To facilitate the exploration of the issues and expertises mobilised by the Global Forum on Artificial Intelligence for Humanity, we have created an atlas of the recent scientific literature on AI for humanity. Starting from a query targeting both the societal implications of AI technologies and the use of AI in human and social sciences (cf. query), we collected a corpus of more than 23 thousand bibliographic records corresponding to journal articles and conference proceedings published on these topics in the last five years. Using co-citation techniques and force-directed network layout, we created a base map of the relevant scientific literature and use it to locate the keywords, subject areas and institutions appearing in our bibliographic corpus (cf. protocol).
The results of this work revealed the great diversity of matters of reflections to be addressed during the Forum (and by the International Panel on AI that it prefigures), but also the seamless connection between technical questions related to advanced techniques of computation of robotics innovation and social and human implications of their implementation (cf. interpretation).

Click here for a high-resolution PDF of the maps   (to download, right click and save the target file)
 

Presentation and interpretation

The Global Forum on AI for Humanity marks an important step in the reflection on the social and human implications of the scientific and industrial developments associated with artificial intelligence. This reflection comes with the recognition that computer and information sciences are not the only disciplines concerned by AI nor the only ones that should help in steering its course. Inviting diversity of experts to the discussion table, the Global Forum (and the International Panel on AI that it prefigures) encourages a broad discussion on the political dimensions of AI and on its use in human and social sciences.

To foster the exploration of this diversity of issues and expertises, our atlas proposes a mapping of the scientific literature related to AI for humanity. Starting from a query targeting both the research on the societal questions connected to advanced computation technologies and their use in human and social disciplines (cf. query) , we collected a corpus of more than 23 thousand bibliographic records corresponding to the articles and conference proceedings published on these topics in the last five years. The first step in our analysis consists in extracting the references present in three or more records of our corpus (note that, while these records have all been issued in the last five years, there is no temporal restriction on the references they cite). We then build a network of references connected if cited together by at least two records in the corpus. We apply a force-directed layout to obtain the network of co-cited references that constitutes the base map of our atlas. Finally, we project on this base map the article keywords, subject areas and institutions appearing in our bibliographic corpus, sizing them according to the number of occurrences. (cf. protocol)

The resulting maps (cf. maps) display three main regions. The first, located at the left of the map, focusses on different types of interactions between humans and digital technologies, spanning from the robots (at the bottom of the diagram), to the smart cities (upper-left corner) and business intelligence (slightly to the right of the previous cluster), passing through debates on work automation, data privacy and more generally critical data studies (middle-left of the diagram). The right part of the map is occupied by technical questions related to machine learning, neural networks, genetic algorithms, etc., but also suicide prevention and e-learning. Bridging the left and right regions, a dense centre occupied by digital media and related conversations about data journalism, filter bubbles, misinformation, but also more technical discussion about how data from electronic media can be used for social research through digital methods and text mining.

Protocol for generating the maps

  1. Design a query capable of identifying
    • the research on the societal questions raised by artificial intelligence techniques
    • and the use of AI techniques in human and social sciences.
  2. Use the query to extract from ISI Web of Knowledge the ≈ 23.000 bibliographic records related to artificial intelligence for humanity and published in the last 5 years
    • ≈ 15.000 journal articles
    • ≈ 9.000 papers in conference proceedings 
  3. Extract the references present in the corpus keeping only the ≈ 28.000 that
    • are cited in 3 or more records of the corpus
    • are cited together in 2 or more records
    • (NB. while the corpus contains only records issued in the last five years, there is no temporal restriction on the references they cite)
  4. Build a network of references connected if they co-appear in the bibliography of at least 2 records with a weight proportional to the number of co-citations.
  5. Spatialise the network with a force-directed layout and fix the position of the nodes-references to obtain the base map of the atlas.
  6. Project on this base map
    • the ≈ 1200 article keywords appearing in > 3 records of the corpus
    • the ≈ 130 subject areas appearing in > 1 records 
    • the ≈ 450 institutions appearing in > 25 records
      (generic institutions, e.g. “business school” are removed)
    • (NB. nodes are positioned according to the references with which they co-appear and sized according to their number of occurrences)

