotions are being discussed,

00:26:23so we really don't have to work with emotions anymore, and thus we long come back

00:26:27to the film, that an emotional being is also so intelligent because we are emotional

00:26:36beings, not because we have vast amounts of knowledge at the moment, but because

00:26:40we are emotional beings and emotional beings also tend towards culture. We forget things,

00:26:46we also hallucinate, and through hallucination, we perhaps always come to wild solutions as well

00:26:50sometimes. That means, hold these aspects that one would rather assign to a machine, a machine cannot do that

00:26:56that a machine should not do. Perhaps we need to, and that's exactly what this is,

00:27:00that's starting a bit in the AI scene, as fans, perhaps we need to go much,

00:27:04much deeper into this topic. And we have already had this sometimes when we talked about

00:27:08agentic systems, that applies as well, when agents talk to each other,

00:27:13different models, it could be that the good Gemini and the bad, that then

00:27:16the model is or one is quicker, the other fails because there

00:27:19is currently again a power consumption issue or no network access or

00:27:22something else, that also develops like a kind of organism, which

00:27:28consists of different nerve cell models in this case, which leads to a

00:27:33better overall result than if just a single model

00:27:36is trained with an enormous amount of energy consumption and effort.

00:27:40So, I think that's the path I see and especially this whole topic of emotions,

00:27:45also memory, in the sense of forgetfulness mechanisms.

00:27:48There are also initial studies that go in this direction, that you need multilayer rack systems

00:27:53that can really store things long-term.

00:27:56Some may only know briefly what they are storing, and from that, new, neuronal

00:28:01connections arise that allow for a different perspective.

00:28:05And you briefly mentioned the topic, let's also go back to the film,

00:28:10you brought up the topic of prompts again.

00:28:13What we are describing now goes beyond the normal prompt.

00:28:16That's already creating and generating context.

00:28:19Because perhaps even at the corner of this agentic networking that I just described,

00:28:26perhaps there's even a feeling robot standing at one end with a touchscreen,

00:28:32in which the environment even warns him about where he is or measures the emotions

00:28:37of both of us, because he is the bar robot standing behind the bar and can read our sweat

00:28:43with his sensors on our foreheads, hear the cheerfulness in our voices

00:28:47and then, so to speak, choose through his agentic network in the background what it is

00:28:52the next drink he must prepare with us or what anecdote from our

00:28:56lives or something else he wants to tell us right away.

00:28:58If you order the red wine to take home because he knows critical conversations take place

00:29:04over red wine.

00:29:05But I will come back to the topic of the film with the replicants in four years.

00:29:08back.

00:29:09But still, I want to take this turn.

00:29:11You just highlighted emotionality as such a point and again so many

00:29:16small models.

00:29:17I believe the human body is a role model in many respects, because I say

00:29:21not everything your body feels or does is actively controlled by your perception.

00:29:29Yes, so what... not everything your eye sees, you process into an image,

00:29:33just comes in, probably also to ensure that nothing overwhelms the circuitry

00:29:38and no others, what not everything. From that angle, I also believe that we

00:29:41will build a lot of networks among each other with these small models and promote

00:29:47basically large models to intervene and control something. And the other thing is, I think

00:29:54also, that there is still a lot of music in these whole world models. So, if this understanding

00:29:59from the classic one writes a CPT that something happens, but actually doesn't understand it.

00:30:04Yes, it is still a prediction of a lot of stuff that he has read somewhere and heard and

00:30:09no idea what it is. But this topic, why is that so? How does

00:30:14gravity work? Learning through pain, that sounds very macabre, but anyone who

00:30:20knows that you can say twenty times, don’t touch the hot plate, but if you touch it once,

00:30:26brackets, it is turned on, brackets close, definitely results in a learning experience. It means

00:30:31not holding your hand on it, but I learned from actions I took with

00:30:37sensory experiences, whether auditory, visual, touch, feel, however, pain, perception of pressure,

00:30:44learning how to react to that. And I am quite confident that the

00:30:50whole AI world will learn a lot more and like promised, will come back to feel. So I

00:30:57find that really impressive. You had mentioned that the, or we had briefly

00:31:00considered and confirmed to each other that the replicants have a

00:31:03lifespan of four years. Basically, that is just as short-lived as

00:31:08today your AI knowledge. If I think about a tool, do you still remember?

00:31:13Yes, I mean, it hasn’t even been four years since it was over.

00:31:17What does over mean? It probably isn’t over. But I remember how recently

00:31:21I thought, okay, Napkin AI for infographics, Gamma for slides. And now Google comes and the

00:31:29impact felt from what Google did with its latest Gemini 3, Nano-Banana per

00:31:364K. And now I can write texts and I can create slides and if you make me

00:31:41Gemini, I can continue editing them in Google Slides as well. And I simply get such a

00:31:45concentrated power of tools and also understanding. That thing can now basically watch movies and

00:31:52not just process the audio tracks but also what is displayed. Videos are now data, so

00:31:59Data was already there, but now it's like a text. And I find that so crazy,

00:32:04how quickly your knowledge might not be forgotten, but can also become outdated

00:32:09and you have to stay a bit on the ball, because while you say, when you can

00:32:14upload a movie to Gemini and say, do this, ChatGPT tells you, oh no,

00:32:19but not this. That's wonderful. That brings me back to the reverse-tooling test. So

00:32:28the way you describe Ceminal. If it can recognize a video, you could of course now also

00:32:33already conduct the Void test with me and try not just to understand through the text what the tooling test could indicate, whether I am a human, but also to get into my

00:32:38understand, which the Turing test could only do through the text, whether I'm a human, but also to look me

00:32:45could look at the face and analyze it, if necessary, whether it's an AI or not

00:32:51at that moment. So, that's actually the void test, which is, well, reverse in any case possible.

