added wise AI paper

cicl-stanford · Nov 6, 2024 · 61792f0 · 61792f0
1 parent e3cddbf
commit 61792f0
Show file tree

Hide file tree

Showing 16 changed files with 726 additions and 49 deletions.
diff --git a/content/publication/johnson2024wise.md b/content/publication/johnson2024wise.md
@@ -0,0 +1,33 @@
++++
+# 0 -> 'Forthcoming',
+# 1 -> 'Preprint',
+# 2 -> 'Journal',
+# 3 -> 'Conference Proceedings',
+# 4 -> 'Book chapter',
+# 5 -> 'Thesis'
+
+title = "Imagining and building wise machines: The centrality of AI metacognition"
+date = "2024-11-06"
+authors = ["S. G. B. Johnson","A. Karimi","Y. Bengio","N. Chater","T. Gerstenberg","K. Larson","S. Levine","M. Mitchell","B. Schölkopf","I. Grossmann"]
+publication_types = ["1"]
+publication_short = "_arXiv_"
+publication = "Johnson, S. G. B., Karimi, A., Bengio, Y., Chater, N., Gerstenberg, T., Larson, K., Levine, S., Mitchell, M., Schölkopf, B., Grossmann, I. (2024). Imagining and building wise machines: The centrality of AI metacognition. _arXiv_."
+abstract = "Recent advances in artificial intelligence (AI) have produced systems capable of increasingly sophisticated performance on cognitive tasks. However, AI systems still struggle in critical ways: unpredictable and novel environments (robustness), lack transparency in their reasoning (explainability), face challenges in communication and commitment (cooperation), and pose risks due to potential harmful actions (safety). We argue that these shortcomings stem from one overarching failure: AI systems lack wisdom. Drawing from cognitive and social sciences, we define wisdom as the ability to navigate intractable problems---those that are ambiguous, radically uncertain, novel, chaotic, or computationally explosive---through effective task-level and metacognitive strategies. While AI research has focused on task-level strategies, metacognition---the ability to reflect on and regulate one's thought processes---is underdeveloped in AI systems. In humans, metacognitive strategies such as recognizing the limits of one's knowledge, considering diverse perspectives, and adapting to context are essential for wise decision-making. We propose that integrating metacognitive capabilities into AI systems is crucial for enhancing their robustness, explainability, cooperation, and safety. By focusing on developing wise AI, we suggest an alternative to aligning AI with specific human values---a task fraught with conceptual and practical difficulties. Instead, wise AI systems can thoughtfully navigate complex situations, account for diverse human values, and avoid harmful actions. We discuss potential approaches to building wise AI, including benchmarking metacognitive abilities and training AI systems to employ wise reasoning. Prioritizing metacognition in AI research will lead to systems that act not only intelligently but also wisely in complex, real-world situations."
+image_preview = ""
+selected = false
+projects = []
+#url_pdf = "papers/johnson2024wise.pdf"
+url_preprint = "https://arxiv.org/abs/2411.02478"
+url_code = ""
+url_dataset = ""
+url_slides = ""
+url_video = ""
+url_poster = ""
+url_source = ""
+#url_custom = [{name = "Github", url = ""}]
+math = true
+highlight = true
+[header]
+# image = "publications/johnson2024wise.png"
+caption = ""
++++
diff --git a/docs/404.html b/docs/404.html
@@ -237,6 +237,10 @@ <h1>Page not found</h1>
 
   <h2>Publications</h2>
 
+  <ul>
+    <li><a href="https://cicl.stanford.edu/publication/johnson2024wise/">Imagining and building wise machines: The centrality of AI metacognition</a></li>
+  </ul>
+
   <ul>
     <li><a href="https://cicl.stanford.edu/publication/prinzing2024purpose/">From Artifacts to Human Lives: Investigating the Domain-Generality of Judgments about Purposes</a></li>
   </ul>
@@ -253,10 +257,6 @@ <h2>Publications</h2>
     <li><a href="https://cicl.stanford.edu/publication/beller2024causation/">Causation, Meaning, and Communication</a></li>
   </ul>
 
-  <ul>
-    <li><a href="https://cicl.stanford.edu/publication/gandhi2024affective/">Human-like Affective Cognition in Foundation Models</a></li>
-  </ul>
-
 
 
 

diff --git a/docs/bibtex/cic_papers.bib b/docs/bibtex/cic_papers.bib
@@ -1,13 +1,22 @@
 %% This BibTeX bibliography file was created using BibDesk.
 %% https://bibdesk.sourceforge.io/
 
-%% Created for Tobias Gerstenberg at 2024-10-26 12:42:26 -0700 
+%% Created for Tobias Gerstenberg at 2024-11-06 11:16:38 -0600 
 
 
 %% Saved with string encoding Unicode (UTF-8) 
 
 
 
+@article{johnson2024wise,
+	abstract = {Recent advances in artificial intelligence (AI) have produced systems capable of increasingly sophisticated performance on cognitive tasks. However, AI systems still struggle in critical ways: unpredictable and novel environments (robustness), lack transparency in their reasoning (explainability), face challenges in communication and commitment (cooperation), and pose risks due to potential harmful actions (safety). We argue that these shortcomings stem from one overarching failure: AI systems lack wisdom. Drawing from cognitive and social sciences, we define wisdom as the ability to navigate intractable problems---those that are ambiguous, radically uncertain, novel, chaotic, or computationally explosive---through effective task-level and metacognitive strategies. While AI research has focused on task-level strategies, metacognition---the ability to reflect on and regulate one's thought processes---is underdeveloped in AI systems. In humans, metacognitive strategies such as recognizing the limits of one's knowledge, considering diverse perspectives, and adapting to context are essential for wise decision-making. We propose that integrating metacognitive capabilities into AI systems is crucial for enhancing their robustness, explainability, cooperation, and safety. By focusing on developing wise AI, we suggest an alternative to aligning AI with specific human values---a task fraught with conceptual and practical difficulties. Instead, wise AI systems can thoughtfully navigate complex situations, account for diverse human values, and avoid harmful actions. We discuss potential approaches to building wise AI, including benchmarking metacognitive abilities and training AI systems to employ wise reasoning. Prioritizing metacognition in AI research will lead to systems that act not only intelligently but also wisely in complex, real-world situations.},
+	author = {Johnson, Samuel G B and Karimi, Amir-Hossein and Bengio, Yoshua and Chater, Nick and Gerstenberg, Tobias and Larson, Kate and Levine, Sydney and Mitchell, Melanie and Sch{\"o}lkopf, Bernhard and Grossmann, Igor},
+	date-added = {2024-11-06 11:16:21 -0600},
+	date-modified = {2024-11-06 11:16:21 -0600},
+	journal = {arXiv},
+	title = {{Imagining and building wise machines: The centrality of AI metacognition}},
+	year = {2024}}
+
 @article{jin2024marple,
 	abstract = {Reconstructing past events requires reasoning across long time horizons. To figure out what happened, we need to use our prior knowledge about the world and human behavior and draw inferences from various sources of evidence including visual, language, and auditory cues. We introduce MARPLE, a benchmark for evaluating long-horizon inference capabilities using multi-modal evidence. Our benchmark features agents interacting with simulated households, supporting vision, language, and auditory stimuli, as well as procedurally generated environments and agent behaviors. Inspired by classic ``whodunit'' stories, we ask AI models and human participants to infer which agent caused a change in the environment based on a step-by-step replay of what actually happened. The goal is to correctly identify the culprit as early as possible. Our findings show that human participants outperform both traditional Monte Carlo simulation methods and an LLM baseline (GPT-4) on this task. Compared to humans, traditional inference models are less robust and performant, while GPT-4 has difficulty comprehending environmental changes. We analyze what factors influence inference performance and ablate different modes of evidence, finding that all modes are valuable for performance. Overall, our experiments demonstrate that the long-horizon, multimodal inference tasks in our benchmark present a challenge to current models. Project website: https: //marple-benchmark.github.io/.},
 	annote = {Comment: NeurIPS 2024. First two authors contributed equally. Project page: https://marple-benchmark.github.io/},

diff --git a/docs/home/index.xml b/docs/home/index.xml
@@ -15,7 +15,7 @@
     <item>
       <title>Causality in Cognition Lab</title>
       <link>https://cicl.stanford.edu/home/home/</link>
-      <pubDate>Sun, 15 Oct 2017 00:00:00 -0700</pubDate>
+      <pubDate>Sun, 15 Oct 2017 00:00:00 -0500</pubDate>
 
       <guid>https://cicl.stanford.edu/home/home/</guid>
       <description>The Causality in Cognition Lab at Stanford University studies the role of causality in our understanding of the world and of each other.</description>
@@ -53,7 +53,7 @@ We are looking to hire a lab manager to start in the fall of 2018. For informati
     <item>
       <title>Selected&lt;br&gt;publications</title>
       <link>https://cicl.stanford.edu/home/publications_selected/</link>
-      <pubDate>Wed, 20 Apr 2016 00:00:00 -0700</pubDate>
+      <pubDate>Wed, 20 Apr 2016 00:00:00 -0500</pubDate>
 
       <guid>https://cicl.stanford.edu/home/publications_selected/</guid>
       <description></description>

diff --git a/docs/index.html b/docs/index.html
@@ -110,7 +110,7 @@
   <meta property="og:description" content="">
   <meta property="og:locale" content="en-us">
 
-  <meta property="og:updated_time" content="2024-10-26T00:00:00&#43;00:00">
+  <meta property="og:updated_time" content="2024-11-06T00:00:00&#43;00:00">
 
 
 

diff --git a/docs/index.xml b/docs/index.xml
@@ -6,9 +6,18 @@
     <generator>Hugo -- gohugo.io</generator>
     <language>en-us</language>
     <copyright>&amp;copy; 2024 Tobias Gerstenberg</copyright>
-    <lastBuildDate>Sat, 26 Oct 2024 00:00:00 +0000</lastBuildDate>
+    <lastBuildDate>Wed, 06 Nov 2024 00:00:00 +0000</lastBuildDate>
     <atom:link href="/" rel="self" type="application/rss+xml" />
 
+    <item>
+      <title>Imagining and building wise machines: The centrality of AI metacognition</title>
+      <link>https://cicl.stanford.edu/publication/johnson2024wise/</link>
+      <pubDate>Wed, 06 Nov 2024 00:00:00 +0000</pubDate>
+
+      <guid>https://cicl.stanford.edu/publication/johnson2024wise/</guid>
+      <description></description>
+    </item>
+
     <item>
       <title>From Artifacts to Human Lives: Investigating the Domain-Generality of Judgments about Purposes</title>
       <link>https://cicl.stanford.edu/publication/prinzing2024purpose/</link>
@@ -135,14 +144,5 @@
       <description></description>
     </item>
 
-    <item>
-      <title>Resource-rational moral judgment</title>
-      <link>https://cicl.stanford.edu/publication/wu2024resource/</link>
-      <pubDate>Wed, 01 May 2024 00:00:00 +0000</pubDate>
-
-      <guid>https://cicl.stanford.edu/publication/wu2024resource/</guid>
-      <description></description>
-    </item>
-
   </channel>
 </rss>
diff --git a/docs/member/tobias_gerstenberg/index.html b/docs/member/tobias_gerstenberg/index.html
@@ -356,6 +356,45 @@ <h2 id="publications">Publications</h2>
 
 
        <div class="pub-list-item" style="margin-bottom: 1rem" itemscope itemtype="http://schema.org/CreativeWork">
+  <span itemprop="author">
+    S. G. B. Johnson, A. Karimi, Y. Bengio, N. Chater, T. Gerstenberg, K. Larson, S. Levine, M. Mitchell, B. Schölkopf, I. Grossmann</span>
+
+    (2024).
+
+  <a href="https://cicl.stanford.edu/publication/johnson2024wise/" itemprop="name">Imagining and building wise machines: The centrality of AI metacognition</a>.
+  <em>arXiv</em>.
+
+
+
+
+  <p>
+
+
+
+
+<a class="btn btn-outline-primary my-1 mr-1 btn-sm" href="https://arxiv.org/abs/2411.02478" target="_blank" rel="noopener">
+  Preprint
+</a>
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+  </p>
+
+</div>
+<div class="pub-list-item" style="margin-bottom: 1rem" itemscope itemtype="http://schema.org/CreativeWork">
   <span itemprop="author">
     M. Prinzing, D. Rose, S. Zhang, E. Tu, A. Concha, M. Rea, J. Schaffer, T. Gerstenberg, J. Knobe</span>
 

diff --git a/docs/publication/index.html b/docs/publication/index.html
@@ -1610,6 +1610,19 @@ <h1>Publications</h1>
 
 
 
+
+
+
+
+
+
+
+
+
+
+
+
+
 
 
 
@@ -1746,6 +1759,57 @@ <h1>Publications</h1>
 
 
 
+				<div class='grid-sizer col-md-12 isotope-item pubtype-1 year-2024 author-'>
+
+						<div class="pub-list-item" style="margin-bottom: 1rem" itemscope itemtype="http://schema.org/CreativeWork">
+  <span itemprop="author">
+    S. G. B. Johnson, A. Karimi, Y. Bengio, N. Chater, T. Gerstenberg, K. Larson, S. Levine, M. Mitchell, B. Schölkopf, I. Grossmann</span>
+
+    (2024).
+
+  <a href="https://cicl.stanford.edu/publication/johnson2024wise/" itemprop="name">Imagining and building wise machines: The centrality of AI metacognition</a>.
+  <em>arXiv</em>.
+
+
+
+
+  <p>
+
+
+
+
+<a class="btn btn-outline-primary my-1 mr-1 btn-sm" href="https://arxiv.org/abs/2411.02478" target="_blank" rel="noopener">
+  Preprint
+</a>
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+  </p>
+
+</div>
+
+
+				</div>
+
+
+
+
+
+
+
 				<div class='grid-sizer col-md-12 isotope-item pubtype-2 year-2024 author-'>
 
 						<div class="pub-list-item" style="margin-bottom: 1rem" itemscope itemtype="http://schema.org/CreativeWork">

diff --git a/docs/publication/index.xml b/docs/publication/index.xml
@@ -12,6 +12,15 @@
 	<atom:link href="https://cicl.stanford.edu/publication/index.xml" rel="self" type="application/rss+xml" />
 
 
+    <item>
+      <title>Imagining and building wise machines: The centrality of AI metacognition</title>
+      <link>https://cicl.stanford.edu/publication/johnson2024wise/</link>
+      <pubDate>Wed, 06 Nov 2024 00:00:00 +0000</pubDate>
+
+      <guid>https://cicl.stanford.edu/publication/johnson2024wise/</guid>
+      <description></description>
+    </item>
+
     <item>
       <title>From Artifacts to Human Lives: Investigating the Domain-Generality of Judgments about Purposes</title>
       <link>https://cicl.stanford.edu/publication/prinzing2024purpose/</link>