阿裡為何重金押寶AI晶片和量子計算
上周在雲棲大會的開幕式上，阿裡雲總裁胡曉明邀請大會主持人上臺品茶，不過品茶之前，主持人需要辨別手工沖的茶和機器沖的茶的區別。
正當主持人看著兩杯差不多的茶，無從分辨時，胡曉明掏出手機，每杯茶拍了照片，發送到天貓開發的應用程式，這個程式經過專門訓練的演算法可以分辨不同的茶葉。應用程式輕鬆地解決了這個問題。
這是阿裡巴巴對基礎技術研究與其業務需求之間相互作用的一個小例子。之後的會議上，阿裡巴巴CTO、達摩院院長張建鋒宣佈，將阿裡在 4 月份收購的中天微與達摩院自研晶片業務進行整合，成立平頭哥半導體有限公司，從事晶片的自研開發與戰略佈局。張建鋒同時表示，阿裡將在明年年中推出首款神經網路晶片，並計畫在兩三年內造出一款真正的量子晶片。
會後，張建鋒接受了《麻省理工科技評論》的採訪，圍繞公司未來計畫展開了討論。
《麻省理工科技評論》：阿裡巴巴為什麼要成立晶片公司？
Why did Alibaba create a chip company?
張建鋒：在晶片變得更加專業化的今天，自研晶片能為我們帶來更大的優勢。如今很多公司都在開發 AI 晶片，但每家公司都有不同的業務需求，所以不同的公司都需要根據自己的軟體量身定制晶片，這是傳統晶片公司無法做到的。
Zhang：This gives us an advantage. Chips are becoming more specialized. Many companies are developing AI chips, but each company has different data loads and business needs. This means each company needs customized chips optimized for its own software. This is something traditional chip companies can’t do.
But of course, we will also utilize some already widely used chip designs, and combine these with our own designs to create more value.
《麻省理工科技評論》：“平頭哥”將採用什麼樣的商業模式？
What business model will the chip company pursue?
張建鋒：大致有三個方向:首先是智慧財產權許可；二是向各個行業提供處理器解決方案，比如我們為家用電器和汽車製造定制化的處理器；第三則是為阿裡雲的資料中心製造晶片。
There are three routes. The first is IP licensing. We can license our chip designs for other companies to use. The second is providing processor solutions for various industries through IP aggregation. For example, we can provide customized processors for home appliances and cars. The third is making chips for our own cloud and data centers.
《麻省理工科技評論》：那阿裡巴巴的量子計算的發展路線呢？
What about Alibaba’s road map for quantum computing?
張建鋒：我們在量子計算上還有很長時間要走，因為在目前階段，我們還沒有自己的處理器。而有了處理器之後，我們還需要解決如何使用的問題，並不是所有任務都適用於量子電腦。 我們沒有把量子比特的數量作為我們唯一的目標，由於需要確保所有的量子比特能夠無縫可靠地協同工作，所以你擁有的量子比特數量越多，其中遇到的困難就越多。因此我們要解決的是量子計算的工程問題，比如如何在量子處理器上運行現有的程式？以及確定哪些計算任務適用於量子電腦。 同時，我們正在杭州總部建造一台超導量子電腦，希望能夠實現可擴展的目標。
This is going to play out over a longer term, because we don’t have a processor for it yet. Once we have a processor, we’ll need to answer the question of what to use it for. Quantum computers might be good for applications in cryptography or large simulations of materials, but they will not be suitable for all kinds of tasks. And the more qubits you have, the more challenges there are, because you need to make sure all these qubits can work together seamlessly and reliably.
The number of qubits is not our only goal. We want to solve the engineering problems of quantum computing. How do you run existing programs on quantum processors? You’re not going to run every program on quantum processors, right? So you’ll need to determine which computation tasks are for quantum computers and which are for classical ones. We are building a superconducting quantum computer in our Hangzhou headquarters, and we want to reach a point where quantum computing is scalable.
麻省理工科技評論：你如何看待達摩院目前的各項研究工作？
How do you prioritize DAMO Academy’s research?
張建鋒：當下，資料和 AI 是達摩院研究工作的核心，所有的工作也都圍繞這兩者展開。目前幾乎所有的資料都要通過我們的 AI 進行處理，這裡的AI部分則包括了演算法和處理器。
Right now all of our work revolves around data and intelligence. We’ll have more and more data, so how do you collect, store, and process it? We process almost all our data using AI, which is the intelligence part of DAMO’s research. And this includes algorithms and processors.
麻省理工科技評論：阿裡巴巴是如何吸引技術人才？
How does Alibaba attract technical talent?
張建鋒：阿裡巴巴的大部分業務都是基礎服務，如電商、雲計算和物流等，同時這些也是整個中國的基礎服務，因此，中國的不斷發展也將為阿裡巴巴提供更多的空間。同時很多人來到阿裡，是因為他們認為他們的工作可以改變許多人的工作和生活，甚至可以在某個領域內推動國家的進步。這些都是因為阿裡巴巴是一個平臺，你的每一個微小的工作都能通過這個平臺被不斷放大，這帶來的經濟價值和社會價值對許多技術人才來說非常有吸引力。
Most of Alibaba’s businesses are basic services, such as e-commerce, cloud computing, and logistics, which are services fundamental to the entire country of China. The fact that China wants to keep growing will provide a lot more space for Alibaba. People want to come here because they feel what they do can make a difference for many people, or even a whole country’s progress in a particular area. This is how Alibaba is unique as a platform. It means every bit of work you do on this platform can be amplified. I think that kind of social and economic value is very attractive for technical experts.
