It seems that the term Artificial Intelligence has not dropped from our mouths for a couple of years, but the truth is that it was in 1956 when a university professor, John McCarthy, coined it in an academic co-management. Although the Laws of Turing and Asimov are from 1950, it was necessary to wait six years to coin this term and refer to that set of technologies with which the machines are intended to be intelligent.

Since then (more than six decades have passed), the advances have been striking: computers capable of winning chess champions, machines that learn by themselves, cars that drive solely …

But what will Artificial Intelligence really be able to do in 10 years? Will we still call it “Artificial Intelligence”? What about reality, and what about science fiction or hope in what is said about these technologies?

Why is it difficult to make these forecasts

If making predictions of the future is difficult, doing so on a technology-related aspect such as Artificial Intelligence seems complicated even for the experts who work in its development. To understand it a little better, let's know that all the advances we are seeing today come from the development of deep learning. A matter that, precisely 10 years ago, “did not work on a practical level.” This gives us a vision of the complexity of making predictions in the future.

Therefore, the only thing we can do, as a starting point, is to review what has been achieved in this decade to try to see what they will be able to do in a few years.

During this time, technology has learned to recognize objects and faces, to perform predictive analyses (which are applied in advertisements or in medicine), or to create games ( Pokemon go type ). These are all predictive models, but there are also generative models, which are capable of creating something: writing, painting … Or inventing people’s faces, for example.

Autonomous driving will be technically viable

Therefore, and underlining that it is very difficult to know what the technology will be able to do in the future because, if we continue to advance in the same areas of the last 5 years (recognition of patterns in video, text, images, and audio, that is, in unstructured data), autonomous driving will be a reality from a technical point of view. We will have the technical skills to have safe autonomous driving, although there are other non-technological factors that will determine whether it will be a reality or not in the market.

It is a fact that, right now, cars are already able to react to what the sensors see and what they have learned. But in the future, they will be able to predict the behaviors of other cars, driven by AI or people. A component that is not present today, but in the future, imitating what humans do -

When talking about autonomous driving, it is divided into 5 levels. If level 5 is the maximum autonomy, in 10 years, many cities will already be at level 4 (which still requires a human to take their hands on the wheel). At level 5, it will not be struck by purely technical issues, except in very closed and controlled environments.

Diagnostics, personalized medicine and molecules under the prism of AI

Medicine will be another area in which we will see more progress. Although today it is still one of the most talked-about topics, it is not yet deployed with its full potential, so when this occurs, we will see important advances. Today it is already helping in the study of drugs, but the molecules have to go through clinical trials. AI will design molecules, although, rightly, it is still a very regulated sector.

So, it will be a breakthrough that will be slow, because it has to be validated very well. But in the purpose of treatments, in the diagnosis, it will be increasingly common for Artificial Intelligence to be applied to subjects of images and radiographs. Sure they will be very good medical diagnoses, even better than those made by the expert doctors themselves. And it is something that is already done. But, the true contribution to the upcoming ten years will be that it will also be possible to explain how this diagnosis has been reached. Although Machine Learning and AI can now make these diagnoses, it is very difficult to explain this diagnosis, both to the professional and to the individual. In 10 years it can be explained. Both at a general level and in specific cases.

Artificial Intelligence will also help to suggest treatments. In 10 years, we will see it very deployed, and this will help the effectiveness of the doctor. 

Of course, we should not let AI leave in many facets without human supervision because technology will have its own limitations.

At this point, as in the case of driving, it will be other factors that determine whether we will actually use AI in the field of health. We generate a lot of our health data. All medical tests are digital, DNA sequencing, medical reports, and radiographs. It is a great source of unstructured data on which AI can interpret and find patterns. In radiology, it is already providing results. The algorithms can now detect tumors of different types with a precision superior to that of a radiologist.

However, that does not mean that all the medicine of the future will be based on Artificial Intelligence. It is not something simple because a single medical test does not have the answer, but several tests are needed, but the algorithms will help us to advance in this field.

Language, writing and other communications

AI will develop so much in the next few years that it will surely be difficult for us to distinguish when we are communicating with a human or a machine.

We will talk with machines as with humans, with a 98% reliability

One of the great promises seen is the ability to interact naturally with technology. Now we are doing it with assistants and phones. Over time it will improve more because there are many things that we still do not understand. The progress in natural language processing will be great.

Google Duplex might be much more widespread over more domains, and may have a natural language, have a conversation with a machine.

But this interpretation of natural language will not only be oral but also written. The machines of the future will be able to write and do quite well. They already write a text comment, but I would dare to say, being very adventurous, that they will write consistent and long stories.