Query

AI synonyms & techniques (1) OR AI synonyms & techniques (1)
AND AND
social & human issues (2) social & human disciplines (3)
    AND
document type = article  OR  proceedings published between 2015 and 2019
(1)   Title, Abstract or Keywords =
("abstraction and micro-operator*" OR "action recognition" OR "activit* recognition" OR "adversarial learning" OR "adversarial network*" OR "agent-based simulation*" OR "agent* communication" OR "agent* model" OR "agent* societ*" OR "ai-chipset*" OR "algorithmic game theory" OR "algorithms" OR "answer set programming" OR "ant colony optimisation" OR "ant colony optimization" OR "artificial evolution" OR "artificial intelligence" OR "association rule learning" OR "automated planning" OR "automated reasoning" OR "automated scheduling" OR "automation" OR "autonomous car*" OR "autonomous system*" OR "autonomous vehicle*" OR "bayesian method*" OR "bayesian network*" OR "big data" OR "biometric" OR "BOT" OR "BOTs" OR "case-based reasoning" OR "character recognition" OR "cognitive computing" OR "cognitive robotic" OR "combinatorial search" OR "computational complexity of reasoning" OR "computational control theory" OR "computational models of argument*" OR "computational photography" OR "computational social choice*" OR "computer vision" OR "constraint* programming" OR "constraint* satisfaction problem*" OR "continuous space search" OR "cooperative game*" OR "coordination and cooperation" OR "cost-sensitive learning" OR "data mining" OR "data stream" OR "deep learning" OR "dependable robot*" OR "description logic" OR "developmental robotic" OR "dimensionality reduction" OR "discrete space search" OR "distributed ai" OR "distributed constraint*" OR "distributed search" OR "driverless car" OR "driverless vehicle" OR "dynamic programming" OR "ensemble method*" OR "ensemble model*" OR "evolutionary algorithm*" OR "evolutionary programming" OR "evolutionary robotic" OR "expert system*" OR "face recognition" OR "feature* selection" OR "fuzzy logic" OR "fuzzy system*" OR "game tree search" OR "genetic algorithm*" OR "genetic programming" OR "gesture* recognition" OR "handwriting detection" OR "hardware for ai" OR "heuristic function construction" OR "heuristic search" OR "hierarchical planning" OR "image segmentation" OR "inference engine*" OR "instance segmentation" OR "instance-based learning" OR "intelligent agent*" OR "intelligent system*" OR "kernel method*" OR "knowledge based system*" OR "knowledge discovery" OR "knowledge engineering" OR "knowledge extraction" OR "knowledge representation" OR "knowledge-based learning" OR "latent representation*" OR "learning generative model*" OR "learning graphical model*" OR "learning in robotic" OR "learning sparse model*" OR "lexical semantic" OR "logic programming" OR "logics for knowledge representation" OR "machine learning" OR "machine translation" OR "machine vision" OR "manifold learning" OR "markov decision process*" OR "markov model*" OR "massive data analysis" OR "maxsat" OR "minsat" OR "mobile agent*" OR "model-based reasoning" OR "motion estimation" OR "motion path planning" OR "motion tracking" OR "multi-agent* learning" OR "multi-agent* planning" OR "multi-agent* system*" OR "multi-robot system*" OR "multi-task learning" OR "naive bayes" OR "natural language generation" OR "natural language processing" OR "natural language semantic" OR "natural language summarisation" OR "natural language summarization" OR "neural network*" OR "neuromorphic architecture" OR "neuromorphic computing" OR "neuromorphic engineering" OR "nlp" OR "non-classical logics for knowledge representation" OR "non-monotonic reasoning" OR "noncooperative game*" OR "nonprobabilistic model*" OR "ontology engineering" OR "optical character recognition" OR "optimisation for learning" OR "optimization for learning" OR "pattern search" OR "planning algorithm" OR "planning under uncertainty" OR "planning with incomplete information" OR "pomdp" OR "predictive analytics" OR "preference* modeling" OR "preference* modelling" OR "probabilistic graphical model*" OR "probabilistic inference*" OR "probabilistic reasoning" OR "random forest*" OR "randomized search" OR "recommendation system*" OR "recommender system*" OR "regression tree*" OR "reinforcement learning" OR "robot fleet*" OR "robot interaction*" OR "robot learning" OR "robot planning" OR "rule* learning" OR "sat analysis" OR "sat application*" OR "sat evaluation*" OR "sat solver*" OR "sat tool*" OR "satisfiability" OR "search engine*" OR "search in image bank*" OR "search in planning and scheduling" OR "search in video bank*" OR "self-driving car*" OR "semantic network*" OR "semantic ontolog*" OR "semantic web" OR "semi-supervised learning" OR "sensor network*" OR "sentiment analysis" OR "sequential decision making" OR "sequential learning" OR "sequential processing" OR "serial learning" OR "social robot*" OR "soft computing" OR "spatiotemporal reconstruction" OR "speaker recognition" OR "speech identification" OR "speech processing" OR "speech recognition" OR "speech synthesi" OR "speech verification" OR "speech-to-speech" OR "supervised learning" OR "support vector* machine*" OR "support vector*" OR "swarm intelligence" OR "swarm optimisation" OR "swarm optimization" OR "temporal and hybrid planning" OR "tensor processing" OR "text classification" OR "text mining" OR "text recognition" OR "unsupervised learning" OR "video segmentation" OR "visual servoing" OR “artificial neural network*” OR “autoencoder*” OR “backprop*” OR “boltzmann machine*” OR “convolutional neural network*” OR “deductive database*” OR “deep belief network*” OR “deep learning” OR “deep neural network*” OR “expert system*” OR “hopfield network*” OR “inference engine*” OR “knowledge based system*” OR “knowledge representation*” OR “lisp” OR “logic programming” OR “minimax” OR “nonmonotonic reasoning*” OR “perceptron*” OR “prolog” OR “recurrent neural network*” OR “search tree*” OR “theorem prover*” OR “tree search” OR "automated trading system" OR "algorithmic trading" OR "smart cit*" OR "differential privacy")

