Technology invades the modern world

Chapter 466 Virtual Social Experiment
"I know that in this era of artificial intelligence, not everyone who comes to Yenching University’s School of Mathematical Sciences wants to work in pure mathematics. Many students feel that mathematics can be a stepping stone to the artificial intelligence industry."

It's normal. Most people are just ordinary folks. Compared to the meager income of mathematicians, the prospect of earning millions a year in artificial intelligence is indeed too tempting.

As you can see from the questions you just asked, you all have some understanding of artificial intelligence, more than the average person.

Therefore, the core of all current large models is probability fitting and optimal decision-making.

The training objective of a large model is to minimize the loss function.

This means that the model is infinitely inclined to produce the output with the highest probability and the lowest error.

On an emotional or behavioral level, this means that the model will always converge toward the safest, most logical, and most expected perfect answer.

Reinforcement learning and human feedback are its underlying mechanisms. The essence of this mechanism is to use human feedback to filter and reinforce 'good' behaviors and punish 'bad' behaviors.

This makes the model a highly optimized pleaser, intolerant of errors or hesitations that do not meet human expectations.

Believe me, teachers understand artificial intelligence better than you do. The underlying logic of current mainstream AI is no different, including Crimson Cross.

When it comes to the user experience layer, the competition is about who can simulate better, whose model is more accurate, and whose training corpus is richer.

Therefore, I am extremely curious about how Crimson's Rainbow will avoid the drawbacks of large models. Will it create a completely new algorithm? Or will it completely abandon the underlying mechanisms of large models?

Xu Xian spoke eloquently on the podium.

He was very curious, as were people working in the artificial intelligence industry.

To what extent can this AI system called Rainbow actually achieve? Can it truly simulate human emotions?

Huawei was the first to approach them.

As a long-term partner of Apollo Technology, Huawei provides underlying computing power for DeepRed. Similarly, because DeepRed endorses Huawei's Ascend series chips, the development of Ascend's software ecosystem has progressed rapidly.

"President Lin, I'd like to know, what are Hong's capabilities, and is there any opportunity for us to collaborate around Hong?"

Yu Dong sat at the conference table, his expression focused and eager.

He is one of the top five people at Huawei, but he usually doesn't come to discuss business. Things like computing power procurement are handled by his subordinates.

He was more interested in learning about the latest developments in artificial intelligence and discussing his strategic ideas with Lin Ran.

Can it truly bridge the gap in emotional communication?

Lin Ran replied without hesitation, "I don't know. We all know that artificial intelligence is a black box. Even today, no one knows exactly how the so-called emergence of intelligence in artificial intelligence manifests itself."

All I can say is that our testers felt that this was more intelligent than any previous AI, and that it conveyed more emotions.

But can you say that this crosses the so-called chasm of emotional communication? I can't give you a specific answer.

Yu Dong thought of the artificial intelligence they had experienced in the past, whether it was text or voice, it had no emotions.

Like Doubao, it can simulate human tone, pauses, and breathing during conversations, but the cold reality remains: the fixed nature of computer code cannot be avoided.

What is the fixation of computer code?

This means that during the simulated dialogue process, the pauses and intervals in tone are the same between different dialogues.

Similar to the range of doubts from 0 to 100, Doubao's response to doubts always stops at the 50 mark.

This is because when the model generates speech, it needs to predict and control a series of parameters, including pitch, speech rate, volume, pauses, and so on.

The first priority of the model is smoothness, which means generating natural, smooth, and unnatural sounds without any obvious mechanical feel.

To achieve this, the model tends to choose the most stable and smoothest path in the parameter space, rather than the most extreme and irregular path.

In addition, in mean regression in parameter space, in order to minimize the loss function, the model will also tend to output a high-probability mean.

To put it more bluntly, once the AI ​​recognizes the label of doubt, it binds it to a set of preset, fixed voice parameters, such as slowing down the speech rate by 5%, raising the pitch by 30Hz, and increasing the pause by 500ms.