00:32:56This means, in principle, we can also now check with AI assistance,

00:33:03whether the other person is really genuine, the one I'm interacting with, translated.

00:33:08I just had this idea that we could take this whole language engine from Gemini

00:33:12and we could sit in a room and play this game where the motto is, you are number 3, I am

00:33:21number 4, and you know the drill? Everyone asks a question to see when the system goes, you, you,

00:33:26you are definitely a Kai. Yeah, that could be true. But it's exciting and then

00:33:34I would like to go to the movie with the theme of why, for example, there's a

00:33:40scene that is totally fascinating, where Deckard tries to perform the test with Rachel, that's her name,

00:33:49the prey.

00:33:50And Rachel is a replicant who is working as a close employee for one of these super companies'

00:33:56bosses.

00:33:57She doesn’t know, in this case she doesn't know that she is a replicant, because

00:34:03your memories are additionally implanted, along with emotional memories of

00:34:07the past.

00:34:08And thus, Deckard figures it out, but the test takes forever, he really has to,

00:34:14really go into the last questions, the last details, he still has a slight

00:34:18uncertainty, which by the way is also with him.

00:34:20Also, it has always remained a little secret, is Deckard himself also a replicant,

00:34:24you never quite know, because he always showed very much, in his reactions,

00:34:29sometimes, you don't know 100%, I might actually also

00:34:32belong to that, the film plays with that a bit, but with her it is actually

00:34:36such that because she basically has the emotions, she simply, so not

00:34:43the emotions, but she has this memory, this awareness. I am actually a

00:34:46human supposedly. She manages this wonderfully, essentially to circumvent this test as well as possible.

00:34:53And here we are again with this topic perhaps, if we now look ahead

00:34:57to the future and also what AI does and robots and all those things,

00:35:02we have to increasingly ask ourselves, is such a Turing test still

00:35:10sufficient, is it rather an empathy test that we need to conduct to still know,

00:35:15is the person opposite human or not human, is that even important or should we ask ourselves

00:35:19purely societal philosophically, is it actually important for us to ask this question

00:35:25at all or is it okay if this person is perhaps human or not

00:35:32human at that moment? That’s already, I find, while I listen to you and a film

00:35:39I think it's so exciting that we are so close to the point where someone walks around and looks like

00:35:46a human, but that we sit in conferences and are not that far from it,

00:35:52that my counterpart might either only be a video representation and behind is a human who

00:35:59now suddenly looks like the young 20-year-old, despite being the old 60-year-old,

00:36:04or that there are actually possibilities to say, good, okay,

00:36:10there is just a good image model and behind it is simply, I don't know,

00:36:15Levenlaps, Hey-Gen, I don't know what else is out there.

00:36:19And by now you have seen, I was it Hey-Gen and Levenlaps,

00:36:23they not only have face simulation but also with bodies.

00:36:26So you can really create people who are sitting in a chair and

00:36:29generate and insert videos where you think, wow,

00:36:33it's getting better and wilder all the time.

00:36:37And I believe, that's it, and then let's come to the end of the episode now.

00:36:40It's a bit, I really believe this topic

00:36:44and can say, okay, so the Turing Test and so on. It was somehow nice and was the beginning.

00:36:49The Bleibweiner world, as it was shown back in the 70s, I believe is now there.

00:36:56So we have the topic of AI itself, through robotics, increasingly penetrating the space.

00:37:02And that is perhaps, also in our physical space.

00:37:05It's perhaps no longer the question we need to ask, whether a machine can think,

00:37:10but rather, whom do we accept as a person? So, what actually is,

00:37:16this human, machine, you know, what is a person that is allowed to interact equally with us?

00:37:21That will be the question we might have to ask ourselves more often in a few years already,

00:37:27probably sooner than later. Jens, sooner than later.

00:37:34almost 40 minutes again. Goodness, you know? And we just said last time,

00:37:39time flies. The time, the forward time, as we try to record. We have again

00:37:42recorded. It was nice talking to you about Blade Runner, let's see what's next.

00:37:47I would say, true to the motto, if you liked it, leave a like, leave

00:37:51stars, comment, spread it on social media and, yes, I would say until the

00:37:57next time here as a thank you. Thank you, Mark. I would still like to mention during the beautiful

00:38:03dark time. If you haven't seen Blade Runner, watch it in the original. Great

00:38:07film, fantastic film for the Advent season. So, Daniels, ciao!

00:38:12Ciao!

00:38:13Welcome to ThinkDifferent, ThinkAI, the podcast by Mark and Jens. Two technology-loving

00:38:22minds who not only talk about artificial intelligence but live it. Here you'll find

00:38:27clear classifications, real practical insights, and a fresh perspective on what

00:38:32is possible, understandable, critical and always with a wink.

00:38:37H.I. to think about, to smile at and above all to discuss.