人工智能ai發展前景
An off-shelf Artificial intelligence language processing system can generate as much as 1,400 pounds’ carbon emission [1]. An AI-based language system can create around 80,000 pounds’, which is twice of what a human being breathes in a lifetime [1]. The amount of power for searching and training a neural network architecture can be anywhere around 6,16,000 pounds [2]’ — roughly five times what an average U.S. car emits!
現成的人工智能語言處理系統可以産生多達1,400磅的碳排放量[1]。 基于AI的語言系統可以産生大約80,000磅的體重，這是人類一生中呼吸的兩倍[1]。 搜尋和訓練神經網絡體系結構所需的能量大約為6,16,000磅[2]'-大約是美國普通汽車排放量的五倍！
You heard it right; the massive carbon footprint that comes with artificial intelligence is posing risks for the technology landscape.
您沒聽錯； 人工智能帶來的巨大碳足迹給技術格局帶來了風險。
Right, A.I. continues to rise exponentially and is here to stay — but with environmental concerns at stake. A.I. has not only become a focal point of ethical concerns; it is also contributing to around 3% of total carbon dioxide globally[3].
沒錯，人工智能将繼續呈指數級增長，并且将繼續存在，但是環境問題已成問題。 人工智能不僅成為道德關注的焦點； 它也占全球二氧化碳總量的3％左右[3]。
Question yourself — Machine-driven society for who?
問問自己-機器驅動的社會是誰的？
Technology and society — these two are at crossroads. Think of roads with bots, human beings with machines and everything else driven by numbers — What do you see? Do you see smart cities or engineered ideologies with biased views splashed over them, or do you imagine a world where social good has converted into social justice?
技術與社會-兩者正處于十字路口。 想想帶機器人的道路，帶機器的人類以及其他一切由數字驅動的事物-您看到了什麼？ 您是否看到智慧城市或工程意識形态的觀點views貶不一，或者您想象一個社會福利已轉變為社會公正的世界？
Think.
認為。
Engineered education is what we like to call it, and surprisingly, it doesn’t come from the foundational knowledge. It’s only confined to the education that is imparted from the school and community. Teaching something over the education system is only helping people with clouded thoughts about how everything will switch to a data-driven ecosystem with absolutely no implications on the environment. Dr. Erin A Cech [4], in 2013 spoke about how the U.S. is trying to emphasize the importance of training ethical, socially conscious engineers but in contrary engineering education itself is failing to encourage the neophytes to look upon public welfare as their professional responsibility — and thus giving a pseudo-reality with A.I. and uninformed opinions, in equal parts.
工程教育是我們想要稱呼的，而且令人驚訝的是，它并非來自基礎知識。 它僅限于學校和社群提供的教育。 在教育系統上教些東西隻會幫助那些對一切都将如何切換到資料驅動的生态系統，而對環境完全沒有影響的想法雲霧people繞的人。 Erin A Cech博士[4]在2013年談到美國如何試圖強調教育訓練具有道德意識，具有社會意識的工程師的重要性，但相反，工程教育本身卻未能鼓勵新手将公益視為其職業責任-進而在同等程度上利用AI和無知的見解給出了僞現實。
There is more to it.
還有更多。
Photo by Ridham Parikh on Unsplash Ridham Parikh在 Unsplash上 拍攝的照片
Depoliticization — It states that cultural and social concerns of the society don’t hold good in the ‘real’ engineering work and hence should be cast out. So to say, somewhere we do bring a notion of only technological advancement, but how it will impact the human or environment around us, we are least concerned about that. Engineering work done without any thought around the people in major is just the technology of no good. As technology and society are seen as two different pillars, that makes public welfare difficult for engineers to understand what they are doing, how it will impact society. We do need metrics, which can be brought into the picture crafted by ‘socio- techno’ guys.
非政治化-它指出，社會的文化和社會問題在“實際”工程工作中并不令人滿意，是以應予以消除。 可以這麼說，在某個地方，我們确實帶來了僅技術進步的概念，但是它會如何影響我們周圍的人或環境，我們對此最不用擔心。 未經專業人士的考慮而完成的工程工作隻是一項不好的技術。 由于技術和社會被視為兩個不同的Struts，這使工程師難以了解他們的所作所為及其對社會的影響，進而使公共福利變得更加困難。 我們确實需要度量标準，這些度量标準可以納入“社交技術”專家制作的照片中。
There are a lot of things the engineered minds talk about including data-privacy, IoT, ease of accessibility, 5G, so on and so forth — At the same time, it opens a world that is close to being easily exploited. One of the best instances for this could be cybersecurity in the education system. Due to increased surveillance, policing, etc. in today’s society for better justice, it also opened a backdoor for unethical means. In modern society, even though we talk about smart cities, most of the funding does come in from the government and business people.
工程師頭腦中涉及很多事情，包括資料隐私性，物聯網，易通路性，5G等，同時，它打開了一個幾乎容易被利用的世界。 最好的例子之一可能是教育系統中的網絡安全。 由于當今社會越來越多的監視，維持治安等目的，以實作更好的司法公正，這也為不道德手段打開了後門。 在現代社會中，即使我們談論的是智慧城市，大部分資金的确來自政府和商人。
The foundation for everything? Trust. And that’s where machines lack.
一切的基礎？ 相信。 那就是機器缺乏的地方。
Well, the solution? A consortium, which can be trusted by people who are nurtured in different circumstances so that everyone can put faith in them. At least, we can look upon somebody who can make decisions regarding the price that human beings will be paying now so to correct the old deeds and for the brighter future of humankind. There is a need for social justice-based society. Emergence in research around decolonizing research methodologies, do show we have started, and we are moving towards the better world. The new generation does know what we lost and the current situation we are facing.
好吧，解決方案？ 一個财團，可以由在不同情況下受過養育的人們信任，以便每個人都可以對他們充滿信心。 至少，我們可以找一個可以為人類現在付出的代價做出決定的人，以便糾正舊的行為和人類的美好未來。 需要建立以社會正義為基礎的社會。 圍繞非殖民化研究方法的研究的出現确實表明我們已經開始，并且我們正在朝着更美好的世界前進。 新一代确實知道我們失去了什麼以及我們目前面臨的情況。
The big question — is the technology just smarter, not greener?
最大的問題-技術是否更智能，而不是綠色？
Businesses are scurrying to assimilate data and derive insights, which is giving A.I. an impetus to become stronger and, therefore, deliver better results. But the pylon holding up the burden is the environment. More extensive models, more consumption, and a negative impact on the environment.
企業急于吸收資料并擷取見解，這為AI帶來了強大的動力，是以可以提供更好的結果。 但是承受重擔的塔架是環境。 更廣泛的模型，更多的消耗以及對環境的負面影響。
When the ‘data-hungry’ machines speak, the world listens.
當“資料饑渴”的機器說話時，全世界都在聆聽。
Carbon emissions. Energy consumption. Reduction in greenhouse gas emissions. –They all veil themselves under the influence of shiny insights that machines offer.
碳排放量。 能源消耗。 減少溫室氣體排放。 –他們都在機器提供的閃亮見解的影響下掩蓋自己。
Photo by Franck V. on Unsplash Franck V.在 Unsplash上的 照片
From 2012–2018, the energy required for the computation of deep learning has increased around 3,00,000 times [5]. Machine learning models usually require more data and are prone to consume more power. To make those models even more skilled and accurate, one needs more training and execution it becomes a never-ending consumption process.
從2012年至2018年，深度學習計算所需的能量增加了約300萬倍[5]。 機器學習模型通常需要更多資料，并且傾向于消耗更多功率。 為了使這些模型更加熟練和準确，人們需要更多的教育訓練和執行，這是一個永無止境的消費過程。
Case in point — OpenAI recently launched its biggest AI-based language model — GPT3, trained on around 500bn words dataset against the previous GPT2 model that was trained on a dataset of 40bn words [6]. Before this, in 2018, the BERT, the best NLP model, was trained on a dataset of 3bn words and BERT outperformed by XLNet, which was trained on 32bn words [7]. — These numbers sound very optimistic at the outset, but the risk they come with includes lengthy training sessions translating into more energy consumption, and eventually a significant carbon emission.
一個很好的例子-OpenAI最近推出了最大的基于AI的語言模型-GPT3，該模型在約5000億個單詞的資料集上進行了訓練，而之前的GPT2模型在400億個單詞的資料集上進行了訓練[6]。 在此之前，2018年，最佳的NLP模型BERT在30億個單詞的資料集上進行了訓練，而BERTNet則優于XLNet，後者在320億個單詞上進行了訓練[7]。 —這些數字一開始聽起來很樂觀，但随之而來的風險包括冗長的教育訓練課程，這轉化為更多的能源消耗，最終導緻大量的碳排放。
Is there some light at the end of this data tunnel?
該資料隧道的末端是否有亮燈？
Definitely, there is. A.I. is unarguably saving the environment and boosting the country’s economy to provide transparency in governance. A.I. for environmental applications has the potential to ramp the GDP of a nation by 3.1–4.4% if a recent report by PwC has to be believed [8].
肯定有 。 毫無疑問，人工智能正在保護環境并促進該國經濟的發展，進而提供治理的透明性。 如果必須相信普華永道的最新報告，用于環境應用的人工智能有可能使一個國家的GDP增長3.1–4.4％[8]。
S. on S.上 Unsplash Unsplash
It can reduce global greenhouse gas emissions by around 1.5–4.0% by 2030 if the business is done as usual, while raising the GDP by a significant margin [8]. The early GDP gains are visible in a few parts of the world like Europe, North America, and East-Asia, accounting for around 1trillion USD [8] For the energy and transport sector, there could be a cut of about 2% and 1.7% respectively in total greenhouse emissions [8]. Having said that, more focus is still needed towards the water and agriculture in particular as they play a significant role in the environment in a broader sense.
