Artificial intelligence improves itself

We saw it dozens - or hundreds - of times in movies, novels and science fiction series: suddenly, artificial intelligence (AI) takes control and things go out of control for humans. With this background that is not at all reassuring, we are moving towards a world in which machines gain more and more abilities.

We saw it dozens - or hundreds - of times in movies, novels and science fiction series: suddenly, artificial intelligence (AI) takes control and things go out of control for humans. With this background that is not at all reassuring, we are moving towards a world in which machines gain more and more abilities.

A short historical account takes us to that moment in 1997 when Deep Blue, an IBM creation, defeated none other than the multi-time chess champion Garry Kasparov. The next leap came in 2014, when Eugene, a simple chatbot, managed to pass the Turing Test, defined by the famous mathematician, creator of computing as we know it where an algorithm must show behavior similar to that of a human being, without a human realizing that the answers come from a machine. And in 2017 Google's AlphaGo beat Ke Jie, the Go champion, breaking a new milestone.

 Some few examples of how certain algorithms, organizing data based on algebraic and matrix logic, can continuously improve and achieve increasingly ambitious objectives. This occurs due to various factors. The first of them is, without a doubt, an increasingly greater and cheaper processing capacity, a phenomenon accelerated and democratized thanks to cloud architecture. Added to this is an amount of data - which is the source of training for these robots to become more “intelligent” - that reaches limits incomprehensible to the human brain.

 The World Economic Forum estimates that by 2025 about 450 exabytes will be generated per day. To get an idea of the size, it is equivalent to almost 16 billion average-sized games. Our mobile devices recording every movement we make 24 hours a day, seven days a week, the corporate systems that keep track of all transactions, the content that is uploaded to social networks, the messages that are exchanged incessantly and the More and more Internet of Things sensors, for example, feed that figure.

 The conclusion is simple: there are more places to test these algorithms (at affordable prices, by the way) and more data to train them. This allows technological advances that, in a virtuous circle, stimulate investments and enhance improvements. This is how we come to amazing research.

The evolution continues

For example, the formulation of the matrix multiplication problem represented as a tensor decomposition. For that, the researchers put the Alpha Tensor artificial intelligence to search for “better algorithms” using reinforcement learning, that is, the learning method based on machine learning that rewards the desired behaviors and punishes those that do not meet the parameters to motivate the agent to perceive and interpret its environment and learn through the trial and error model. The conclusion? He managed to improve the Strassen Algorithm, the best of all known so far and which after fifty years of use was considered unsurpassed as a mechanism for multiplying matrices.

But in this world in continuous beta, Alpha Tensor managed to find the best way to program the functions used for artificial intelligence and unlocked a virtuous circle: it processes in less and less time, optimizes the use of data and generates a better algorithm. Details can be found at https://github.com/deepmind/alphatensor, free source.

 We are perhaps at the beginning of a path in which algorithms begin to improve themselves. Far from that threatening posture that we described in science fiction, artificial intelligence is consolidating itself as the ally that allows us to exploit our human abilities to the maximum at work, predict climate catastrophes or accelerate scientific research. Therefore, looking to the future, it is exciting that more and more algorithms, like Eugene at the time, manage to pass the Turing Test. Of course, I confess that I would be a little scared if one day they decided to intentionally “fail” it.

By Manuel Allegue – October 14, 2022

Be part of the Cloud world

Subscribe to our periodic summary of Technology News.

en_US