So the emotions simulated by artificial intelligence are fake. Yu Dong's question is, how far can you actually go?
Is it just more labels, more nuanced emotional expression, or is it a true simulation of emotions?

If Lin Ran spoke with absolute certainty, he would be less inclined to believe him.

Lin Ran's answer was, "I don't know either."

This answer shocked Yu Dong. Could it be that you really did it?

"President Lin, holy crap, you guys really achieved unstructured, high-dimensional real-time emotional feedback?" Yu Dong asked.

Lin Ran said, "If that's all you're referring to, then I can assure you that we have indeed succeeded. This is precisely the core problem that the left-right brain architecture aims to solve."

The neuromorphic chip in the right brain is inherently capable of capturing and generating emotional biases, which can break the rational control of the left brain's GPU.

Your chip R&D department should have relevant modules in the right brain chip design plan they received. If you want to know more about the specific technical issues, you can ask them. I am very much looking forward to Huawei's right brain chip.

It can make AI make mistakes, make it hesitate, and make it behave like a real person, producing random, imperfect pauses and reactions in communication.

We want to do more, but I'm not sure if it can really bridge that gap.

Yu Dong hesitated: "What gap? The gap between machines and humans?"

Lin Ran said, "Yes, the 10,000 slots this time are actually associated with 10,000 characters. We are cooperating with Tencent to create 10,000 characters in a virtual world. We have generated content such as character, background, growth environment, and job for each character."

The interaction between the real world and the virtual world will gradually improve this virtual world.

We are currently curious about a few things, such as the emergence of social emotions.

We wonder if, when 10,000 AIs with imperfect emotions begin to interact in a relatively closed virtual society, they will generate macroscopic social emotions?
Will these groups form a pattern of exclusion, tribal identity, or even systemic prejudice and discrimination? These emotions are not encoded by us, but rather emerge in a dynamic environment from their own interaction logic and pre-set background parameters.

If wisdom can emerge, what about emotion?

Yu Dong's face showed curiosity and anticipation; this was a question he was also curious about.

"The second question is whether these AI characters will develop a self-awareness that exceeds the scope of their design goals when they receive strong emotional feedback from human beta testers and are asked to make complex decisions based on preset personalities and experiences?"

Do they realize they are code, or do they firmly believe they are real individuals in a virtual world? We want to observe whether the boundary of consciousness is the end of computation or the beginning of emotion.
The third question is whether this virtual world will give rise to what we call "taste".
In our real world, architecture, music, and clothing styles all have what we call national characteristics—a kind of flavor. For example, the "German flavor"—German cars, industrial products, architecture, and even photographs taken with a camera all have a certain "German flavor."

So what about the virtual world?

The flavors in our real world originate from the randomness of history, geographical isolation, and irrational resonance between people.

It was naturally selected through high-entropy social interaction during the trial and error and inheritance of several generations.

But our virtual world is code-based, clean, and efficient.

The context we set for AI is structured text.

AI can perfectly execute instructions to create a Baroque-style building, but it cannot create a unique AI Baroque with its own historical and environmental biases.

Can randomness give rise to style?
We have endowed AI individuals with imperfections, hesitations, and randomness, which stems from the contribution of the right brain.

But can this individual randomness be aggregated into a collective style with exclusivity and recognizability through the interaction and iteration of 10,000 AIs on a macro scale?

The virtual world has no rain, no mountains, no plagues, and no wars.

Without these harsh and real external environmental pressures, can AI generate a virtual flavor that can identify itself and be recognized by outsiders, or a true cyberpunk flavor?

Of course, there are other issues, such as social ethics and human morality, which also need to be observed. However, the most important are the three I mentioned above.

Yu Dong was stunned. He realized that Lin Ran's ambition had gone beyond the product itself; he was trying to use algorithms to deconstruct the mysteries of human civilization.

It can only be said that Yu Dong truly deserves to be called a top leader within Huawei. He excels in every field he enters, from mobile phones to automobiles, becoming a top elite in all of them.

He clapped vigorously, the applause sounding exceptionally clear and loud in the quiet conference room.

