Is it possible to create Artificial Intelligence in the short term?
How the Internet is unifying knowledge and social interactions to produce General Intelligence
In the final part of the previous article, we made a comparison between the computing power of computers and that of the human brain, and we concluded that we expect them to equalize between 2025 and 2030.
However, computing ability is not equal to problem solving skill. Today the intelligence of machines is not comparable to that of human beings. Let’s see why.
Biological intelligence
The concept of intelligence is defined as the ability of an organism to adapt to its environment, and it manifests when the living being has to deal with changing or new situations. As a recent example, we have the unprecedented situation caused by the first pandemic of this 21st century, COVID-19. There has been a global adaptation (collective intelligence) that has culminated in obtaining several vaccines in a period of time unthinkable until now. And there has also been an individual adaptation (intelligence per se), in how everyone of us has faced difficulties: home confinement, changes in the way of working, modification of social relationship habits, etc.
Today we know that this adaptation has been successful beacuse we have survived as individuals, and as society the total number of deaths has been minimized. Let us remember that in the previous pandemic, the 1918 flu, at least 50 million people died around the world. In the case of COVID-19, this number rises today to 5 million, when we are facing the end of the pandemic and it is transforming into a endemic virus. In fact, total deaths are only 10% of the aforementioned flu (even so, we all agree that it is still too large a number with an unacceptable human and emotional cost).
Having overcome the pandemic is a direct reflection of the capacity of our intelligence, and it should make us conclude that contemplating it only in its logical-mathematical facet is reductionist; it has many more. Would we have survived COVID just by solving numerical problems?
To close this point on biological intelligence, let us bear in mind that its ultimate goal is our survival as a species, as corresponds to the evolutionary history of living organisms.
Machine intelligence
What happens when we apply the above definition of intelligence to machines? The straightforward conclusion is that they are not smart at all. However, we should not give up so quickly, there are nuances and future prospects.
Historically there have been two approaches to Artificial Intelligence (AI): weak AI and strong AI. The first refers to the ability of computers to solve specific problems in well defined situations (structured environments). A typical case is that of the game of chess: in 1997, Deep Blue — a supercomputer created by IBM — was able to defeat the then world champion, Gary Kaspárov.
Thus, we have today many examples of weak AI, such as recommendation systems for series, music, videos in our mobile applications. They get to know us and identify our tastes from the continuous interaction with them.
In addition, beyond strategy games, we have as relevant examples the military and space missions that are executed without direct human intervention. Thus, after the arrival of man on the Moon in 1969 and subsequent attempts, human-piloted missions beyond our atmosphere have been dropped decades ago (basically due to cost reasons). Robotics and artificial intelligence have allowed us to reach Mars and explore it with a relatively low investment, without the need to physically carry astronauts.
Strong Artificial Intelligence
However, strong AI requires that the machine or robot be able to work in a general, unstructured environment. Let’s cite some examples: having conversations with humans, driving completely autonomously, answering questions of general culture, interpreting jokes and ironies … Some of these things are capable of being done separately by different machines, where each skill is solved by a different robot. In other words, they are different weak AIs, each one specialized in solving a specific task.
Strong AI will emerge when a single or cluster of machines is able to tackle this set of situations without being restricted to a specific task or group of tasks, as well as to face situations that are not previously known. From its point of view, it will be the way to ensure survival in the same way that occurs with the biological intelligence of humans.
Is it conceptually possible? Yes, because the field of action of the different weak AI will expand as they are exponentially improved, and there will come a time when they overlap and cover a continuum of situations.
Is it feasible in practice? We don’t know for sure yet. It takes many technological advances to achieve that hypothetical strong AI.
In any case, I am not going to get into this debate now: first because it is still a speculative terrain, and second because the goal of this article focuses on predictions in the short term, 10 years from now.
Internet as a playground for AI
In the last two decades the development of technologies related to Internet has been exponential. And it will not stop being; we continue to see more and more far-reaching innovations that make us spend more time online. I am referring fundamentally to the 360º experiences provided by modern social networks such as Instagram, Linkedin, Twitter… When I speak of 360º I mean the confluence of written, audio and video messages in a unified user experience. That is, from a single application we access in a multisensory way the particular universe that each person wants to show us about themselves.
In fact, these applications feed on all kinds of data from their users, data that the software uses to train the algorithms. Then they are who decide what information to show us, with which new people we must connect, which market products adapted to our needs we can buy, etc. That is, each of these social networks represents an Artificial Intelligence in itself, with the only restriction that its field of action is limited to a particular scope (leisure, professional interactions, intimate relations).
There is a complementary trend that refers to the integration between applications, that is, that I can have my life unified on the Internet. This is yet partially truth because big companies do not provide full integration given that their value is mainly based on their algorithms and users’ data (we will cover this interesting topic in a future article).
Thus, adding the two ingredients described above- algorithms of social networks and complete integration between applications-, we are already creating a general AI, which has to develop using the data provided by users interacting in the natural online environment (we already feel so good in Internet as in the physical world), and the algorithmic base created by the thousands of engineers who work on them.
We therefore go from the 360º experience that we described above to the immersive experience on the Internet, where everything that happens online will be a mirror the physical world. As a consequence, the border between living online and in the physical world will disappear as such.
A time will come in short that there will not be a distinction between them, but rather they both will part of a continuum. And when that happens, machine intelligence will come implicitly. We can be pretty confident that this will happen before 2030. It will be the first step of strong AI.