2: The impact of machines that "learn" and produce

<p>The results from machine learning have been getting better and better and the results seen so far from <a href="https://openai.com">OpenAI</a>'s <a href="https://en.wikipedia.org/wiki/OpenAI#GPT-3">GPT-3 model</a> look stunningly good. But unlike <a href="https://github.com/openai/gpt-2">GPT-2</a> (which was publicly released under a free license), so far <a href="https://openai.com/blog/openai-api/">GPT-3 is accessible via API-only</a>. What's the reasoning and possible impact of that decision? For that matter, what kind of impacts could machine learning advancements make on FOSS, programming in general, art production, and civic society?</p><p><strong>Links:</strong></p><ul><li><p>The <a href="https://maraoz.com/2020/07/18/openai-gpt3/">OpenAI's GPT-3 may be the biggest thing since bitcoin</a> article</p><ul><li>The <a href="https://news.ycombinator.com/item?id=23886503">quoted comment on Hacker News</a></li></ul></li><li><a href="https://twitter.com/edleonklinger/status/1284251420544372737">Auto-generation of legalese</a> and <a href="https://twitter.com/sharifshameem/status/1282676454690451457">auto-web-design</a> GPT-3 demos</li><li><a href="https://github.com/openai/gpt-2">GPT-2</a></li><li><a href="https://openai.com/blog/openai-api/">GPT-3's API and FAQ page</a></li><li><a href="https://www.tensorflow.org/">Tensorflow</a> and <a href="https://pytorch.org/">PyTorch</a></li><li><a href="https://en.wikipedia.org/wiki/Artificial_neural_networks">(Artificial) neural networks</a> and <a href="https://en.wikipedia.org/wiki/Machine_learning">machine learning</a></li><li><a href="https://thispersondoesnotexist.com/">https://thispersondoesnotexist.com/</a></li><li>Google's <a href="https://deepmind.com/">Deepmind</a> and <a href="https://deepmind.com/blog/article/Agent57-Outperforming-the-human-Atari-benchmark">Agent57</a> (be sure to watch the <a href="https://www.youtube.com/watch?list=PLqYmG7hTraZCHS3JLle_kxwNvImpYVq4z&amp;time_continue=1&amp;v=luZm3jmwGwI&amp;feature=emb_logo">Agent57 videos</a>, they're's impressive)</li><li>Mozilla's <a href="https://voice.mozilla.org/">Common Voice</a> project</li><li><a href="https://deepmind.com/research/case-studies/alphago-the-story-so-far">AlphaGo</a></li><li><a href="https://en.wikipedia.org/wiki/Naive_Bayes_spam_filtering">Bayesian spam filters</a>; see also Paul Graham's highly influential <a href="http://paulgraham.com/spam.html">a plan for spam</a> writeup</li><li><a href="https://en.wikipedia.org/wiki/Markov_chain">Markov chains</a> (we miss you, <a href="https://www.x11r5.com/">X11R5</a>...)</li><li>The <a href="http://www.elsewhere.org/journal/pomo/">Postmodernist Essay Generator</a></li><li><a href="https://en.wikipedia.org/wiki/Postmodernism">Postmodernism</a></li><li>Neural networks' difficulties in explaining &quot;why they did that&quot; and an overview of attempts to make things better: <a href="https://arxiv.org/pdf/1806.00069">An Overview of Interpretability of Machine Learning</a></li><li>Chris had a <a href="https://dustycloud.org/blog/sussman-on-ai/">conversation with Gerald Sussman about AI that was related to the above and influential on them</a>. &quot;If an AI driven car drives off the side of the road, I want to know why it did that. I <em>could</em> take the software developer to court, but I would much rather take the AI to court.&quot;</li><li>The Propagator Model (by Alexey Radul and Gerald Jay Sussman, largely): <a href="https://groups.csail.mit.edu/mac/users/gjs/propagators/">Revised Report on the Propagator Model</a>. See also: <a href="https://www.youtube.com/watch?v=O3tVctB_VSU">We Really Don't Know How to Compute!</a></li><li><a href="http://www.threepanelsoul.com/">Three Panel Soul</a>'s <a href="http://www.threepanelsoul.com/comic/recursion">Recursion</a> comic (cut from this episode, but we also originally mentioned their <a href="http://www.threepanelsoul.com/comic/tethics">Techics</a> comic which is definitely relevant though)</li><li><a href="https://en.wikipedia.org/wiki/Surrealism">Surrealism</a>, <a href="https://en.wikipedia.org/wiki/Abstract_impressionism">Abstract Expressionism</a>, <a href="https://en.wikipedia.org/wiki/Impressionism">Impressionism</a>, and the <a href="https://en.wikipedia.org/wiki/Realism_(art_movement)">Realism</a> movement (Obviously there's also a lot more to say about these art movements than just lumping them as a reaction to photography but... only so much time on an episode.)</li><li><a href="https://play.aidungeon.io/">AI Dungeon 2</a> (nonfree, though you can play it in your browser)</li><li>Episode of <a href="https://ludology.libsyn.com/">Ludology</a> about <a href="https://ludology.libsyn.com/gametek-classic-149-procedural-narrative-generation">procedural narrative generation</a></li><li><a href="https://www.bates.edu/writing/2018/03/12/implicit-bias-and-the-teaching-of-writing/">Implicit Bias and the Teaching of Writing</a></li><li><p>Machine learning's tendency to inherit biases</p><ul><li><a href="https://www.theguardian.com/inequality/2017/aug/08/rise-of-the-racist-robots-how-ai-is-learning-all-our-worst-impulses">Rise of the racist robots -- how AI is learning all our worst impulses</a></li><li><a href="https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing">Machine bias</a> (and its use in deciding court cases)</li><li><a href="https://algorithmwatch.org/en/story/google-vision-racism/">Google's Vision AI producing racist results</a></li><li><a href="https://www.wired.com/story/when-it-comes-to-gorillas-google-photos-remains-blind/">When It Comes to Gorillas, Google Photos Remains Blind</a> (Content warning: this is due to an extremely harmful form of synthesized racism from the biases in the datasets Google has used)</li><li><a href="https://www.technologyreview.com/2020/07/17/1005396/predictive-policing-algorithms-racist-dismantled-machine-learning-bias-criminal-justice/">Predictive policing algorithms are racist. They need to be dismantled.</a></li></ul></li><li><a href="https://arxiv.org/abs/2007.08794">Discovering Reinforcement Learning Algorithms</a> and the subdiscipline of <a href="https://en.wikipedia.org/wiki/Meta_learning_(computer_science)">Learning to Learn</a></li><li><a href="https://www.youtube.com/watch?v=BfxlgHBaxEU">Hayao Miyazaki's criticism of an AI demonstration not considering its impact</a></li></ul>