In a world controlled by data, Big Data opens up a brand new field of possibilities. It puts us before a new series of decisions to make, that are often dependent on variables that radically change our decisional instincts. Eric Biernat offers a striking plunge into a world where machines can learn. According to the physicist Werner Heisenberg, it is impossible to simultaneously and precisely measure the speed and the position of a particle. This is not due to a technical defect in the technology used, but to the particles’ structural inability to be measured according to those to axes. It is on the basis of this postulate which is the foundation of quantum physics that Eric Biernat starts his talk on the possibilities that machine learning offers.
Predicting as accurately as possible
For a long time, it was thought that machine learning would be limited to repetitive tasks as was the case with industrial automation. But looking at machine learning from autonomous cars to the recent victory of a robot versus the Go world champion, it is clear that the learning ability of machines is exponential. To optimize the process, data scientists apply repetition and inundate the Artificial Intelligence with examples. To teach it to recognize blue, they submit images with several nuances of the color, together with counter-examples with elements of green, red, yellow… The higher the number of examples, the deeper the learning is. Once this step is completed, the essential phase of prediction can start.
“Predicting means identifying the critical population of our objective” according to Eric Biernat’s definition. He adds that the result isn’t decisive given that the margin of error tends to evolve continuously.
The margin of error defines the level of performance of a machine and it is invariably indexed to its ability to predict. Because the more accurate the prediction, the more efficient the action plan.
An algorithm’s margin of error is invariably linked to the time at our disposal, the time it takes to position the risk cursor. As Heinsenberg theorized, the more I define my population, the less reaction time I have to put in place the parameters necessary to the precision of my model. But are mistakes a bad thing? Eric Biernat takes the example of marketing to affirm that “error is structural”.
For instance, all digital, banner or mailing strategies are only around 1% efficient.
The rest is a failure, people who never open their emails. The mission of a manager is to assess the mistakes in an algorithm and to “rethink their processes in depth”, which is what Netflix did in 2009 with a prize that handed out one million dollars to the data scientist team able to improve its predictive algorithm by 10%. This was good operation which improved client satisfaction and helped gather more new data. According to Eric Biernat, this virtuous circle can optimize the performance of many companies. However, if intelligent machines and their predictions are increasingly adapting to the professional world, doubts remain as to their ability to create and be artistically expressive.
More to read : Andrew Mcafee – The Second Machine Age
Let us create
In May 2016, Google unveiled Magenta, a robot able to create a 1.30mn long melody. This was a technical prowess, but like all machines, regardless of how elaborate they are, Magenta, so far, can only reproduce what man has taught it by improving through examples.
Yet is reproducing truly an act of creation?
The very use of the words to express the idea evokes a disruption. Moving away from the path, facing difficulties, moving out of one’s comfort zone, thinking out of the box… it all leans on novelty, innovation.
According to Eric Biernat, “machine learning can create, but only by learning to reproduce the past”.
Nothing’s new, it’s all already in the data. Rather than trying to artificially duplicate our arts, shouldn’t we use machines as assistants instead, aids that would relieve us from constraining tasks, so we can focus on more thriving activities? Eric Biernat insists that “the progress of machine learning must lead us to ask how our time as a biological brain can be employed, how we could use or intellectual abilities”. The evolution of machine learning must enable us to focus on what is worthy of our attention, such as emotions and creativity.
To go further, discover Eric Biernat’s talk on our Youtube channel: