|
Next revision
|
Previous revision
|
featured_articles:evidence_against_current_methods_leading_to_human_level_artificial_intelligence [2022/09/21 07:37]
127.0.0.1 external edit |
featured_articles:evidence_against_current_methods_leading_to_human_level_artificial_intelligence [2022/10/19 02:01] (current)
katjagrace |
| // Published 12 August, 2019; last updated 07 October, 2019 // | // Published 12 August, 2019; last updated 07 October, 2019 // |
| |
| <HTML> | This is a list of published arguments that we know of that current methods in artificial intelligence will not lead to human-level AI. |
| <p>This is a list of published arguments that we know of that current methods in artificial intelligence will not lead to human-level AI.</p> | |
| </HTML> | |
| |
| |
| |
| |
| <HTML> | We take ‘current methods’ to mean techniques for engineering artificial intelligence that are already known, involving no “qualitatively new ideas”.((“It now seems possible that we could build ‘prosaic’ AGI, which can replicate human behavior but doesn’t involve qualitatively new ideas about ‘how intelligence works’”— Christiano, Paul. [[https://openreview.net/pdf?id=H18WqugAb|Prosaic AI Alignment]]. 2017. Medium. Accessed August 13 2019. https://ai-alignment.com/prosaic-ai-control-b959644d79c2.)) We have not precisely defined ‘current methods’. Many of the works we cite refer to currently //dominant// methods such as machine learning (especially deep learning) and reinforcement learning. |
| <p>We take ‘current methods’ to mean techniques for engineering artificial intelligence that are already known, involving no “qualitatively new ideas”.<span class="easy-footnote-margin-adjust" id="easy-footnote-1-1938"></span><span class="easy-footnote"><a href="#easy-footnote-bottom-1-1938" title='“It now seems possible that we could build ‘prosaic’ AGI, which can replicate human behavior but doesn’t involve qualitatively new ideas about ‘how intelligence works’”. &#8212; Christiano, Paul. <a href="https://openreview.net/pdf?id=H18WqugAb">&#8220;Prosaic AI Alignment&#8221;</a>. 2017. Medium. Accessed August 13 2019. https://ai-alignment.com/prosaic-ai-control-b959644d79c2.'><sup>1</sup></a></span> We have not precisely defined ‘current methods’. Many of the works we cite refer to currently <em>dominant</em> methods such as machine learning (especially deep learning) and reinforcement learning.</p> | |
| </HTML> | |
| |
| |
| <li><div class="li"><strong>Innate knowledge:</strong> Intelligence relies on prior knowledge which it is currently not feasible to embed via learning techniques, recapitulate via artificial evolution, or hand-specify. — Marcus (2018)<span class="easy-footnote-margin-adjust" id="easy-footnote-6-1938"></span><span class="easy-footnote"><a href="#easy-footnote-bottom-6-1938" title='Section 3.1, “Deep learning thus far is data hungry” &#8212; Marcus, Gary. 2018. <a href="https://arxiv.org/pdf/1801.00631.pdf">&#8220;Deep Learning: A Critical Appraisal&#8221;</a>. <em>arXiv</em>. Accessed August 12 2019. https://arxiv.org/abs/1801.00631.'><sup>6</sup></a></span></div></li> | <li><div class="li"><strong>Innate knowledge:</strong> Intelligence relies on prior knowledge which it is currently not feasible to embed via learning techniques, recapitulate via artificial evolution, or hand-specify. — Marcus (2018)<span class="easy-footnote-margin-adjust" id="easy-footnote-6-1938"></span><span class="easy-footnote"><a href="#easy-footnote-bottom-6-1938" title='Section 3.1, “Deep learning thus far is data hungry” &#8212; Marcus, Gary. 2018. <a href="https://arxiv.org/pdf/1801.00631.pdf">&#8220;Deep Learning: A Critical Appraisal&#8221;</a>. <em>arXiv</em>. Accessed August 12 2019. https://arxiv.org/abs/1801.00631.'><sup>6</sup></a></span></div></li> |
| <li><div class="li"><strong>Data hunger:</strong> Training a system to human level using current methods will require more data than we will be able to generate or acquire. — Marcus (2018)<span class="easy-footnote-margin-adjust" id="easy-footnote-7-1938"></span><span class="easy-footnote"><a href="#easy-footnote-bottom-7-1938" title='Section 3.1, “Deep learning thus far is data hungry” &#8212; Marcus, Gary. 2018. <a href="https://arxiv.org/pdf/1801.00631.pdf">&#8220;Deep Learning: A Critical Appraisal&#8221;</a>. <em>arXiv</em>. Accessed August 12 2019. https://arxiv.org/abs/1801.00631.'><sup>7</sup></a></span></div></li> | <li><div class="li"><strong>Data hunger:</strong> Training a system to human level using current methods will require more data than we will be able to generate or acquire. — Marcus (2018)<span class="easy-footnote-margin-adjust" id="easy-footnote-7-1938"></span><span class="easy-footnote"><a href="#easy-footnote-bottom-7-1938" title='Section 3.1, “Deep learning thus far is data hungry” &#8212; Marcus, Gary. 2018. <a href="https://arxiv.org/pdf/1801.00631.pdf">&#8220;Deep Learning: A Critical Appraisal&#8221;</a>. <em>arXiv</em>. Accessed August 12 2019. https://arxiv.org/abs/1801.00631.'><sup>7</sup></a></span></div></li> |
| | <li><div class="li"><strong>Long tails:</strong> There are a vast number of improbable situations which an AI system might not encounter during training on real-world scenarios, and which it is hence likely to perform much worse on. — Mitchell (2019)<span class="easy-footnote-margin-adjust" id="easy-footnote-8-1938"></span><span class="easy-footnote"><a href="#easy-footnote-bottom-8-1938" title='Mitchell, M., 2019. Artificial Intelligence: A Guide for Thinking Humans p. 125-127. Picador.'><sup>8</sup></a></span></div></li> |
| </ul> | </ul> |
| </HTML> | </HTML> |
