And then she would call again and a different elephant would sort of lift her head up and, you know, rumble very loudly.

然後她再叫一聲，另一頭大象就會把頭抬起來，你知道，發出非常響亮的隆隆聲。

This is Joyce Poole.

這位是 Joyce Poole。

She's been studying African elephants and their communication for 50 years.

50 年來，她一直在研究非洲大象及其交流方式。

Then I started to think, well, okay, so maybe they have a way of directing a call to a specific individual.

然後我開始想，好吧，也許他們有辦法把電話打給某個特定的人。

But we had no way of detecting that.

但我們沒有辦法發現這一點。

Decades later, she partnered up with Mickey Pardo, who designed a study around her observations.

幾十年後，她與 Mickey Pardo 合作，後者根據她的觀察結果設計了一項研究。

I went out into the field.

我走到田野裡。

I recorded calls with careful behavioral observations.

我記錄了呼喚聲，並進行了細緻的行為觀察。

So we knew who made each call.

這樣我們就知道每次是在向誰呼喚。

We knew who the call was addressed to.

我們知道是在呼喚誰。

We knew the context of the call.

我們知道呼喚的背景。

They encoded the acoustic information from the recordings into a long string of numbers, along with the data Mickey collected about the calls.

他們將錄音中的聲音資訊和米奇收集到的通話數據一起編碼成一長串數字。

They fed nearly 500 different calls like this into a statistical model.

他們將近 500 個類似的不同電話輸入統計模型。

And when given the acoustic structure of a new call, the model could predict who the receiver of the call was much better than chance.

當得到一個新呼叫的聲音結構時，該模型可以比概率更好地預測誰是呼叫的接收者。

In other words, evidence suggesting African savannah elephants give each other names.

換句話說，有證據表明非洲大草原上的大象會給對方起名字。

When we posted it on Facebook, somebody wrote back, said that the earth just shifted a little bit.

當我們把它發到 Facebook 上時，有人回信說，地球只是移動了一點點。

And I think that's true.

我認為這是事實。

This is just one example of how machine learning is decoding complexities in animal communication that humans can't detect.

這只是機器學習如何解碼人類無法發現的動物交流複雜性的一個例子。

And now some AI researchers want to take the next step.

現在，一些 AI 研究人員希望更進一步。

Large language models like the ones that power chatbots, but built for interspecies communication.

大型語言模型，就像為聊天機器人提供動力的模型，但專為物種間交流而建。

Can we talk a little bit about love?

我們能談談愛嗎？

There is still much to be learned about whales.

關於鯨魚，我們還有很多東西需要了解。

When researchers study animal communication, they usually employ a few methods.

研究人員在研究動物交流時，通常會採用幾種方法。

Recording their vocalizations, observing and documenting the behavior and context around those sounds, and sometimes doing a playback to measure the animal's response.

記錄它們的發聲，觀察和記錄它們的行為以及這些聲音周圍的環境，有時還進行回放以測量動物的反應。

All of these areas are already being impacted by AI.

所有這些領域都已受到 AI 的影響。

Recordings from the field don't usually sound like this.

現場錄製的聲音通常不是這樣的。

They often sound like this.

他們經常這樣說。

Multiple animals vocalizing on top of one another in a noisy environment.

在嘈雜的環境中，多隻動物同時發聲。

This is known as the cocktail party problem.

這就是所謂的雞尾酒會問題。

And it's a common issue in the field of animal research.

這也是動物研究領域的一個常見問題。

But machine learning solved a similar problem in human speech recognition.

但機器學習解決了人類語音識別中的類似問題。

AI researchers trained a model called deep karaoke on lots of music tracks where instruments and vocals were recorded separately.

AI 研究人員在大量分別錄製了樂器和人聲的音樂曲目上訓練了一個名為深度卡拉 OK 的模型。

Then also on the fully mixed tracks.

然後也在完全混合的音軌上。

Until it was able to do the task of separating out instruments and vocals in new music clips.

直到它能夠完成將新音樂片段中的樂器和人聲分離出來的任務。

Recently, AI researchers have had some success applying similar algorithms to animal sound recordings.

最近，AI 研究人員在將類似算法應用於動物聲音記錄方面取得了一些成功。

Which means you can take a clip of a group of macaque monkeys and single out one discernible call.