Source: MIT Technology Review
Translated by Coral Zhong
上周在雲棲大會的開幕式上，阿裡雲總裁胡曉明邀請大會主持人上臺品茶，不過品茶之前，主持人需要辨別手工沖的茶和機器沖的茶的區別。
正當主持人看著兩杯差不多的茶，無從分辨時，胡曉明掏出手機，每杯茶拍了照片，發送到天貓開發的應用程式，這個程式經過專門訓練的演算法可以分辨不同的茶葉。應用程式輕鬆地解決了這個問題。
這是阿裡巴巴對基礎技術研究與其業務需求之間相互作用的一個小例子。之後的會議上，阿裡巴巴CTO、達摩院院長張建鋒宣佈，將阿裡在 4 月份收購的中天微與達摩院自研晶片業務進行整合，成立平頭哥半導體有限公司，從事晶片的自研開發與戰略佈局。張建鋒同時表示，阿裡將在明年年中推出首款神經網路晶片，並計畫在兩三年內造出一款真正的量子晶片。
會後，張建鋒接受了《麻省理工科技評論》的採訪，圍繞公司未來計畫展開了討論。
《麻省理工科技評論》：阿裡巴巴為什麼要成立晶片公司？
Why did Alibaba create a chip company?
張建鋒：在晶片變得更加專業化的今天，自研晶片能為我們帶來更大的優勢。如今很多公司都在開發 AI 晶片，但每家公司都有不同的業務需求，所以不同的公司都需要根據自己的軟體量身定制晶片，這是傳統晶片公司無法做到的。
Zhang：This gives us an advantage. Chips are becoming more specialized. Many companies are developing AI chips, but each company has different data loads and business needs. This means each company needs customized chips optimized for its own software. This is something traditional chip companies can’t do.
But of course, we will also utilize some already widely used chip designs, and combine these with our own designs to create more value.
《麻省理工科技評論》：“平頭哥”將採用什麼樣的商業模式？
What business model will the chip company pursue?
張建鋒：大致有三個方向:首先是智慧財產權許可；二是向各個行業提供處理器解決方案，比如我們為家用電器和汽車製造定制化的處理器；第三則是為阿裡雲的資料中心製造晶片。
There are three routes. The first is IP licensing. We can license our chip designs for other companies to use. The second is providing processor solutions for various industries through IP aggregation. For example, we can provide customized processors for home appliances and cars. The third is making chips for our own cloud and data centers.
《麻省理工科技評論》：那阿裡巴巴的量子計算的發展路線呢？
What about Alibaba’s road map for quantum computing?
張建鋒：我們在量子計算上還有很長時間要走，因為在目前階段，我們還沒有自己的處理器。而有了處理器之後，我們還需要解決如何使用的問題，並不是所有任務都適用於量子電腦。 我們沒有把量子比特的數量作為我們唯一的目標，由於需要確保所有的量子比特能夠無縫可靠地協同工作，所以你擁有的量子比特數量越多，其中遇到的困難就越多。因此我們要解決的是量子計算的工程問題，比如如何在量子處理器上運行現有的程式？以及確定哪些計算任務適用於量子電腦。 同時，我們正在杭州總部建造一台超導量子電腦，希望能夠實現可擴展的目標。
This is going to play out over a longer term, because we don’t have a processor for it yet. Once we have a processor, we’ll need to answer the question of what to use it for. Quantum computers might be good for applications in cryptography or large simulations of materials, but they will not be suitable for all kinds of tasks. And the more qubits you have, the more challenges there are, because you need to make sure all these qubits can work together seamlessly and reliably.
The number of qubits is not our only goal. We want to solve the engineering problems of quantum computing. How do you run existing programs on quantum processors? You’re not going to run every program on quantum processors, right? So you’ll need to determine which computation tasks are for quantum computers and which are for classical ones. We are building a superconducting quantum computer in our Hangzhou headquarters, and we want to reach a point where quantum computing is scalable.
麻省理工科技評論：你如何看待達摩院目前的各項研究工作？
How do you prioritize DAMO Academy’s research?
張建鋒：當下，資料和 AI 是達摩院研究工作的核心，所有的工作也都圍繞這兩者展開。目前幾乎所有的資料都要通過我們的 AI 進行處理，這裡的AI部分則包括了演算法和處理器。
Right now all of our work revolves around data and intelligence. We’ll have more and more data, so how do you collect, store, and process it? We process almost all our data using AI, which is the intelligence part of DAMO’s research. And this includes algorithms and processors.
麻省理工科技評論：阿裡巴巴是如何吸引技術人才？
How does Alibaba attract technical talent?
張建鋒：阿裡巴巴的大部分業務都是基礎服務，如電商、雲計算和物流等，同時這些也是整個中國的基礎服務，因此，中國的不斷發展也將為阿裡巴巴提供更多的空間。同時很多人來到阿裡，是因為他們認為他們的工作可以改變許多人的工作和生活，甚至可以在某個領域內推動國家的進步。這些都是因為阿裡巴巴是一個平臺，你的每一個微小的工作都能通過這個平臺被不斷放大，這帶來的經濟價值和社會價值對許多技術人才來說非常有吸引力。
Most of Alibaba’s businesses are basic services, such as e-commerce, cloud computing, and logistics, which are services fundamental to the entire country of China. The fact that China wants to keep growing will provide a lot more space for Alibaba. People want to come here because they feel what they do can make a difference for many people, or even a whole country’s progress in a particular area. This is how Alibaba is unique as a platform. It means every bit of work you do on this platform can be amplified. I think that kind of social and economic value is very attractive for technical experts.
Source: MIT Technology Review
Translated by Coral Zhong
最新評論