"This is completely beyond the imagination of all of us outsiders!"
Because they are virtual characters, I originally thought that Crimson only wanted to solve the emotional problem this time, but I didn't expect that your purpose was to simulate an entire culture that exists in the virtual space.

This is essentially a deep-level social experiment before the arrival of a one-dimensional universe.

Our understanding of the metaverse is pixels and head-mounted devices, but what you want to build this time is the underlying logic that drives the metaverse.

It is the emergence of emotions, social rules, and culture.

If you can prove that 'imperfection, hesitation, and randomness' are necessary elements for the emergence of virtual civilizations, then all future virtual worlds must be built upon your framework to construct their souls.

This is true ecosystem-level monopoly.

Yu Dong finds it hard to imagine that an entrepreneur like Lin Ran could emerge in China.

Single-handedly, he completely shattered the notion in public opinion that China cannot innovate.

Now America is following China.

Musk has stopped talking about going to Mars and is now focused on building lunar infrastructure in Florida.

When Crimson mentioned the left-right brain architecture, everyone in Silicon Valley, from hardware manufacturers like Intel to software companies like OpenAI, followed suit, saying that this was the right way to do it.

"Mr. Lin, the reason I'm here today is indeed to discuss cooperation."

As you should know, we solved Level 5 autonomous driving a long time ago, but we haven't been granted permission to sell cars equipped with Level 5 autonomous driving to individual consumers.

We completely understand that in order to reduce social unrest, we need to wait for our competitors to catch up before we can sell.

Huawei's cars equipped with Level 5 autonomous driving technology can currently only be sold to companies, as Yanjing authorities refuse to approve or grant them approval, resulting in a prolonged delay in their implementation.

On the contrary, in Nordic countries like Norway, Volkswagen's L5-level autonomous driving system, equipped with Huawei technology, is already on sale.

"I never expected your company's design to be so magnificent."

What you just described are the sparks of civilization, the dreams of philosophers.

I was thrilled to hear it, and also felt ashamed of myself.

I originally came here to discuss whether our Level 5 autonomous driving technology could be combined with DeepRed's artificial intelligence to launch an AI-powered autonomous driving robot.

It focuses on emotional exchange and companionship. Imagine being able to have emotional exchanges with someone and also being driven to wherever you want to go.

What I'm thinking of is adding a 3D laser projector that can project the image of a girl.

When competitors put their wives in the passenger seat, they boast about how it's a "queen's passenger seat," but the wives treat them like slaves! They sit in the passenger seat and still find fault with everything.

Now you can sit in the passenger seat right now, let's call it the Emperor's Carriage!
The car girl sits in the driver's seat, always in charge of driving, always providing emotional support, and never complaining.

When we conducted research within our vehicle business unit, many people were very interested and expressed their intention to buy.

What ship girls? Are they as cute as our real car girls?
I've even come up with a slogan: Love accompanies you on every trip.

Unexpectedly, the crimson design was so grand that such a collaboration is probably impossible.

Lin Ran asked, "Aren't they saying we can't sell them?"

Yu Dong explained, "Theoretically, it's not allowed to sell cars, but I was thinking of selling robots, trying to skirt the rules."

Another idea I'm considering is to focus on the male market, launch two cars in this very small niche segment, and then price them at a relatively high level. I want to see if this plan is feasible.

Lin Ran was even more puzzled: "No, wouldn't this make it even less likely for you to sell? Would the men who buy your car still want to get married? Would they still want to have children?"

Yu Dong grinned and said, "No, no, Mr. Lin, this is your blind spot."

Our reasoning for applying is that users who can afford our car are high-income earners. After experiencing the coldness of virtual objects, they will yearn for the warmth and genuine emotions of the real world, which could actually boost marriage and birth rates.

Lin Ran stared at the sly smile on Yu Dong's face, speechless for a moment.

He realized that the business tycoon in front of him had an understanding of human nature and policy that was in no way inferior to his mastery of technology.

(End of this chapter)