(2)   Title =
("explainability" OR "interpretability" OR "societal" OR "disinformation" OR "data privacy" OR "gender" OR "labour" OR "labor" OR "transparency" OR "fairness" OR "oversight" OR "bias" OR "legal" OR "education" OR "educational" OR "business" OR "intelligibl*" OR "journalis*" OR "suicid*" OR "commons" OR "sustainability" OR "misinformation" OR "information disorder*" OR "propagand*" OR "well-being" OR "wellbeing" OR "fake")
OR Title, Abstract or Keywords =
("accountab*" OR "ethic*" OR "moral*" OR "prosperity" OR "human control" OR "*justice" OR injustice OR "benevol*" OR "wellfare" OR "censorship" OR "hate speech" OR "fake news" OR "filter bubbles" OR "civil libert*" OR "human right*" OR "surveillance capitalism" OR "algorithmic justice" OR "critical AI studies" OR "critical algorithm studies" OR "platform studies" OR "app studies" OR "media studies" OR "gender studies" OR "private life" OR "humanity" OR "legislative" OR "political" OR "stakeholder*" OR "debias*" OR "GDPR" OR "racial discrimin*" OR "gender discrimin*" OR "predictive policing" OR "deepfake" OR "future of work" OR "gig economy" OR "uberisation" OR "uberization" OR "precari*" OR "public polic*" OR "attention economy" OR "quantified self" OR "whistleblow*" OR "whistle-blow*" OR "snowden" OR "internet surveillance" OR "mass surveillance" OR "gender equality" OR "decent work" OR "peace" OR "conspira*" OR "microwork*" OR "micro-work*" OR "free will" OR "biopolitics" OR "human autonomy" OR "psychology")

(3)   Research Area =
("Architecture" OR "Art" OR "Arts & Humanities Other Topics" OR "History" OR "History & Philosophy of Science" OR "Literature" OR "Music" OR "Philosophy" OR "Biomedical Social Sciences" OR "Communication" OR "Geography" OR "Government & Law" OR "International Relations" OR "Public Administration" OR "Social Issues" OR "Sociology" OR "Urban Studies" OR "Women's Studies")
AND NOT Research Area =
("Physical Geography" OR "Neurology")