<\/p>\n
JAKE ANDERSON
\nThe Mind Unleashed<\/h3>\n
(TMU<\/a>)<\/strong>\u00a0\u2014 One of the biggest global players in artificial intelligence (AI) is Google and their high-tech Brain division has been pushing the envelope for years.<\/p>\n Now scientists working for their\u00a0AutoML<\/a>\u00a0project have a new paper in which they claim to be developing algorithms that can evolve on their own without human input. Even more stunning is their claim that they can induce \u201cmutations\u201d into new generations of algorithms, which mimics principles of Darwinian evolution, namely \u201csurvival of the fittest.\u201d<\/p>\n The team started with one of the most basic ideas in modern AI: machine learning. Machine learning tools allow us to use algorithms to search through massive troves of data and quickly identify patterns. But the traditional problem with this method is the element of human bias.<\/p>\n As the\u00a0team paper<\/a>\u00a0states: \u201cHuman-designed components bias the search results in favor of human-designed algorithms, possibly reducing the innovation potential of AutoML. Innovation is also limited by having fewer options: you cannot discover what you cannot search for.\u201d<\/p>\n To bypass this problem, the team wanted to develop a system by which AI can grow on its own.<\/p>\n The team used simple math equations to develop machine learning algorithms that author 100 \u201ccandidate algorithms.\u201d These candidates compete using basic machine learning tools like neural network image differentiation tests and the best-performing algorithms then mutated, or evolved, via random code alteration.<\/p>\n The system can cull through tens of thousands of algorithms each second in search of a solution while dismissing\u00a0\u201cevolutionary dead-ends\u201d<\/a>\u00a0and duplicates. Over multiple generations, the process grows a library of high-performance algorithms. According to the Google team, these new algorithms have already\u00a0reproduced decades<\/a>\u00a0worth of human-led AI discoveries in only days.<\/p>\n Perhaps most astonishingly, the new AI algorithmic evolution is able to eliminate the problem of human bias that is often introduced during data input. The AutoML-Zero can essentially \u201cautomatically discover\u201d unknown algorithms and develop new previously undiscovered AI programs without any human intervention, using only basic mathematical concepts.<\/p>\n Haran Jackson, the chief technology officer (CTO) at Techspert,\u00a0explains why<\/a> the new paper\u00a0is so interesting:<\/p>\n \u201cAs exciting as AutoML is, it is restricted to finding top-performing algorithms out of the, admittedly large, assortment of algorithms that we already know of. There is a sense amongst many members of the community that the most impressive feats of artificial intelligence will only be achieved with the invention of new algorithms that are fundamentally different to those that we as a species have so far devised.<\/em><\/p>\n \u201cThis is what makes the aforementioned paper so interesting. It presents a method by which we can automatically construct and test completely novel machine learning algorithms.\u201d<\/em><\/p>\n As noted before, the scientists say the AI programs can improve upon each previous generation, producing a kind of \u201csurvival of the fittest\u201d that resembles Darwin\u2019s view of evolution in natural biological systems.<\/p>\n While much testing and review still await the Google team, their\u00a0new paper<\/a>, titled \u201cEvolving Machine Learning Algorithms From Scratch,\u201d suggests the tantalizing, albeit unnerving likelihood that AI of the future will be designed by other AI machines.<\/p>\n One can only impishly guess at what self-replicating AI algorithms evolving on their own will mean for the future of life on Earth.<\/p>\n