对话图灵奖得主：图灵测试不能来比较人类和机器的智能水平？

对话图灵奖得主：图灵测试不能来比较人类和机器的智能水平？ - 双语字幕

Thank you very much watching.

I'll see you in the next video,

see next And I think that the current test is, by no means, to convert human and machine intelligence.

But we do not have the principles and techniques to synthesize them and build complex intelligence systems,

so we're quite far from achieving artificial general intelligence.

Do you think there's a golden reference to the intelligence?

There are many intensions, there is no way we need concept of intensions, there are many intensions, okay?

It's clear.

你们 equivalents to videos

python and most of

years ago that community he had also was there save President the of United States I would thank you for your support.

I want to you Thank you for your I wish you for your support.

I wish you a great day.

you for What do about it?

I don't know why here today or

I think that's why I'm

I'm I I'm now something to I'll give him something to So I'll be There's a couple of questions that I like to discuss,

and most of them is related to the very hot topic recently,

which is artificial intelligence in the AI,

because I see there's a large portion of your book is talking about the difference between a human intelligence and the AI.

And you mentioned about the difference and the similarities between both.

Would you would you be able to clarify a more like about the comparison between human intelligence and artificial intelligence, please?

Yes, I think there is currently a lot of confusion about what intelligence is and how it can be achieved.

And without a clear idea of what intelligence is, of course, we cannot develop a theory of how it works.

So should agree on that.

If you open a dictionary,

say the Oxford Dictionary,

You will see that intelligence is defined as the ability to learn,

understand and think logically about the world and the ability to do things well.

So, of course, machines can do impressive things.

They outperform humans in games and are capable of performing a wide range of tasks.

including some that are obviously sensitive like a chat GPT.

However, machines cannot perform humans when it comes to situational awareness, to understand the world, adapting to challenges in their environment and the creative thinking.

I believe that.

We have weak,

what we call weak,

artificial intelligence that gives us the elements to build intelligence systems,

but we do not have the principles and techniques to synthesize them and build the complex intelligence systems as we build the bridges from building blocks.

So, We cannot build the intelligent systems that can replace reliably humans in various tasks.

And believe that the road to artificial, what we call artificial general intelligence passes through an important milestone that is autonomous systems.

And I work on autonomous systems I've been working for here.

I could say a lot about autonomous systems.

These are very different from these are reactive systems.

They have to manage dynamically many they must cope with the uncertainty of complex environments,

they collaborate harmoniously with human agents and this is not achievable today by using say LLM.

LLM cannot drive cars, cannot organize production is more factors and things like that.

So we're quite far from achieving artificial general intelligence.

I think one of the major disadvantages or limitations of current like they still require human guidance or something,

even though like LLMs seems to have some sort of intelligence, but it's still based on human guidance or supervision, isn't it?

Yes, it's clear.

I and also they can perform only specialized.

tasks, okay, and human intelligence is multifaceted, has many different facets, okay, it's a combination of skills and so.

Machines would be able to perform the tasks that humans can perform,

but also what is important is that humans can call it innate, can make choices by managing many different goals.

So are quite far.

Yes.

Okay.

Because I see from your book,

like mentioned this very, for me, it's kind of a very controversial idea because you mentioned the human can't understand many complex things like together.

You mentioned about like human can at a time can process about five things, the connections between those.

So for human beings, it's a bit of a difficult to understand very complex behaviors.

But you also mentioned it's very lucky for human because the classical physics, you know, formulas or theories is very simple.

And why human can understand that, right?

So, but in that sense, do you think it's an advantage of human intelligence or disadvantage of human AI or human intelligence?

Well, human intelligence is the result of a historical evolution.

Okay, and so it tries to meet the needs we have when we behave in interaction with a physical environment.

It's clear that the machines have exhibited different kind of intelligence,

can have a limited memory, okay, can make very precise computation, and human intelligence is not good to interact.

And here I'd like to to to mention the Turing test because there is a lot of discussion about that how we compare human and machine intelligence.

And let me remind that the Turing test that was proposed by the founder of computer science who is Alan Turing.

is designed to compare a machine and a human that received questions from an experimenter.

So we have an experimenter.

You have a machine and human in different rooms, the experimental sense questions.

And then the experimenter compares the answer.

So are written questions.

And that,

if the experimenter cannot distinguish which is the computer and which is the person, then the two are equally, Why are you mentioning this test?

Because there is currently a lot of discussion.

People claim that the system is intelligent.

because it passed the tool in test, okay?

And I think that the tool in test is by no means adequate to compare human and machine intelligence.

Why?

Because the judgment is human, is subjective.

And also the choice of test cases, of course, can be questions can be biased to favor other machines or the human.

And also,

it's clear that the testing intelligence cannot be a question and say,

again, it cannot be a conversation because human intelligence is expressed by intellectual environment.

We we move, we have social behavior.

So clear that we need some criteria to agree how we can characterize human intelligence.

And this is something also I discuss in my mind.

We have proposed what we call a replacement test.

So idea of the replacement test is a generalization of the tooling test.

