If AI has anything to say to 2020, it\u2019s \u201cyou can\u2019t touch this.\u201d<\/p>\n\n\n\n
Last year may have severed our connections with the physical world, but in the digital realm, AI thrived. Take NeurIps<\/a>, the crown jewel of AI conferences. While lacking the usual backdrop of the dazzling mountains of British Columbia or the beaches of Barcelona, the annual AI extravaganza highlighted a slew of \u201cbig picture\u201d problems\u2014bias, robustness, generalization\u2014that will encompass the field for years to come.<\/p>\n\n\n\n On the nerdier side, scientists further explored the intersection between AI and our own bodies. Core concepts in deep learning, such as backpropagation, were considered a plausible means by which our brains \u201cassign fault\u201d in biological networks\u2014allowing the brain to learn. Others argued<\/a> it\u2019s high time to double-team intelligence, combining the reigning AI \u201cgolden child\u201d method\u2014deep learning\u2014with other methods, such as those that guide efficient search.<\/p>\n\n\n\n Here are four areas we\u2019re keeping our eyes on in 2021. They touch upon outstanding AI problems, such as reducing energy consumption, nixing the need for exuberant learning examples, and teaching AI some good ole\u2019 common sense.<\/p>\n\n\n\n You\u2019ve heard this a billion times: deep learning is extremely greedy, in that the algorithms need thousands (if not more) examples to showcase basic signs of learning, such as identifying a dog or a cat, or making Netflix or Amazon recommendations.<\/p>\n\n\n\n It\u2019s extremely time-consuming, wasteful in energy, and a head-scratcher in that it doesn\u2019t match our human experience of learning. Toddlers need to see just a few examples of something before they remember it for life. Take the concept of \u201cdog\u201d\u2014regardless of the breed, a kid who\u2019s seen a few dogs can recognize a slew of different breeds without ever having laid eyes on them. Now take something completely alien: a unicorn. A kid who understands the concept of a horse and a narwhal can infer what a unicorn looks like by combining the two.<\/p>\n\n\n\n In AI speak, this is \u201cless than one-shot<\/a>\u201d learning, a sort of holy-grail-like ability that allows an algorithm to learn more objects than the amount of examples it was trained on. If realized, the implications would be huge. Currently-bulky algorithms could potentially run smoothly on mobile devices<\/a> with lower processing capabilities. Any sort of \u201cinference,\u201d even if it doesn\u2019t come with true understanding, could make self-driving cars far more efficient at navigating our object-filled world.<\/p>\n\n\n\n Last year, one team from Canada suggested the goal isn\u2019t a pipe dream. Building on work from MIT analyzing hand-written digits\u2014a common \u201ctoy problem\u201d in computer vision\u2014they distilled 60,000 images into 5 using a concept called \u201csoft labels.\u201d Rather than specifying what each number should look like, they labeled each digit\u2014say, a \u201c3\u201d\u2014as a percentage of \u201c3,\u201d or \u201c8,\u201d or \u201c0.\u201d Shockingly, the team found that with carefully-constructed labels, just two examples could in theory encode thousands of different objects. Karen Hao at MIT Technology Review<\/em> gets into more detail here<\/a>.<\/p>\n\n\n\n For everything AI can do, it\u2019s flawed at defending insidious attacks targeting its input data. Slight or seemingly random perturbations to a dataset\u2014often undetectable by the human eye\u2014can enormously alter the final output, something dubbed \u201cbrittle\u201d for an algorithm. Too abstract? An AI trained to recognize cancer from a slew of medical scans, annotated in yellow marker by a human doctor, could learn to associate \u201cyellow\u201d with \u201ccancer.\u201d A more malicious example is nefarious tampering. Stickers placed on a roadway can trick Tesla\u2019s<\/a> Autopilot system to mistake lanes and careen into oncoming traffic.<\/p>\n\n\n\n Brittleness requires AI to learn a certain level of flexibility, but sabotage\u2014or \u201cadversarial attacks\u201d\u2014is becoming an increasingly recognized problem. Here, hackers can change the AI\u2019s decision-making process with carefully-crafted inputs. When it comes to network security, medical diagnoses, or other high-stakes usage, building defense systems against these attacks is critical.<\/p>\n\n\n\n This year, a team from the University of Illinois proposed a powerful way<\/a> to make deep learning systems more resilient. They used an iterative approach, having two neural nets battle it out\u2014one for image recognition, and the other for generating adversarial attacks. Like a cat-and-mouse game, the \u201cenemy\u201d neural net tries to fool the computer vision network into recognizing things that are fictitious; the latter network fights back. While far from perfect, the study highlights one increasingly popular approach to make AI more resilient and trustworthy.<\/p>\n\n\n\n One of the most impressive algorithms this year is GPT-3<\/a>, a marvel by OpenAI that spits out eerily human-like language. Dubbed \u201cone of the most interesting and important AI systems ever produced<\/a>,\u201d GPT-3 is the third generation of an algorithm that produces writing so \u201cnatural\u201d that at a glance it\u2019s hard to decipher machine from human.<\/p>\n\n\n\n Yet GPT-3\u2019s language proficiency is, upon deeper inspection, just a thin veil of \u201cintelligence.\u201d Because it\u2019s trained on human language, it\u2019s also locked into the intricacies and limitations of our everyday phrases\u2014without any understanding of what they mean in the real world. It\u2019s akin to learning slang from Urban Dictionary instead of living it. An AI may learn to associate \u201crain\u201d with \u201ccats and dogs\u201d in all situations, gaining its inference from the common vernacular describing massive downpours.<\/p>\n\n\n\n One way to make GPT-3 or any natural language-producing AI smarter is to combine it with computer vision. Teaching language models to \u201csee\u201d is an increasingly popular area in AI research. The technique combines the strength of language with images. AI language models, including GPT-3, learn through a process called \u201cunsupervised training,\u201d which means they can parse patterns in data without explicit labels. In other words, they don\u2019t need a human to tell them grammatical rules or how words relate to one another, which makes it easier to scale any learning by bombarding the AI with tons of example text. Image models, on the other hand, better reflect our actual reality. However, these require manual labeling, which makes the process slower and more tedious.<\/p>\n\n\n\n Combining the two yields the best of both worlds. A robot that can \u201csee\u201d the world captures a sort of physicality\u2014or common sense\u2014that\u2019s missing from analyzing language alone. One study in 2020 smartly combined both approaches. They started with language, using a scalable approach to write captions for images based on the inner workings of GPT-3 (details here<\/a>). The takeaway is that the team was able to connect the physical world\u2014represented through images\u2014by linking it with language on how we describe the world.<\/p>\n\n\n\n Translation? A blind, deaf, and utterly quarantined AI learns a sort of common sense. For example, \u201ccats and dogs\u201d can just mean pets, rather than rain.<\/p>\n\n\n\n The trick is still mostly experimental, but it\u2019s an example of thinking outside the artificial confines of a particular AI domain. By combining the two areas\u2014natural language processing and computer vision\u2014it works better. Imagine an Alexa with common sense.<\/p>\n\n\n\n Speaking of thinking outside the box, DeepMind is among those experimenting with combining different approaches to AI into something more powerful. Take MuZero<\/a>, an Atari-smashing algorithm they released just before Christmas.<\/p>\n\n\n\n Unlike DeepMind\u2019s original Go, poker, and chess-slaying AI wizard, MuZero has another trick up its sleeve. It listens to no one, in that the AI doesn\u2019t start with previous knowledge of the game or decision-making processes. Rather, it learns without a rulebook, instead observing the game\u2019s environment\u2014akin to a novice human observing a new game. In this way, after millions of games, it doesn\u2019t just learn the rules, but also a more general concept of policies that could lead it to get ahead and evaluate its own mistakes in hindsight.<\/p>\n\n\n\n Sounds pretty human, eh? In AI vernacular, the engineers combined two different approaches, decision trees and a learned model, to make an AI great at planning winning moves. For now, it\u2019s only been shown to master games at a level similar to AlphaGo. But we can\u2019t wait to see what this sort of cross-fertilization of ideas in AI can lead to in 2021.<\/p>\n\n\n\n Image Credit:\u00a0Oleg Gamulinskiy<\/a>\u00a0from\u00a0Pixabay<\/a><\/em><\/p>\n\n\n\n <\/a>SHELLY FAN<\/a> Shelly Xuelai Fan is a neuroscientist-turned-science writer. She completed her PhD in neuroscience at the University of British Columbia, where she developed novel treatments for neurodegeneration. While studying biological brains, she became fascinated with AI and all things biotech. Following graduation, she moved to UCSF to study blood-based factors that rejuvenate aged brains. She is the …\u00a0Learn More<\/a><\/p>\n\n\n\n Original Article<\/a><\/strong><\/p>\n","protected":false},"excerpt":{"rendered":" If AI has anything to say to 2020, it\u2019s \u201cyou can\u2019t touch this.\u201d Last year may have severed our connections with the physical world, but in the digital realm, AI thrived. Take NeurIps, the crown jewel of AI conferences. While lacking the usual backdrop of the dazzling mountains of British Columbia or the beaches of Barcelona, […]\n","protected":false},"author":1,"featured_media":1075,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"episode_type":"","audio_file":"","podmotor_file_id":"","podmotor_episode_id":"","cover_image":"","cover_image_id":"","duration":"","filesize":"","filesize_raw":"","date_recorded":"","explicit":"","block":"","footnotes":""},"categories":[13],"tags":[],"series":[],"class_list":["post-1074","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-articulos-ingles"],"episode_featured_image":"https:\/\/singularityumexico.com\/wp-content\/uploads\/2021\/08\/abstract-4626114_1280-AI.jpg","episode_player_image":"https:\/\/singularityumexico.com\/wp-content\/uploads\/2023\/05\/11711533-1673157178559-89a95be153719-4-scaled.jpg","download_link":"","player_link":"","audio_player":false,"episode_data":{"playerMode":"dark","subscribeUrls":{"apple_podcasts":{"key":"apple_podcasts","url":"","label":"Apple Podcasts","class":"apple_podcasts","icon":"apple-podcasts.png"},"stitcher":{"key":"stitcher","url":"","label":"Stitcher","class":"stitcher","icon":"stitcher.png"},"google_podcasts":{"key":"google_podcasts","url":"","label":"Google Podcasts","class":"google_podcasts","icon":"google-podcasts.png"},"spotify":{"key":"spotify","url":"","label":"Spotify","class":"spotify","icon":"spotify.png"}},"rssFeedUrl":"https:\/\/singularityumexico.com\/en\/feed\/podcast\/the-feedback-loop-by-singularity","embedCode":"Greed: Less Than One-Shot Learning<\/h3>\n\n\n\n
Brittleness: A Method to Keep AI Hacker-Proof<\/h3>\n\n\n\n
AI Savant Syndrome: Learning Common Sense<\/h3>\n\n\n\n
Deep Learning Fatigue<\/h3>\n\n\n\n
\n\n\n\nAuthor:<\/h4>\n\n\n\n
2021 Could Be a Banner Year for AI\u2014If We Solve These 4 Problems<\/a><\/blockquote>