We continue our column on theartificial intelligence trying to answer a tricky question: How and when will artificial intelligence surpass human intelligence?
In the previous article we talked about which artificial intelligence to use and for what purposes. We have already made distinctions clear to you in that context. It is necessary not to generalize to have a broad and concrete idea of a diverse technology and its different tools.
Here too we use the same approach, trying to give you information as objectively and specifically as we can. the perspective ofartificial intelligence that surpasses the human one, and of a challenge between the two, in fact it may seem alarming. But the issue is more complex than that.
What is Intelligence?
We didn't miss the word "artificial" in the question. To understand what we are talking about, we need to identify what intelligence is, so as to have clear what the "field of challenge" is between human and AI.
According to the definition of Treccani, which we quote verbatim, intelligence would be: “Complex of psychic and mental faculties that allow man to think, understand or explain facts or actions, develop abstract models of reality, understand and be understood by others, to judge, and together make it capable of adapt to new situations and to change the situation itself when it presents obstacles to adaptation; proper to man, in which it develops gradually from childhood and in which it is accompanied by awareness and by 'self-awareness. "
So already here we can collect the elements highlighted in bold and add other information.
The different types of intelligence
There are several theories, all criticized and currently debated, that seek to classify different types of intelligence. Here we propose the theory of multiple intelligences of the psychologist Gardner elaborated in 1983. Initially there were 7 and today we have reached 12:
As we have already told you, not all and not for all of these types of intelligence are considered valid. In any case, we have proposed them to you because they concern talents and abilities closely linked to intelligence, beyond academic definitions.
This list is indicative to understand that it is simplistic to compare AI to human intelligence in a general sense.
Types of artificial intelligence
Again, not to be reductive, we point out that there are two distinctions to be taken into consideration when defining theartificial intelligence. Within these two, then we go to frame the different tools that we know and that will come.
THEweak artificial intelligence it is based on systems that can act successfully in some specific functions, such as translations and processing of texts, images, videos and sounds.
It basically simulates a thinking brain, but only from a functional point of view.
So one does not abstract from rational elaboration tasks by asking questions that lead to self-awareness.
Basically it cannot be autonomous from man for an instrumental question.
Meanwhile, strong AI is theoretical, so it hasn't been implemented yet, but it is being worked on.
The characteristics aimed at are autonomous learning, reasoning, understanding of context, creativity, judgment and adaptation.
So when it comes to the challenge between AI and Human in an alarming sense, strong AI is the one to consider. And, we repeat, it does not yet exist.
The risks of artificial intelligence
In 2015 Stephen Hawking warned humanity and the scientific community about the risks of developingartificial intelligence strong. So he involved 1000 other influential entrepreneurs, scientists and programmers in signing and publishing a memorandum comparing strong AI to the "new nuclear bomb". Then to an avoidable catastrophe. To the reckless use of a technology that is beyond human control.
Today rather the most heartfelt fears are towards weak AI, because it is already available to everyone. Text processing and audiovisual software could make some technical and artistic professions disappear or revolutionize their very nature. The automation of agricultural, industrial and manufacturing processes would take away millions and millions of jobs, creating disastrous social and economic consequences.
But there's more.
Today you can go to the supermarket to buy everything, so it's not necessary for everyone to know how to cultivate the land, raise an animal, hunt, or plant and take care of a fruit tree. So many of us are not able to do it in the slightest.
Similarly, if we delegate too much to artificial intelligences the resolution of logical problems or the processing of information found around, we risk finding ourselves not needing to think.
So to weaken some of our cognitive abilities.
The data problem
Often one gets an idea of theartificial intelligence related only to the product. So how it manifests itself, the results it gives and the functions it performs. And it is on the basis of these factors that then theartificial intelligence defines itself, enters our imagination and gives rise to disastrous theories.
What is not considered, however, is that theartificial intelligence it develops and learns based on the data it is given. So thanks to the systems of machine learning. Beyond the fact that there are different systems of machine learning, it is enough for us to reflect on the data, because it is these that make us understand how aartificial intelligence process and why.
Not knowing the provenance of this data exposes you to the risk of inaccuracy or willful distortion of information, copyright infringement, government abuse… but aren't these already problems that have existed with current and previous media for at least a century?
Therefore we feel like saying that, rather than deprecating or being scared of a tool, we need to worry about who uses it and how.
And with this statement we do not think we are betraying the objectivity promised at the outset.
In what fields has artificial intelligence surpassed human intelligence?
Now we show you which are the fields of knowledge or skills in which weak artificial intelligences manage to overcome some human capabilities.
- Game of chess and go: go (picture above) is a Chinese game that is over 2500 years old and has more combinations than the number of atoms in the observable universe.
- Transcription, processing and translation of texts: in terms of speed and the amount of sources that the AI can draw on to perform these functions.
- Medical diagnostics: in relation to the correlation between the data provided to him and in some cases of specific diseases, such as macular degeneration, which he diagnoses with a 94% accuracy percentage.
- Industrial automation and maintenance.
Conclusions on how and when artificial intelligence will surpass human intelligence
With what you have read we hope to have avoided reductive polarizations or the idea of catastrophic countdowns.
Certainly there are risks associated with developing some artificial intelligences, both weak and strong, to the use of data, just as the threat to some occupations is concrete. But the questions we should perhaps be asking aren't about when theartificial intelligence will surpass the human one, but: how do the different types of artificial intelligence? How do they learn to perform their duties? Where does the data that is provided to them come from?
Who are those involved in developing the artificial intelligences? What goals do they have? What values do they have? How are they framed in the geopolitical balance?
How are the artificial intelligences and why?
We understand that these are complex questions to ask and to which it is difficult to answer, but it is also true that we need to refine ourselves before claiming to judge, use and understand refined means.
It's too much easier to criticize, be afraid and imagine catastrophes. We are not in a film where the distinctions between good and evil are clear-cut.
The reality we live in today is far more nuanced.
And the mission to which every human being is called to be aware of the world in which he lives is to be able to relate to its complexity.