這就意味著，你可以把一群獼猴的叫聲剪輯下來，然後從中挑選出一種可以辨別的叫聲。

Researchers could also start using AI in how they use playbacks in the field.

研究人員還可以開始將 AI 應用於如何在現場使用回放。

You may have seen AI models that can be trained on lots of examples of a sound recording.

您可能見過 AI 模型，它可以在大量錄音實例的基礎上進行訓練。

And then generate another unique version of it.

然後生成另一個獨特的版本。

AI researchers are starting to develop similar models for animal recordings.

AI 研究人員正開始開發類似的動物記錄模型。

These are all types of supervised learning.

這些都是監督學習的類型。

That means that the model gets trained on lots of examples labeled by humans.

這意味著，模型要在大量由人類標註的示例上進行訓練。

And in the elephant name study, researchers were able to feed a model their observations.

在 "大象的名字 "研究中，研究人員能夠將他們的觀察結果反饋給模型。

Which, along with the sound data, helped them detect something in elephant calls they couldn't through observation alone.

這與聲音數據一起，幫助他們從大象的叫聲中發現了一些僅靠觀察無法發現的東西。

You need to annotate a lot of data.

您需要註釋大量數據。

Yossi Yovel trained a statistical model on 15,000 Egyptian fruit bat vocalizations.

Yossi Yovel 對 15,000 次埃及果蝠發聲進行了統計模型訓練。

Which was able to identify the emitter of the call, the context of the call, its behavioral response, and who the call was addressed to.

它能夠識別呼叫的發出者、呼叫的背景、呼叫的行為反應以及呼叫的對象。

And we annotated them manually.

我們對它們進行了人工標註。

You know, I'm already saying this is a restriction of the study because maybe we're missing something.

你知道，我已經說過這是研究的侷限性，因為也許我們遺漏了什麼。

We're humans, we're not bats.

我們是人類，不是蝙蝠。

And that's the problem with supervised learning models.

這就是監督學習模型的問題所在。

They are limited by what we humans already know about animal communication in order to label the training data.

為了給訓練數據貼標籤，它們受到了我們人類已經瞭解的動物交流知識的限制。

And we don't know a lot.

我們知道的並不多。

That's why some AI researchers say self-supervised models hold the most potential for decoding animal communication.

這就是為什麼一些 AI 研究人員說，自我監督模型在解碼動物交流方面最具潛力。

This is how natural language processing models like ChatGPT are trained.

ChatGPT 等自然語言處理模型就是這樣訓練出來的。

Instead of human-labeled examples, they are trained on a large amount of unlabeled data.

它們不使用人類標註的示例，而是使用大量未標註的數據進行訓練。

And they can sort it according to patterns and categories it detects all on its own.

它們還能根據自己檢測到的模式和類別進行分類。

In the example of ChatGPT, it learned from all the books, websites, social media feeds, and anything else it could scrape from the internet and came to its own conclusions about how language works.

在 ChatGPT 的例子中，它從所有的書籍、網站、社交媒體和其他任何它能從互聯網上搜索到的東西中學習，並得出自己關於語言如何運作的結論。

Every language has a shape that AI discovers.

每種語言都有 AI 發現的形狀。

This is Eze Raskin.

我是 Eze Raskin。

He co-founded the Earth Species Project, one of a few organizations that want to build models like this for animal communication.

他是地球物種項目的創始人之一，該項目是少數幾個希望建立類似動物交流模式的組織之一。

What he means by language having a shape is that language models are built out of relationships among words.

他所說的 "語言有形 "是指語言模型是由詞與詞之間的關係建立起來的。

Words that mean similar things are placed near each other.

意思相近的詞語放在一起。

Words that share a relationship, share a distance and direction.

有關係的詞語，有距離的詞語，有方向的詞語。

So man is to king as woman is to queen.

男人之於國王，就如同女人之於王后。

So this is the shape of all those relationships among the English language's 10,000 most common words, visualized here by the Earth Species Project.

這就是 "地球物種項目 "所展示的英語中最常見的一萬個單詞之間的關係。

Flattened out, it looks something like this.

平鋪開來，看起來是這樣的。

Something really miraculous happened in 2017.

2017 年發生了一件非常神奇的事情。

And that was researchers discovered that you could take the shape of any one language and match it to the shape of any other language and the point which is dog ends up in the same spot.

