Search

OK ladies, let’s talk AI.



Or artificial intelligence.

If we take our knowledge about AI from Hollywood movies, it would be hard to lay down at night in the anticipation of Dooms Day, when your blow-dryer will strangle you in your sleep.

Truth is …this cannot be further from the truth.

Google has announced last week a program called LaMDA, fully based on AI. Here, a computer can embody ''the personality of an object'' and carry on human-like conversations on behalf of that object, as if it were alive. All based on information the computer ‘’learned’ by itself’. It can formulate answers in fully natural language.

Here you can see a conversation between a person and the planet Pluto : https://www.youtube.com/watch?v=aUSSfo5nCdM

Scarry isn’t it?

Well not really! So what is AI? What is so different about AI?

The way that computers usually work is very simple: there is a question, a problem or a task…Programmers write a code exploring step by step every possibility and, by following this path called algorithm, the solution is generated.

In contrast, AI works similar to the human brain through training and repetition. It gets a lot of input data for a specific question and needs to ''be told ‘’ many times what the correct answer is... over and over again, so that at a certain point it can come to the same conclusion by itself.

For example: if we think about the program used for self-driving cars, when a pedestrian would appear in front of the car: if this program were a classic algorithm, it would take a picture and think…is this a cat? NO!….is this a dog? NO!…is this a horse? NO!….etc is this a pedestrian? YES! ...is it moving? NO...is it in the middle of the road? YES. Then BREAK. This could take a while…

If this is done through AI: the program has been trained in prior with many images of cats, dogs, pedestrians from different distances, postures and angles, so that the first time it sees the image it can say: it is a pedestrian in the middle of the street...So BREAK

AI is an almost exact digital simulation of how a human brain learns. While a human brain has 100 billion neurons, the average AI has approx 16 billion, but growing rapidly.(the best super supercomputer can reach 86 billion).

How is it used?

Taking the example before, we can conclude that AI would only work if it had ‘’seen’’ before enough images to recognise the pedestrian among all the possibilities. Thus, an AI is as good as the training data, similar to a person who is as healthy as he eats or as smart as he reads.

AI is present in our lives through Siri (from your iPhone) or Amazon Alexa. And with every question we address Siri or Alexa we are training them for free. Social media and search engines like Google use AI to predict behaviors or to place advertising to be as relevant as possible. Is this dangerous? Only as much as advertising can be, with the watch out that personalised advertising can be much more convincing and sometimes manipulating.

Natural language processing is a branch of AI, which was perfected lately by Apple, Amazon and Google and used in their assistants like Alexa or Siri. The objective is to make a computer answer and interact like a human and not sound like a robot.

The Turing test, created by Alan Turing in 1950 was long time considered the borderline to claim that machines can reach human level intelligence. In the simplest form, the Turing test consists in a person evaluating the conversation between a human and a machine, without being able to distinguish which one is which. Google has presented already 3 years ago a conversation with an AI that is very close to that: https://www.youtube.com/watch?v=D5VN56jQMWM, but without a commercial roll out since, still work in progress. So we are still some steps away from human level intelligence invading our lives.

But can AI be a danger to humanity for the future? More about that in the next post.

43 views0 comments