What is an exploration engine?
There is search, and there is exploration. Search is when you are looking for a train table, the birth date of Queen Elizabeth, the best way to Chicago. You know what you are looking for, and there is a simple answer. Exploration is when you are looking for a good movie, an idea for a birthday gift, an opinion about a political debate. You don’t know what you are looking for, and there is no simple answer.
Exploration is no better than search. We don’t want to explore the landscape of train timetable websites. We just want to know when our train leaves. And we don’t want to explore all the possibilities to go to Chicago. We just want the fastest, or the cheapest, or the greenest.
Search and exploration are thus two different activities. As Kalinov et al. (2012) write:
“Search engines by design satisfy the information locating need only (the user knows something exists and needs to find out where it is). They do not address the complementary information discovery need (the user does not know something and needs to find out that it exists).” (Kalinov et al., 2012)
Yet, at the moment, we only have one tool, search engines. Search engines, like Google, are great, but they are mostly great at searching. They give you the most relevant response to your specific needs. And they are increasingly good at that. But they're not good at exploring.
Consider the way search engines respond to queries. They return a ranked list of results. This is good when you know what you want, and when there is one best answer. You look at the first ranked result, maybe the second one, the third one, and if the engine is good, then you’re done. You got your answer.
But what if you want to find a good movie? Then you get an endless list of "good movies". But "good" according to whom? Maybe you don't like superheroes or fantasy. Sure, you can specify your criteria: "best romantic movie", "best romantic movie of the last ten years", "most acclaimed romantic movie of the last ten years", and so on. This is a tiring process. You have to think about the specific criteria you are most likely to like. And it's not a very efficient process: you may end up watching the same kind of things over and over again, simply because you don't have a better idea of what you want to watch.
As Pearce et al. (2011) write: “Search engines are especially ineffective when exploring unfamiliar domains – domains in which you know the kind of things you like, but may not be able to articulate them using terms that would return useful results in a traditional search engine.” For this kind of situations, you need an exploration engine.
So what is the difference between a search engine and an exploration engine? The former gives you the best answer within billions of websites while the latter aggregates and summarizes billions of websites so that you have a view of all the options.
Kalinov, P., Sattar, A., & Stantic, B. (2012, April). Towards real intelligent web exploration. In Asia-Pacific Web Conference (pp. 411-422). Springer, Berlin, Heidelberg.
Pearce, J., Chang, S., Alzougool, B., Kennedy, G., Ainley, M., & Rodrigues, S. (2011, November). Search or explore: do you know what you're looking for?. In Proceedings of the 23rd Australian Computer-Human Interaction Conference (pp. 253-256).
The human need for exploration
At BUNKA, we are convinced that people’s motivation to explore should be taken seriously. Exploring is a ubiquitous requirement of life. Animals as well as humans explore to find mates and habitats, avoid predators, and learn new action–outcome associations. Scientists have long studied exploration in many different species, from bacteria to wolves to humans. They have demonstrated that similar cognitive and neural processes, and in particular the dopaminergic system, are present across species. All species indeed face the same challenges. Hills et al. (2018) sum up these challenges well: “Individual organisms must strike the proper balance between global exploration and local exploitation to survive – exploring sufficiently to find resources and exploiting sufficiently to harvest them.”
Scientists have also demonstrated that many kinds of exploration - animal foraging, visual search, information search, search within one's own memory, search in problem solving - present similar challenges and recruit the same neural processes. These neural processes originally evolved to explore physical landscapes and discover new sources of food and energy, but were progressively recycled to control cognitive exploration, that is the exploration of ideas, strategies and knowledge.
Humans exploring information on the web is thus one kind of exploration. It obeys the same logic and recruits the same cognitive mechanisms. For instance, people switch adaptively between local and global search: people leave a local patch of web pages when they perceive that its value has fallen below what can be found globally elsewhere.
Exploring is thus not like searching. When searching, you are exploiting: you are looking for something you already know exists and you want to know where it is located. Exploring is different: you do not know whether what you are looking for exists. In fact, you may not even know what you are looking for.
And because exploration is key in surviving, animals and humans alike have a strong drive for exploration. Experiments have long demonstrated that animals, from rats to macaques to dolphins, explore spaces and information even in the absence of any reward. They just like exploring, for the sake of it. And this is all the more true for humans. The human species has long been a generalist species, constantly exploring new environments, new ways of life and new habitat. Exploration is a defining feature of humans.
Figure from Roberts & Stewart (2018).
At BUNKA, we think that the need for exploration is a key aspect of human behavior, and that this need is not fully fulfilled by current search engines. People would like to search less, and to explore more, but they can’t. This is all the more true that the need to explore, and the pleasure to explore, tend to increase when people have more resources and more time. Studies with rats, but also parrots, vampire bats, wild-spotted hyenas and orang-utangs, show that individuals explore more when they are safe and satiated.