研究人員發現，你可以將任何一種語言的形狀與任何其他語言的形狀相匹配，而 "狗 "這個點最終會出現在同一個地方。

This idea that similar words can be located in other languages in roughly the same place is what gives the Earth Species Project hope we could do a version of this for animal communication.

在其他語言中，類似的詞可以在大致相同的地方找到，正是這種想法給了 "地球物種項目 "希望，我們可以在動物交流中使用這種方法。

To do a translation without needing any examples, without needing a Rosetta stone.

翻譯時不需要任何範例，不需要羅塞塔石碑。

This is complicated though because we know that animals don't just communicate with sound but with other senses too.

不過這也很複雜，因為我們知道動物不僅用聲音交流，還用其他感官交流。

But Eiza points out that we can learn from the fact that image generation models like Dali and Midjourney are built on the same large language model structure used for text.

但 Eiza 指出，我們可以從 Dali 和 Midjourney 等影像生成模型建立在與文本相同的大型語言模型結構上這一事實中吸取經驗教訓。

It turns out behind the scenes, it's again these kinds of shapes.

原來在幕後，又是這樣的形狀。

There's the shape that represents sound, the shapes that represents images.

有代表聲音的形狀，也有代表影像的形狀。

Those two shapes get aligned and now you can translate between images and text.

將這兩個圖形對齊後，就可以在影像和文本之間進行轉換了。

Their expectation is that where non-human animals communication would line up with ours will tell us even more about what we have in common.

他們的期望是，非人類動物的交流方式與我們的交流方式的一致性將告訴我們更多關於我們的共同點。

Dolphins look in mirrors and recognize themselves.

海豚會照鏡子，認出自己。

Elephants too.

大象也是。

That's a kind of self-awareness.

這是一種自我意識。

One concern with this plan is related to a step in self-supervised learning called validation.

該計劃的一個問題與自我監督學習中的一個步驟--驗證--有關。

Meaning humans still need to refine these models by grading them on their answers.

這意味著人類仍然需要通過對答案進行評分來完善這些模型。

How would we do that in a communication so foreign from our own?

在一個與我們的交流如此陌生的環境中，我們該如何做到這一點呢？

We also might have too high of expectations of this overlap or the capacity for having a conversation with a non-human animal in a shared language and about shared experiences.

我們也可能對這種重疊或與非人類動物用共同語言和共同經歷進行對話的能力期望過高。

Hey Kurt, how are you doing, dude?

庫爾特，你好嗎？

So I'm about to translate that into a meow.

所以我要把它翻譯成 "喵喵"。

We said hi, hi, hi, hi.

我們說嗨，嗨，嗨，嗨。

You know, next time you want to say how are you.

你知道，下次你想說你好嗎？

I do not think that humans should be considered more important than other species,

我不認為人類應該比其他物種更重要，

but that doesn't mean that there's no usefulness in distinguishing between language which is this very specific behavior that at least based on what we currently know seems to be unique to humans and other forms of communication.

但這並不意味著區分語言和其他交流形式就沒有用處，因為語言是一種非常特殊的行為，至少根據我們目前所知，它似乎是人類獨有的。

In order to build these models, the first step is collecting a lot more data on animal sounds than exists right now.

為了建立這些模型，第一步是收集比現在更多的動物聲音數據。

And so I'm actually at the moment building up a database with all the individual calls so close to 10,000 records in that which is very small actually.

我目前正在建立一個包含所有個人通話記錄的數據庫，其中有接近 10,000 條記錄，這實際上是非常小的。

Around the world, animal researchers are in the midst of a massive data collection effort tagging and recording animals with video and sound and spatial data to feed these data-thirsty models.

在世界各地，動物研究人員正在進行大規模的數據收集工作，用影片、聲音和空間數據對動物進行標記和記錄，為這些渴求數據的模型提供資訊。

Time will tell whether true interspecies communication will be facilitated by AI but researchers hope that discoveries along the way will continue to have an impact on our appreciation and protection of the species we share the planet with.

AI 能否促進真正的物種間交流尚待時日，但研究人員希望，AI 的發現將繼續對我們欣賞和保護與我們共同生活在這個星球上的物種產生影響。

We're not the only ones on the planet who can communicate, who care about one another, who have thoughts about the past and about the future.

我們不是地球上唯一能夠交流、彼此關心、對過去和未來有想法的人。

They also have a right to be here and a reason for being here. you

他們也有權利來到這裡，也有理由來到這裡。