如果照常開展業務，到2030年，它可以将全球溫室氣體排放量減少1.5％至4.0％，同時将GDP大幅提高[8]。 早期的GDP增長在歐洲，北美和東亞等世界各地可見，約合1萬億美元[8]對于能源和交通運輸業，可能分别削減約2％和1.7分别占溫室氣體總排放量的百分比[8]。 話雖如此，但仍然需要更多地關注水和農業，因為它們在更廣泛的意義上在環境中發揮着重要作用。
While A.I. can help make the right decisions, improve climate predictions, and work on allocating renewable resources — there have to be a few solutions to mitigate the risks.
盡管人工智能可以幫助做出正确的決定，改善氣候預測并緻力于配置設定可再生資源，但必須有一些解決方案來減輕風險。
● A.I. model training sessions can be moved to the cloud and hosted near the location where there is a more significant consumption of renewable resources. Since a cloud can store more datasets, it’s also easier to leverage data from different locations.
●可以将AI模型教育訓練課程移至雲中，并在可再生資源消耗更多的位置附近進行托管。 由于雲可以存儲更多資料集，是以利用來自不同位置的資料也更加容易。
● Making efficient A.I. algorithms can help — A recent study carried out by the Stanford group evaluated different algorithms for the same task. Results revealed that the difference in the electricity consumption of tuned and un-tuned algorithms was nearly about 880 kilowatt-hours, which is a typical consumption of American households for a month [1]. If we write better code or better models, we can make a daunting impact to reduce the carbon footprint of the application.
●制定有效的AI算法可以有所幫助-斯坦福大學小組最近進行的一項研究評估了針對同一任務的不同算法。 結果顯示，調整後的算法和未調整後的算法的耗電量差異接近880千瓦時，這是美國家庭一個月的典型耗電量[1]。 如果我們編寫更好的代碼或更好的模型，則會對減少應用程式的碳足迹産生巨大影響。
● Evaluating programs based on the default configuration, and fine-tune the model once it’s fixed. As we know, not only training of the machine learning model demands high energy, but consuming such an A.I. system consumes far more power than training. Evolution not only from algorithmic side needed but also infrastructure built on sustainable energy will also be required to facilitate such a vast A.I. dependent application.
●根據預設配置評估程式，并在修複模型後對其進行微調。 衆所周知，不僅機器學習模型的訓練需要高能量，而且消耗這樣的AI系統比訓練還要消耗更多的動力。 不僅需要從算法方面進行進化，而且還需要基于可持續能源的基礎設施來發展，以促進如此龐大的依賴于AI的應用程式。
Well, let’s hope for the A.I. pipeline to become environmentally sustainable — wouldn’t that be the literal best of both worlds?
好吧，讓我們寄希望于AI管道在環境上可持續發展-這難道不是兩全其美嗎？
[1] https://hai.stanford.edu/blog/ais-carbon-footprint-problem[2] https://www.technologyreview.com/2019/06/06/239031/training-a-single-ai-model-can-emit-as-much-carbon-as-five-cars-in-their-lifetimes/[3] https://www.seai.ie/data-and-insights/seai-statistics/key-statistics/co2/[4] https://journals.sagepub.com/doi/abs/10.1177/0162243913504305[5] https://www.sciencedaily.com/releases/2020/05/200518144908.htm[6] https://www.forbes.com/sites/robtoews/2020/06/17/deep-learnings-climate-change-problem/#2bd06ae56b43[7] Yang, Zhilin and Dai, Zihang and Yang, Yiming and Carbonell, Jaime and Salakhutdinov, Russ R and Le, Quoc V, Xlnet: Generalized autoregressive pretraining for language understanding (2019), Advances in neural information processing systems[8] https://www.pwc.co.uk/sustainability-climate-change/assets/pdf/how-ai-can-enable-a-sustainable-future.pdf
[1] https://hai.stanford.edu/blog/ais-carbon-footprint-problem [2] https://www.technologyreview.com/2019/06/06/239031/training-a-single-ai -model-can-emit-as-much-carbon-as-five-cars-in-their-lifetimes / [3] https://www.seai.ie/data-and-insights/seai-statistics/key- statistics / co2 / [4] https://journals.sagepub.com/doi/abs/10.1177/0162243913504305 [5] https://www.sciencedaily.com/releases/2020/05/200518144908.htm [6] https ：//www.forbes.com/sites/robtoews/2020/06/17/deep-learnings-climate-change-problem/#2bd06ae56b43 [7]楊志林和戴，子行和楊，宜明和卡博乃爾，海梅和Salakhutdinov，Russ R和Le，Quoc V，Xlnet：用于語言了解的廣義自回歸預訓練(2019年)，神經資訊處理系統的進展[8] https://www.pwc.co.uk/sustainability-climate-change/assets / pdf /如何啟用一個可持續的未來.pdf
翻譯自: https://towardsdatascience.com/ai-for-a-sustainable-society-731fe5116471
人工智能ai發展前景