This is also true in humans. As we have demonstrated in our team, individuals with high and steady levels of resources are more ready to explore novel information and new rewards, and levels of artistic and technological innovation are higher in more affluent and safer societies. In line with this idea, large-scale surveys and psychological questionnaires demonstrate that, as economic development increases, the younger generations are becoming less and less conformists, more and more open to novelty, and more and more interested in exploring rather than exploiting. Not surprisingly, this is reflected in pop culture. Fictions with imaginary worlds, where people can explore and discover new things draw acclaim from the public, the critics, and the industry, making them both best-selling and most-appreciated fictions, from novels (e.g., Lord of The Ring and Harry Potter) to films (e.g., Star Wars and Avatar), video games (e.g., The Legend of Zelda and Final Fantasy), graphic novels (e.g., One Piece and Naruto), and TV series (e.g., Star Trek and Game of Thrones).
Exploration is an important drive of human nature, and increasingly so in the younger generations. Exploration engines are needed.
Baumard, N. (2019). Psychological origins of the industrial revolution. Behavioral and Brain Sciences, 42
Dubourg, E., & Baumard, N. (2021). Why Imaginary Worlds?: The psychological foundations and cultural evolution of fictions with imaginary worlds. Behavioral and Brain Sciences, 1-52..
Hills, T. T., Todd, P. M., Lazer, D., Redish, A. D., Couzin, I. D., & Cognitive Search Research Group. (2015). Exploration versus exploitation in space, mind, and society. Trends in cognitive sciences, 19(1), 46-54.
Roberts, P., & Stewart, B. A. (2018). Defining the ‘generalist specialist’niche for Pleistocene Homo sapiens. Nature Human Behaviour, 2(8), 542-550.
What is collective intelligence?
Collective intelligence refers to the fact that a team of cooperating agents can solve problems more efficiently than when these agents work in isolation. Collective intelligence is used by insects living in colonies, by teams of humans, and even by collaborative robots.
Collective intelligence has added value for everyone - citizens, consumers, media, public actors, industries, information management systems. In fact, modern societies increasingly rely on collective intelligence via peer review, crowdsourcing, open-source intelligence or collaborative filtering. And with the advent of Web 2.0 (or Participatory Web), collective intelligence is leveraged every single day: we use Twitter to explore others’ points of view, Youtube tutorials to repair appliances, Instagram to discover new trends, and Reddit and Wikipedia to learn about a multitude of facts.
For the moment, search engines do not provide access to collective intelligence. They have been tailored for the Web 1.0 and strangely they still are organized around websites, not user-generated content. True, there is a growing trend to use collective intelligence. Google for example lists not only the best sites, but also the most relevant results in Wikipedia and in the media ('top stories'). Google also informs us about users’ related searches ('People also ask', 'People also search for'.) And obviously, the very idea of PageRank is based on the principle of collective intelligence.
Yet, collective intelligence is hard to grasp. It does not lie in a single post or website, but rather in the whole of these posts and websites, their diversity, their differences, their oppositions and their convergences. Ranking websites, or presenting the most relevant news articles cannot be the solution.
The solution lies in aggregation. At BUNKA, our algorithm does not only give you the most relevant websites but also aggregate all websites, posts and reviews according to their semantic similarity (i.e. how much they talk about the same thing, they agree on the same issues), and present you with a map of all these aggregations, with their main topics, their respective popularity, their convergences and their distance from one another. You can then explore, zoom in, zoom out, and access to collective intelligence.
One domain where collective intelligence and aggregation might be especially useful is the fiability of information, and the diffusion of fake news. One obvious solution is to implement tools for assessing the content of a website or an article. Typical examples are debunking and fact-checking. But fact-checking and debunking have limits. They take time, and are difficult to automatize. Maybe more importantly, debunking and fact-checking adopt a confrontational approach which can backfire.
At BUNKA, our approach is different. Rather than evaluating the content of a post, we use collective intelligence to evaluate the context: who are the people who posted it? Who are the people who liked and shared it? Who disagrees with the post? It is through exploration that you get to be able to evaluate the quality of the information.
BUNKA doesn't tell people if they are right or wrong. It contextualizes what they are reading: Is it popular? Is it consensual? Is it fringe? Is it close to the alt-right? Is it endorsed by evidence-based websites? Is it shared by the anti-vaccine movement?
Picture adapted from Faris et al. (2017)
Using collective intelligence - what others think - is actually nothing new. Humans, as a species, consume an enormous amount of culturally derived information and, lacking sufficient expertise, constantly need the social context to assess the epistemic quality of this information (source, status of the speaker, popularity of the information, etc.).