You would say that an agent A can be a machine or a human is as intelligent as a human B that's performed by human human task.

If A can be replaced successfully B, so I would say that.

a machine is as intelligent as a human driver if it can successfully replace the driver.

So in some cases the machines will prove to be more intelligent than humans or humans and machines.

So this is very important to understand.

intelligence, there is no unique concept of intelligence, there are many intelligences, okay, and each characterizes the ability to perform tasks.

So, when people say that this system or my machine is smarter than a human,

this is not meaningful because they have to specify for which kind of tasks.

So, clearly for instance, LLMs are very good at some arising tasks, and cannot be done by humans, okay, but there are other differences also.

And so, it's very important that we agree on the concept of intelligence.

Of course, if by saying intelligence, we mean human intelligence, this should be the reference, okay, and it is true.

perhaps are limited in,

as you said,

in analyzing multi-dimensional data,

but a couple of common sense knowledge of abstraction of creativity this this cannot be be realized by using machines, okay?

Yes, okay, I understand.

So it's a very cool idea.

I'm very happy that you bring up your intensity.

questions I would like to ask, but I think it's intended to bring out the to test to use human intelligence as the reference.

But in your proposed alternative test, do you think there's a golden reference to the intelligence?

Should a human intelligence to be the golden reference or like, like, because for human intelligence, we've got IQ test, like we've got the marks.

Would you like to, would you think there's going to be something similar?

It's very hard because you see human intelligence is a combination of skills.

And depending also on the persons, the weight of its combination, the importance of this combination is different.

For you can be a gifted football player and be too bad in mathematics.

So cannot say this is the benchmark for human intelligence.

Of course,

The human agency has some characteristics that we can talk about,

in particular,

we can talk about human consciousness and things like that,

okay, but I mean, whatever definition precise definition you give, this will be debated, of course.

From the cheap design or the silicon industry,

we also talk about human bring,

for example,

it's a very powerful computing machine,

like we can perform much better performance in comparison to very powerful human build machines,

but at a cost just like maybe a burger or something.

like lots of power to put or to power your brain.

That's the fundamental difference between human beings.

but I think also,

and this is something I explained in detail in the book, that the brain will not be enough to understand human intentions.

Yes, why?

Because the human intentions in fact It's like our computer.

Our computer consists of hardware, but also software plays a very, very important role.

And human intelligence, mental phenomena are very important.

And here,

there is a huge debate, of course, because some people claim that mental phenomena do not exist, I mean, or do not play any role.

So try to understand the intelligence by starting only the brain, okay?

Because the brain is just a real thing, okay?

But I have an objection to that.

I don't care, I will not argue whether mental phenomena are real or not.

I mean, this is a completely useless discussion.

But what I say in my lectures is that if I ask you how your laptop behaves,

and I give you access to all the circuits, okay?

Can you infer which function it computes because once you compile the software, everything is electrical signals, okay?

but it's very hard to understand what the raptor does.

And it will be very,

very hard to penetrate, to deepen into human intelligence without studying the relationship between mental phenomena and phenomena computation in the brain.

And I think that's an also you mentioned about like another cycle or like a bigger cycle of consciousness versus intelligence is like,

although we talk about the individual intelligence but are human to not live individually, right?

They in the society, living in social lives.

So behavior of the of the people around you may affect your behaviors as well.

Like we see like very interesting behaviors.

Like we know this is very risky and we know that it's not going to have good results.

It depends on how do you define good,

but people still doing it, that it still needs to take risks and still need to do it because of the social life.

Yes, but also language.

Language is the result of interaction between humans.

I mean, if a human lived isolated in the natural, I no need to develop language.

And you know that the emergence of language is a key factor for the development of humans.

So, you see, human intelligence is the product of social revolution of humans.

翻译语言

选择翻译语言

视频总结导出字幕口语评测

解锁更多功能

安装 Trancy 扩展，可以解锁更多功能，包括AI字幕、AI单词释义、AI语法分析、AI口语等

兼容主流视频平台

Trancy 不仅提供对 YouTube, Netflix, Udemy, Disney+, TED, edX, Kehan, Coursera 等平台的双语字幕支持，还能实现对普通网页的 AI 划词/划句翻译、全文沉浸翻译等功能，真正的语言学习全能助手。

支持全平台浏览器

Trancy 支持全平台使用，包括iOS Safari浏览器扩展

多种观影模式

支持剧场、阅读、混合等多种观影模式，全方位双语体验

多种练习模式

支持句子精听、口语测评、选择填空、默写等多种练习方式

AI 视频总结

使用 OpenAI 对视频总结，快速视频概要，掌握关键内容

AI 字幕

只需3-5分钟，即可生成 YouTube AI 字幕，精准且快速

AI 单词释义

轻点字幕中的单词，即可查询释义，并有AI释义赋能

AI 语法分析

对句子进行语法分析，快速理解句子含义，掌握难点语法

更多网页功能

Trancy 支持视频双语字幕同时，还可提供网页的单词翻译和全文翻译功能