Artificial Intelligence Today

There is still a long way to go till we’re going to have the artificial intelligence known from sci-fi movies that gains consciousness (and eradicates human kind), but in terms of image and voice recognition machines caught up with humans. This is briefly what we learnt from the presentation of Attila Agod and Ferenc Szalai about AI at the event of Café Group, the so-called ‘Breakfast Club’.

Where Are We?

Due to the already-mentioned sci-fi movies and novels, we are inclined to think of artificial intelligence, or AI, as humans, even though technology is far from being there. Currently, we are at a “weak AI” level, which basically means the collection of fundamental cognitive functions. Programs can see and hear, they understand text, recognize patters, but as long as these perceptions do not come together as a whole, un unconscious AU cannot optimize based on the data.


Attila Agod, a researcher of Scenarios AI

According to Attila Agod, a researcher or Scenarios, the promise to integrate AI into economy has been worded on three occasions in the last couple of decades, and contrary to previous “hypes”, today we have everything to realize that promise. The reason for that is the exponential growth of data and computing power every 2 years. The process is similar to flying: people had been trying to build airplanes from the 1860’s, but it took 40 years to build a motor that can lift the weight of the airplane. Capacity has now reached a critical point from where – to reflect on the previous metaphor – the airplane can take off.

Psychedelic Dream

Deep learning brought real breakthrough. The artificial neural network functions like the human brain: layers process the information, starting from the lowest level, i.e. colors, blurs and lines. Following this, AI builds a basic structure (e.g. eyes and ears in terms of animals), then the face and finally it recognizes what is on the image.

The visualization of the process is similar to a psychedelic LSD trip.

Better Than Humans

While browsing a database of hundreds of thousand of pictures, average people cannot recognize 5% of images and their performance is similar in understanding raspy recordings. Due to deep learning, machines caught up with humans in 2015 and they have been enhancing ever since. The same holds true for text comprehension and translation, especially in English and Mandarin Chinese.

These capabilities can be used in practice more and more efficiently. The Adobe product, Sensei, labels images automatically, it recognizes fonts, prepares a summary of a document and sets consumer product reviews in order.

Building Your Knowledge with Lego Bricks

Four bigger companies – Amazon, Google, IBM and Microsoft – develop artificial intelligence at the moment, that build their foundation on cloud-based data storage. According to Ferenc Szalai, a physicist, it is unnecessary to build AI from scratch provided that you can use the already existing knowledge as Lego bricks in the process.

Ferenc Szalai, physicist

The modelling of creative tasks proves to be the biggest challenge now, but there are promising solutions. The trailer of the movie ‘Morgan’ is a good example, as it was cut by the artificial intelligence of IBM. They entered thousands of movies, their trailers and viewer reviews of the trailers into the computer, and an algorithm prepared the trailer to the best of its knowledge based on the data.

Szalai believes that we have to work with artificial intelligence in the future and they have to trained just as a junior. The question for the next couple of years is how far we are going to be from machines: will things remain the same or are they going to be inside us as implants?