To take an example from everyday life, if one is looking for a good restaurant for lunch, it is very likely that one will choose to enter an establishment where there are customers rather than an empty restaurant (the busiest restaurant might reflect its popularity and, therefore, the quality of its service). This is what BUNKA does, but on a grander scale, with many more dimensions.
Faris, R., Roberts, H., Etling, B., Bourassa, N., Zuckerman, E., & Benkler, Y. (2017). Partisanship, propaganda, and disinformation: Online media and the 2016 US presidential election. Berkman Klein Center Research Publication, 6
Lorenz-Spreen, P., Lewandowsky, S., Sunstein, C. R., & Hertwig, R. (2020). How behavioural sciences can promote truth, autonomy and democratic discourse online. Nature human behaviour, 4(11), 1102-1109.
Sperber, D., Clément, F., Heintz, C., Mascaro, O., Mercier, H., Origgi, G., & Wilson, D. (2010). Epistemic vigilance. Mind & language, 25(4), 359-393.
Information geometry: Why maps matter?
At BUNKA, we believe that the best way to explore the web is to use spatial representations: 2D spaces and 3D spaces maps. Humans are a very visual species that evolved to navigate macroscopic landscapes, with hills and rivers, plain and mountains. As a result, we are very good at using visualizations, maps, graphs, projections, and visualizations provide keys to improve knowledge and decision making.
But there is a deeper link between web exploration and spatial visualization. Recent advances in neurosciences suggest that the human brain uses the same neuronal resources to represent physical and abstract spaces. All kinds of information can indeed be represented in a conceptual abstract space. For example, different objects may be factorized according to how ‘‘cuppy’’ they are as well as their color. These abstract maps can be used to describe relational knowledge: distances, similarities and connections between entities.
Crucially, these abstract spaces appear to rely on the same cells, the so-called place and grid-cells of the hippocampus, that are used to encode spatial information. In the figure below, each cell (colored dots) codes for a specific concept (here, a type of car), and specific place in the conceptual place along the dimension of interest (here engine power and car weight). The arrangement of each cell relative to the other cells allows inferences (conceptual distance, conceptual similarity) to be made about the position of the concept relative to other concepts along the dimensions of interest (power and weight) in the domain.
This is exactly what we do at BUNKA. We use computational sciences (natural language processing and machine learning) to create maps of the internet, that is to say abstract spaces along the dimensions that interest users. These maps organize and structure the information. These maps are organized in territories, dedicated to specific topics. They allow users to visualize everything that is said on a subject at a glance, and use their spatial cognition to further explore and discover new ideas.
Bellmund, J. L., Gärdenfors, P., Moser, E. I., & Doeller, C. F. (2018). Navigating cognition: Spatial codes for human thinking. Science, 362(6415), eaat6766.
Behrens, T. E., Muller, T. H., Whittington, J. C., Mark, S., Baram, A. B., Stachenfeld, K. L., & Kurth-Nelson, Z. (2018). What is a cognitive map? Organizing knowledge for flexible behavior. Neuron, 100(2), 490-509.
Empowering people’s search
BUNKA is a cross-platform content aggregator and search engine. The idea is to access the different representations (points of view, ideas, debates, etc.) that people have of a subject (cultural or political). The way to search on BUNKA is topological, using the property of maps and territories, in order to "move" through the informational universe in the most natural way possible. It also allows to distance oneself from the algorithmic bubbles, since the map will display the different debates of ideas around the concepts.
We believe that exploring empowers peoples’ search. Specifically, exploring produces:
- Greater autonomy and greater agency: Search engines are the main tool to browse the web, but users cannot control this tool. They can just read linearly the result of the search. They are passive. BUNKA allows search engine users to have more autonomy by giving them a global vision, a map of the territory they want to explore and by allowing them to control the direction of their searches.
- Greater transparency: Search engines do not give access to their algorithms (i.e. Page Rank for Google). Even though they are working ever more efficiently, these algorithms remain opaque. In contrast, BUNKA's maps and nests make the dimensions used by the algorithm visible and allow users to easily calibrate these parameters (semantic folding).
- Greater diversity of viewpoints: Search engines do not provide access to all viewpoints, which can lead to polarization, information bubbles and false consensus. In contrast, BUNKA provides access to all points of view, even minority opinions: minority and majority opinions are each part of the map's territories.
- Greater epistemic quality of information: Search engines give little indication of the popularity of information, its degree of consensus, its political polarization, its distribution in the population, the quality of its source, the individuals who circulate this information. BUNKA does all that.
- More serendipity: Search engines give access to the most relevant result of the query, but they do not allow users to find things they were not looking for (unlike, for instance, when exploring a book store or shopping in a physical space). By contrast, BUNKA allows users to discover ideas or products they were not looking for, but that are "close" to the original search.
Let's work together
Contact us so we can work together.