{"id":678,"date":"2022-05-25T13:11:01","date_gmt":"2022-05-25T13:11:01","guid":{"rendered":"https:\/\/wellai.health\/blog\/?p=678"},"modified":"2022-05-25T13:57:26","modified_gmt":"2022-05-25T13:57:26","slug":"googles-deepmind-made-a-breakthrough-with-its-gato-model-it-still-aint-even-close-to-a-human-brain","status":"publish","type":"post","link":"https:\/\/wellai.health\/blog\/googles-deepmind-made-a-breakthrough-with-its-gato-model-it-still-aint-even-close-to-a-human-brain\/","title":{"rendered":"Google\u2019s DeepMind made a breakthrough with its \u2018Gato\u2019 model. It still ain\u2019t even close to a human brain."},"content":{"rendered":"<p>Whew\u2026 The Armageddon is averted.\u00a0 The reports of the death of humankind are greatly exaggerated.\u00a0 For now\u2026<\/p>\n<p>&nbsp;<\/p>\n<p>Let\u2019s get to the story and offer our take.<\/p>\n<p>&nbsp;<\/p>\n<p><a href=\"https:\/\/www.deepmind.com\/publications\/a-generalist-agent\">Google\u2019s DeepMind claims it developed a perfect human-level AI with its new \u2018Gato\u2019 model<\/a>. \u00a0One of the developers of the model proclaimed \u201c<a href=\"https:\/\/twitter.com\/NandoDF\/status\/1525397036325019649?s=20&amp;t=5Jx_rN5x814c-OsS5YMI5w\">The Game is Over!<\/a>\u201d\u00a0 Humans have been defeated by robots.\u00a0 AI can do anything humans can.<\/p>\n<p>&nbsp;<\/p>\n<blockquote class=\"twitter-tweet\" data-width=\"550\" data-dnt=\"true\">\n<p lang=\"en\" dir=\"ltr\">Someone\u2019s opinion article. My opinion: It\u2019s all about scale now! The Game is Over! It\u2019s about making these models bigger, safer, compute efficient, faster at sampling, smarter memory, more modalities, INNOVATIVE DATA, on\/offline, \u2026 1\/N <a href=\"https:\/\/t.co\/UJxSLZGc71\">https:\/\/t.co\/UJxSLZGc71<\/a><\/p>\n<p>&mdash; Nando de Freitas \ud83c\udff3\ufe0f\u200d\ud83c\udf08 (@NandoDF) <a href=\"https:\/\/twitter.com\/NandoDF\/status\/1525397036325019649?ref_src=twsrc%5Etfw\">May 14, 2022<\/a><\/p><\/blockquote>\n<p><script async src=\"https:\/\/platform.twitter.com\/widgets.js\" charset=\"utf-8\"><\/script><\/p>\n<p>&nbsp;<\/p>\n<p>First, what does it mean to \u201cachieve human-level AI\u201d?\u00a0 Last time we checked a teenager could learn to drive in less than a week, but after over a decade we still don\u2019t have a self-driving car.<\/p>\n<p>&nbsp;<\/p>\n<p>Second, upon closer examination, the Gato model is cool.\u00a0 But it\u2019s nowhere close to what humans can do.<\/p>\n<p>&nbsp;<\/p>\n<p>Gato is truly a Jack of all trades, a master of none.\u00a0 It can do 604 tasks at the same time.\u00a0 That\u2019s cool.\u00a0 But almost every task it does, it does it at a mediocre level.\u00a0 Well, I guess that would make it human. \ud83d\ude0a<\/p>\n<p>&nbsp;<\/p>\n<p>Its chat dialogue is unnatural.\u00a0 Any human could guess she\/he is talking to a non-human.<\/p>\n<p>&nbsp;<\/p>\n<p>Gato\u2019s image captioning is poor and sometimes contradictory.<\/p>\n<p>&nbsp;<\/p>\n<p>The mediocreness of the Gato algorithm is almost by construction.\u00a0 It doesn\u2019t treat tokens differently whether they are words in a chat or movement vectors in a block-stacking exercise. It\u2019s all the same. It is one neural network that can work with multiple kinds of data to perform multiple kinds of tasks.<\/p>\n<p>&nbsp;<\/p>\n<p>Another Gato\u2019s weakness is its lack of adaptability.\u00a0 Just like GPT-3, Gato is a transformer model.\u00a0 It is very general, also by construction.\u00a0 However, it is weak in transforming learning from one situation into another, like humans do.<\/p>\n<p>&nbsp;<\/p>\n<p>We must also talk about the ethical issues with Gato.\u00a0 As you know, we\u2019ve done a significant amount of work on best practices of AI in healthcare, as part of <a href=\"https:\/\/www.ifcc.org\/ifcc-emerging-technologies-division\/etd-working-groups\/wg-aigd\/\">IFCC\u2019s Working Group on Artificial Intelligence and Genomic Diagnostics (WG-AIGD)<\/a>.<\/p>\n<p>&nbsp;<\/p>\n<p>We strongly believe health equity and unbiasedness are critical parts of the future of the digital health industry.\u00a0 The authors of the DeepMind paper state, \u201cAdditionally, while cross-domain knowledge transfer is often a goal in ML research, it could create unexpected and undesired outcomes if certain behaviors (e.g. arcade game fighting) are transferred to the wrong context\u2026 The ethics and safety considerations of knowledge transfer may require substantial new research as generalist systems advance.\u201d<\/p>\n<p>&nbsp;<\/p>\n<p>DeepMind seems to be doing everything to spite OpenAI.\u00a0 We get it.\u00a0 Healthy competition is always good.\u00a0 But shouldn\u2019t a new model at least address an \u201cold\u201d model\u2019s issues?\u00a0 18 months ago DeepMind critiqued <a href=\"https:\/\/www.theregister.com\/2020\/10\/28\/gpt3_medical_chatbot_experiment\/\">GPT-3 for making simple errors and even suggesting to \u201ckill\u201d someone<\/a>.<\/p>\n<p>&nbsp;<\/p>\n<p>It\u2019s time for DeepMind now to put its own model to a rigorous test.<\/p>\n<p>&nbsp;<\/p>\n<p>We do applaud the fact that Gato is much leaner than GPT-3, using \u201conly\u201d 1.18 billion parameters in its main version vs GPT-3\u2019s 175 billion parameters.\u00a0 However, the cost is a massive issue, as of now.\u00a0 Only corporations could afford GPT-3 and Gato.<\/p>\n<p>&nbsp;<\/p>\n<p>It\u2019s not easy.\u00a0 But we are going to get \u201cthere\u201d one day.\u00a0 That day is not today&#8230;<\/p>\n<p>&nbsp;<\/p>\n<p>Humankind is still going strong.<\/p>\n<p>&nbsp;<\/p>\n<p>Perhaps a \u2018Le\u00f3n\u2019 algorithm would beat humans one day. \ud83d\ude09<\/p>\n<p>&nbsp;<\/p>\n<p>WellAI Team<\/p>\n<p><a href=\"https:\/\/wellai.health\/\">wellai.health<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Whew\u2026 The Armageddon is averted.\u00a0 The reports of the death of humankind are greatly exaggerated.\u00a0 For now\u2026<br \/>\nLet\u2019s get to the story and offer our take.<\/p>\n","protected":false},"author":3,"featured_media":679,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[180,179],"tags":[143,259,252,261,241,262,254,193,229,263,260],"jetpack_featured_media_url":"https:\/\/i0.wp.com\/wellai.health\/blog\/wp-content\/uploads\/2022\/05\/Googles-DeepMind-made-a-breakthrough-with-its-\u2018Gato-model.-It-still-aint-even-close-to-a-human-brain.-e1653484249482.jpg?fit=698%2C400&ssl=1","jetpack_sharing_enabled":true,"jetpack_likes_enabled":true,"_links":{"self":[{"href":"https:\/\/wellai.health\/blog\/wp-json\/wp\/v2\/posts\/678"}],"collection":[{"href":"https:\/\/wellai.health\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/wellai.health\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/wellai.health\/blog\/wp-json\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/wellai.health\/blog\/wp-json\/wp\/v2\/comments?post=678"}],"version-history":[{"count":4,"href":"https:\/\/wellai.health\/blog\/wp-json\/wp\/v2\/posts\/678\/revisions"}],"predecessor-version":[{"id":683,"href":"https:\/\/wellai.health\/blog\/wp-json\/wp\/v2\/posts\/678\/revisions\/683"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/wellai.health\/blog\/wp-json\/wp\/v2\/media\/679"}],"wp:attachment":[{"href":"https:\/\/wellai.health\/blog\/wp-json\/wp\/v2\/media?parent=678"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/wellai.health\/blog\/wp-json\/wp\/v2\/categories?post=678"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/wellai.health\/blog\/wp-json\/wp\/v2\/tags?post=678"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}