AI Building AI: Mankind Losing More Control over Artificial Intelligence
By Makia Freeman
ActivistPost.com
DECEMBER 8, 2017
AI Building AI is the next phase humanity appears to be going through in its technological evolution. We are at the point where corporations are designing Artificial Intelligence (AI) machines, robots and programs to make child AI machines, robots and programs – in other words, we have AI building AI.
While some praise this development and point out the benefits (the fact that AI is now smarter than humanity in some areas, and thus can supposedly better design AI than humans), there is a serious consequence to all this: humanity is becoming further removed from the design process – and, therefore, has less control. We have now reached a watershed moment with AI building AI better than humans can. If AI builds a child AI which outperforms, outsmarts and overpowers humanity, what happens if we want to modify it or shut it down – but can’t? After all, we didn’t design it, so how can we be 100% sure there won’t be unintended consequences? How can we be sure we can 100% directly control it?
AI Building AI: Child AI Outperforms All Other Computer Systems in Task
Google Brain researchers announced in May 2017 that they had created AutoML, an AI which can build children AIs. The “ML” in AutoML stands for Machine Learning. As this article Google’s AI Built Its Own AI That Outperforms Any Made by Humans reveals, AutoML created a child AI called NASNet which outperformed all other computer systems in its task of object recognition:
The Google researchers automated the design of machine learning models using an approach called reinforcement learning. AutoML acts as a controller neural network that develops a child AI network for a specific task. For this particular child AI, which the researchers called NASNet, the task was recognising objects – people, cars, traffic lights, handbags, backpacks, etc. – in a video in real-time. AutoML would evaluate NASNet’s performance and use that information to improve its child AI, repeating the process thousands of times. When tested on the ImageNet image classification and COCO object detection data sets, which the Google researchers call “two of the most respected large-scale academic data sets in computer vision,” NASNet outperformed all other computer vision systems. According to the researchers, NASNet was 82.7 percent accurate at predicting images on ImageNet’s validation set. This is 1.2 percent better than any previously published results, and the system is also 4 percent more efficient, with a 43.1 percent mean Average Precision (mAP).
With AutoML, Google is building algorithms that analyze the development of other algorithms, to learn which methods are successful and which are not. This Machine Learning, a significant trend in AI research, is like “learning to learn” or “meta-learning.” We are entering a future where computers will invent algorithms to solve problems faster than we can, and humanity will be further and further removed from the whole process.
The Rest…HERE
