Ad Code

2020年年度人工智能总结



近十年来最引人关注的人工智能领域突破性成就,大都是以人工神经网络(Artificial Neural Network)为技术核心,而其代表则为数据驱动的深度学习(Deep Learning)以及机器学习(Machine Learning)。刚刚过去的2020年,或许是市场检验数据驱动人工智能能耐的关键年份。

迄今为止,数据驱动人工智能所展现的乐观发展有:

(1) 论证小数据样本学习(low-data training/small data training)成效的论文开始引发关注。这包括“一次或几次性学习”(One- or few-shot learning),甚至是“少于一次的学习”(Less than one shot learning)的科研成果。这些技术旨在克服机器学习模型“数据饥渴”短板(data hungry),让训练模型能够倚赖少量数据见微知著、举一反三,对现实世界的事物与动态做出准确的预判。

(2) 谷歌的子公司DeepMind所开发的AlphaFold 2系统在Critical Assessment of Structure Prediction(CASP)结构预测关键评估挑战赛中,以略高于90%的准确度,成功预测蛋白质的三维形状(proteins' 3D structures)。不少媒体用人工智能解开了困扰人类50年的生物学蛋白质折叠难题来形容这项科研突破。专家认为,这项突破有望协助人类充分理解病毒和癌细胞的蛋白质结构,以寻求在治疗棘手的瘟疫、癌症等生物科技上取得实质的进展。

(3) OpenAI所开发的GPT-2据说能够自学下象棋,还能作曲。在2020年6月公布的GPT-3能写诗、创作剧本,用CSS、JSX、Python等程式语言编程,并应邀在英国的《卫报》撰写一篇由人类编辑合成的文章。

(4) 在2020年8月18日至20日期间,由Heron Systems开发的人工智能系统,在美国国防高等研究计划署(DARPA)举办的AlphaDogfight模拟空战竞赛中,以5比0战胜美军顶尖F-16战斗机飞行员。

(5) 苹果(Apple)和微软(Microsoft)追随谷歌(Google)与亚马逊(Amazon)的步伐,加大对自主研发和生产AI 芯片(AI chips)的投资,放眼不再受制于半导体厂商或中美贸易战博弈的掣肘,达到在软硬体上自给自足的长远维续目的。

无论如何,在2020年,数据驱动人工智能也呈现了一些不太乐观的景象:

(1) 无人驾驶汽车(driverless car)没有落地,也没有在可预期的短期内大规模安全上路的迹象。

(2) 人工智能没有在严格意义上探测到并及时制止新冠疫情肺炎瘟疫的爆发与蔓延。

(3) 数据驱动人工智能模型,在新冠肺炎造成世界各地限行、封城,全民居家工作与上课以免受到感染的情况下,频频出错,无法准确评估消费者和借贷者的信贷评级、消费趋向等。这引发了当前最火红的机器学习和深度学习人工智能技术,只能在常规状况下正常运作的隐忧。

(4) 在美国非裔男子佛洛伊德(George Floyd)遭遇警察暴力致死所引发的全球抗议浪潮压力的驱使下, 基于无法以面部准确识别有色人种尤其是黑人女性的身份,国际商用机器(IBM)、亚马逊(Amazon)以及微软(Microsoft)在各种层度上放弃在短期内允许人脸识别系统应用于警方执法。

(5) 著名非裔女性人工智能伦理守则专家添霓 . 格布鲁(Timnit Gebru),由于坚持以谷歌员工的名誉,发表大型语言模型(large language models)带有难以能轻易根治的性别歧视和种族歧视等偏见,以及极其耗能、耗数据、不利环境等,被谷歌解职。这起事件所反映的矛盾,其实可能远比我们在媒体舆论上观察到的阶级和种族矛盾还要深刻 – 我们当前最热切追求与全力投资的AI技术是不是健全的,还是有难以被逆转的内在缺陷(inherent flaw)?

诚然,人工智能只有在与人类的作息紧密关联的情况下,才会被广泛的采纳与接受。 在资本主义挂帅的年代,使AI融入人类日常生活最直接的途径就是,将AI应用于消费主义(consumerism)当中。这也是当前许多 企业和科研单位构思、设计以及开发AI用例的方式。

在消费主义的驱使之下,AI应用程序的开发虽然不一定会以人为本,却跳脱不了以人类本身为设计主题的中心。 许多能够反映人类好恶的网络活动数据以及线上消费经历,被用于训练AI模型。人类对AI驱动电子事件的反应,被视为评估AI模型的有效性的标杆。 企业和商家希冀消费者将会产生的反应和做出的决定,被定义为AI建模的目标。

遵循这条思路,无可避免的,许多市面上所流行的AI模型,都以影响人类的审美品味和消费观点为目的。而AI模型所触发的数据驱动数码事件,还可能连带改变人类用户的行为与信念。

数据驱动人工智能或许还没有能耐在顷刻间为人类的生活带来翻天覆地的变化;但它却会为商家带来隐藏的商机,为意图引导民众舆论的广告商、营销业者、媒体,甚至是政治势力带来左右普罗大众观点和行为的数码方程式。

总的来说,基于统计学(statistics based)的神经网络(机器学习和深度学习)技术之所以被吹捧为现代AI的支柱,在于它能够持续通过数据所提供的反馈来验证其有效性,以及勾勒未来的长期价值。

从修辞层面来解读,数据驱动人工智能本身是当代企业复兴和大国崛起大叙述里的要角。它能通过形象化的数据图像勾画人类决策者已经平白浪费的过往机会成本,还有在当下能够把握,以及在未来可以从数据中持续发掘的增长契机。

文章发表于2021年1月2日的《东方日报》

参考资料/Reference

关于人工智能乐观前景的参考资料:
[1] MIT Technology Review: A radical new technique lets AI learn with practically no data | “Less than one”-shot learning can teach a model to identify more objects than the number of examples it is trained on.
[2] 滑铁卢大学提出“小于1”次学习,可能极大减少数据集需求
[3] Ilia Sucholutsky, Matthias Schonlau: 'Less Than One'-Shot Learning: Learning N Classes From M<N Samples
[4] Papers With Code: One-Shot Learning
[5] PetaPixel: Samsung AI Can Turn a Single Portrait Into a Realistic Talking Head
[6] DeepMind: AlphaFold: a solution to a 50-year-old grand challenge in biology
[7] Nature: ‘It will change everything’: DeepMind’s AI makes gigantic leap in solving protein structures
[8] Sukanta Saha: The cornerstone to the next revolution — DeepMind’s AlphaFold 2 - How AI just solved the 50-year-old grand challenge of protein folding and modeling
[9] The Science: Has Artificial Intelligence ‘Solved’ Biology’s Protein-Folding Problem?
[10] 学术头条: 又一壮举!GPT-3首次完成剧本创作,AI解决创造性问题的能力正迅速提升
[11] 维科网@OFweek:语言生成器GPT-3《卫报》专栏文章:不要怕我,我不想消灭人类!
https://www.ofweek.com/ai/2020-09/ART-201717-8500-30457682.html
[12] 科技报橘@TechOrange: 英媒《衛報》用 GPT-3 機器人寫了篇文章,主旨是「我不會消滅人類
[13] Wikipedia: Generative Pre-trained Transformer 2 (GPT-2)
[14] Wikipedia: Generative Pre-trained Transformer 3 (GPT-3)
[15] Air Force Magazine: Artificial Intelligence Easily Beats Human Fighter Pilot in DARPA Trial
[16] yahoo!新闻@TechNews: AI 的時代來了?DARPA 舉辦模擬空戰,AI 5:0 完勝美軍 F-16 飛行員
[17] deelearning.ai@The Batch Newsletter Issue August 26, 2020
[18] Wired: With Its Own Chips, Apple Aims to Define the Future of PCs
[19] Bloomberg: Microsoft Designing Its Own Chips for Servers, Surface PCs
[20] CRN: Microsoft Is Working On Arm-Based Chips For Azure, Surface PCs: Report
[21] Bloomberg: Apple Aims to Sell Macs With Its Own Chips Starting in 2021
[22] Tech Wire Asia: Sony and Microsoft develop AI imaging chip – what it means for industry
[23] South China Morning Post: Sony, Microsoft strike deal on tiny AI imaging chip with huge potential for cameras
[24] Unite.AI: Microsoft Partners with Startup Graphcore to Develop AI Chips
[25] The Japan Times: Sony and Microsoft strike deal on tiny AI chip with huge potential
[26] ExBulletin@TECH: Chip Giants Intel and Nvidia face new threats from Amazon to Google and Apple
[27] Tech Republic: What Apple's M1 chip means for big data and analytics
[28] Anuja Ketan: The AI Chip Arms Race: Will Google, Apple, Microsoft, or Amazon Kill The CPU?
[29] Financial Times: Apple chips away at a new strategy for computing


关于人工智能局限的参考资料:
[1] 《经济学人.商论》: 路障 - 无人车显现了当今AI的局限性
The Economist: Driverless cars show the limits of today’s AI
[2] Associated Press: Can AI flag disease outbreaks faster than humans? Not quite
[3] AI could help with the next pandemic—but not with this one
[4] CNBC: A.I. can’t solve this: The coronavirus could be highlighting just how overhyped the industry is
[5] The Brookings Institution’s Artificial Intelligence and Emerging Technology (AIET) Initiative: A guide to healthy skepticism of artificial intelligence and coronavirus
[6] World Economic Forum: AI can help with the COVID-19 crisis - but the right human input is key
[7] The Next Web: AI sent first coronavirus alert, but underestimated the danger
[8] MIT Technology Review: Our weird behavior during the pandemic is messing with AI models
[9] The Batch Newsletter: Bias in Plain Sight, Apple's New AI, Amazon's Firtual Fitting Room, Bot Besties, Researchers Go Ape
[10] MIT Sloan Management Review: Data Science, Quarantined
[11] Vox: Big tech companies back away from selling facial recognition to police. That’s progress.
[12] The Washington Post: Microsoft won’t sell police its facial-recognition technology, following similar moves by Amazon and IBM
[13] The Business Insider: Outrage over police brutality has finally convinced Amazon, Microsoft, and IBM to rule out selling facial recognition tech to law enforcement. 
[14] The New York Times: Google Researcher Says She Was Fired Over Paper Highlighting Bias in A.I.
[15] Wired: Timnit Gebru's Exit From Google Exposes a Crisis in AI 
https://www.wired.com/story/timnit-gebru-exit-google-exposes-crisis-in-ai/
[16] MIT Technology Review: Congress wants answers from Google about Timnit Gebru’s firing
[17] MIT Technology Review: A leading AI ethics researcher says she’s been fired from Google
[18] MIT Technology Review: We read the paper that forced Timnit Gebru out of Google. Here’s what it says.
[19] Reuters: Google told its scientists to 'strike a positive tone' in AI research - documents

Post a Comment

0 Comments