The most popular and comprehensive Open Source ECM platform
Artificial Intelligence: Neuroevolution Points to More General Paths for Artificial Intelligence
The recent wave of artificial intelligence (AI) and deep learning has been powered by teaching machines to recognize patterns. The more data, the more power an algorithm has. More data makes it possible for the AI algorithm to derive deep and not easily discernible patterns and trends.
For example, by feeding an AI algorithm hundreds of thousands of X-Rays, the associated analysis of each of the scans by a trained radiologists, and the actual long-term outcomes of patients, a system that can read X-Ray scans with better accuracy than any one radiologist can be built. In this way, data and algorithms can aggregate the intelligence of humans and the history of outcomes to create a system that is more accurate than any single expert.
But what happens when there is limited data? Not all questions have massive amounts of historical data that can be applied to derive a solution. Deep Learning and the current generation of AI is less helpful in those cases. To address those kinds of problems, scientists are researching more advanced AI techniques.
One of the new techniques is called neuroevolution. This approach builds algorithms that can in turn create new algorithms or specify parameters that can optimize the AI techniques.
Kenneth Stanley, artificial intelligence professor, wrote that “neuroevolution is a subfield within artificial intelligence (AI) and machine learning (ML) that consists of trying to trigger an evolutionary process similar to the one that produced our brains, except inside a computer. In other words, neuroevolution seeks to develop the means of evolving neural networks through evolutionary algorithms.” Neuroevolution is expected to play an important role in developing techniques for robot control and vehicle navigation.