The problem of being overwhelmed and trapped in a flood of information is not only inherent in our daily modern lives, but a challenge faced in research and development in general. Could AI help us out with this debacle?
But first, it is important to understand what the actual issues with data, knowledge and conclusions are. The interconnections between these terms are not only relevant to the philosophy of science, but crucial to professionals of other occupational areas like entrepreneurs and journalists.
Knowledge is uncertain.
Scientists, entrepreneurs and journalists alike access literature and, based on this acquired knowledge, contextualize and conclude. Many factors contribute to a final conclusion when a new piece of data is set into context of the complex world. Ideally, it should be possible to identify the one “true” statement arising from the existing information. Realistically, however, this is not the case. According to Sir Karl Popper, “Science can be described as the art of systemic simplification” and hence our theories are always wrong, but we can refine them to a degree of being very close to the truth. Therefore, they are sufficiently useful for humankind. Due to this nature of science, human knowledge is hypothetical.
Inductive reasoning forces us to predict events.
Restricted by feasibility to survey large populations, scientists look at a small sample of individuals to gain insight into a certain issue. The results derived from few individuals are then used to conclude a general rule – a process know as inductive reasoning. For instance, ten mice fed with compound X live twice as long as mice fed without this compound. Using inductive reasoning, scientists would propose that in general mice (all mice on this planet not only these 10 mice) have a longer lifespan when they eat X.
Similarly, an entrepreneur might infer even further that if compound X increases the lifespan of mice, it potentially does the same for humans. A journalist reporting this scientific breakthrough might go through the same line of thinking. As is obvious with this example, to have a value for us, data is interpreted and used to predict outcomes (with uncertainty).
Superforecasters use the bird’s eye perspective.
Professionals integrating data in their decision-making processes are thus forecasters. In a study, the psychologist and initiator of the “Good Judgment Project” Philip Tetlock identified so called superforecasters. These are persons, who are better at predicting future events than an average person. By studying these talented forecasters, Tetlock could derive general principles that improve making a prognosis. One such is the inside versus outside view. The crucial approach for accurate predictions is to take a step back and look at the problem from the outside. For answering the question of whether the government in Egypt will fall, most people immediately read up on everything that happened in the past in Egypt. More wisely, one should rather ask what percentage of middle eastern authoritarian governments fall yearly (external perspective). Then this number is adjusted with specific knowledge about Egypt.
Hunting for the outside view.
This is where the vast amount of data comes in. How can we find relevant data – this missing puzzle piece from an, at first glance, possibly unrelated field. Translated in Tetlocks terms: How to find the appropriate outside view? There is no one solution, but artificial intelligence (AI) might come in handy.
“Keywords suck”, declares the co-founder of Iris.AI Anita Schjøll Brede. Indeed, extracting the relevant keywords from an unrelated field is a barrier. The science assistant Iris.AI is an AI that should be able to read, understand scientific literature and propose new hypotheses. So much for the future plan. Today, Iris.AI can read a scientific paper or a TED talk and map out relevant topic areas including scientific publications without receiving keywords. As is typical for her kind, Iris can learn and become better the more people train her to understand what is meaningful. This function is freely available on the web. The owners also rent Iris.AI to companies. In Scithons (Science Hackathons) questions can be fed to Iris, who then helps to grasp the overview of a field and assists in finding solutions.
If you want to try using Iris.AI, 1. grab the URL of your favorite TED talk at TED.com. 2. paste the URL into this interface. Iris.AI will mine the scientific literature for publications relevant for the talk.
The output of Iris.AI on the TEDxVienna talk from Carry Poppy – A scientific approach to the paranormal: Left: Categories extracted from the talk. Right: Papers found in one sub category of “Ghost”.
Iris.AI is not the only AI trying to ease the pain of scientific literature research – there are more intelligent search tools emerging.
Journalists are another group of people who have to quickly catch an overview of an unrelated field to place new findings in a meaningful context. Especially science journalism suffers under tight publishing deadlines, which result in often badly researched articles and overstated headlines. Algorithms like Science surveyor might prove as useful. In a collaborative effort of the Universities of Stanford and Columbia, researchers created a tool that supplies journalist with useful information around newly published studies. Their case studies show that for a given publication Science surveyor can retrieve contact details of experts in the field, which can be contacted for further opinions. It also spits out a score, which rates how unusual the findings are to ease the judgement of novelty of discoveries.
Here is an example:
A highly coveted market emerges.
The automized transformation from text to knowledge and understanding and thus meaningful suggestions has emerged as an important quest for the future. Other companies, like the Chan Zuckerberg Initiative-sponsored company Meta, the Stanford offspring Yewno, or the danish startup Unsilo.
For AI to be helpful on the long run, it will be important that AIs can avoid information bubbles and give us the relevant outside view.
Header image credits to shutterstock