Summit -Wisdom of the Crowd
In our previous post, we introduced our readers to the main Citizen Science projects presented at the Summit on Human Problem Solving and Artificial Intelligence last month. Today we continue talking about the power of crowdsourced problem-solving, advances and limitations in machine learning and more!
Wisdom of the Crowd and Social Science Supercolliders
Humans are amazing problems solvers and they are often all the more powerful when solving problems in teams. We have seen the power of teamwork in our own Alice Challenge, and we were keen to hear more success stories from other researchers during the summit. Douglas Merkant started his talk with a pretty amazing demonstration of this by telling us about the USS Scorpion shipwreck that was discovered by putting together seven hypothesis from experts of different fields. Searching in the area where most hypotheses overlapped lead to the discovery of the submarine in this real life Battleship!
Crowdsourced and team-based problem-solving methods are becoming increasingly popular. With many efforts happening online on platforms like Mechanical Turk and even Twitter, social scientists are starting to get very excited about the trends and pattern emerging in these novel settings. John Murray from SRI International called this the social science supercollider - a totally new way of doing social science research in the wild, using digital media, games and environments are much closer to real life than the standard social science laboratory settings. At ScienceAtHome, we can’t wait to be a part of this revolution with games like the Alien Game and Skill Lab!
It is impossible to talk about the problem-solving and games these days without talking about the advances in machine learning. With algorithms that beat Go-grandmasters and start to have human-like image recognition skills, do we need humans to solve problems anymore? I find this an amazing moment in time when we really don’t know what the answer to this question will be. AI solves problems with brutal accuracy and with increasing flexibility, as blogger Arthur Juliani showed us in his talk. Yet humans have an easier time dealing with noise, limited information and abstract tasks, while AI often lacks intuition and high-level problem-solving skills.
Özür Simsek told us of her efforts to get algorithms to think more like humans, who often replace a complex question by a similar but simpler one—and therefore solve hard problems with evident ease. This kind of thinking is hard to code, and we stayed up until the late hours of the night discussing ways of learning similar tricks from our players and incorporating them into our own algorithms.
The last day continued with a similar observation from DTU Copenhagen’s Lars Kai Hansen pointing out that although an algorithm can learn many Atari games, it seriously struggles with games requiring semantic information. Show a player a puzzle with a key and a lock and she will immediately know that they go together—but a computer has to learn this by trial and error. This echoes the concrete results from many citizen science talks we saw in these three days—the human mind is truly a supercomputer.
The summit was truly an inspiring event and we hope to organize it again in a year’s time. With all the fields represented in the summit speeding forward, it will be exciting to see how 12 months of research and development play out! On our side, we have brains buzzing with ideas and our hands full with the new and improved project. Watch this space for all our latest news.