The personalization of education

Another area in which AI will go a long way in the next 10 years is in the education field. In China, they are already conducting experiments, using AI to have personalized education and adapt the curriculums to each child.

Although we will have to see the results that these experiments offer. AI will surely have a great help for us in this sector. It will allow us to customize education, and that right now is not viable. We know that children learn with different styles, rhythms, and techniques, and there is a teacher every 30 students. It is very difficult to offer customization. But with AI, it can be done, also for early detection for difficulties such as dyslexia or disability, so that this concept disappears because it compensates for AI.

Biases? What biases?

Artificial Intelligence systems are based on correlations, in search of patterns. And that is precisely one of the great differences with Human Intelligence and what ends up causing some problems, such as the known biases. People differentiate cause-effect correlations, but not machines. That is why systems fail and have biases and are poorly configured and reinforce negative biases. It is what we have seen and do not understand relationships’ cause-effect today.

Therefore, an active area of ​​research in the advances of AI in data is to address the limitations of existing systems, such as discrimination and bias. The aspiration is that when using data, these algorithms do not suffer the biases that we humans have. We realize that the data is not an objective reflection of reality, but that they are biased. And if you teach an algorithm those biases, you can maximize them.

But, the good news is that the issue of discrimination and data biases can be resolved in 10 years. There will be many tools that can detect biases. It will be like engineering, and they will be able to identify very well and eliminate it. Currently, the problem is sometimes that you can have a set of data in which the race is not established, but it does have those biases. Maybe in 10 years, this issue can also be resolved.

Give me data, and I will give you customization

At this point, all AI is based on data. The most promising and the most successful commercial models are neural networks trained with huge amounts of data. To apply these AI techniques in new areas such as medicine and education or justice, the first thing is to have data. Algorithms learn from data that is supposed to reflect objectively what you want to model.

From that data, if they are enough for each person, you can build a specific model for each person. And adapt to that person. There are already personalization models in the recommendation systems based on your own data and behavior. Personalization is an immense area and is part of the core strategy of most technology companies, especially if they have services that interact with people. The key is how to apply them in non-purely digital sectors.

But they are not the only fields where the application of AI is a challenge. The implementation in robots, which so far has not benefited much from AI, will be a development field. There are many day-to-day tasks, even if they are less newsworthy. A good example is Amazon Go, which recognizes you when you enter and take the product. Although at the moment, it is a prototype, preparing AI to know how much to charge you for the weight of a specific product is a brutal challenge, especially to implement it at scale. But we will get it. Zara allows you to try things virtually in your home, and that suggests clothing based on your tastes and what you have.

Some of the research prototypes might even reap, within 10 years, such as writing an entire Harry Potter novel. Or algorithms that invent an entire movie. There are more dystopian issues such as the generation of fake news that can be done at scale, and that will exist both in text and video.

It’s not what I do, it’s what I should do

All this can be summarized in three main points

• It will create more jobs than it destroys.
• It will help humans in a good number of jobs.
• It will enable a fairer world with more possibilities for poor people who are able to strive in the study and improve their living conditions.

Technological development is much more complex than algorithms. It must be contextualized in reality and with an impact on society and the lives of millions of people. It cannot be developed in isolation. There is a human component. It will still be necessary to control and supervise Artificial Intelligence.

Beyond the development of new algorithms, we must do it in a framework; If not, it will not progress. Not all innovation is progress. That is why we must ensure that it is not discriminating, that it will not increase the difference between rich and poor, or that it only works for one race. 

AI has no superpowers

Surely AI has a lot of development potential, but there are things you won’t be able to do.

Which? The answer to this is related to the Moravec paradox. Apparently, the effort necessary to copy each human ability is proportional to the age with which it appeared in our family tree. “Love,” “creativity,” “empathy” are probably aspects of the human being that will cost to incorporate into an AI system. 

Almost all AI success is based on supervised machine learning with Machine Learning networks. But if you see how a baby learns, he knows nothing, observes, interacts, and receives feedback, and 18 months later, he has built his model of the world in which every concept he knows is anchored in reality. Learning this way is not possible with AI, but steps will be taken to give it. If achieved, it would be a great milestone, it would be to change from supervised learning with ML to unsupervised. We still don’t know how it works.

In addition, it will not have common sense, like humans. The machines do not know how to react, but let’s not rule out that it can be achieved.

AI is not going to be like a human. It will not turn against us. It will not have autonomy. We attribute human qualities to everything, we even say that they learn. But they don’t do it as a person. They make equations and come to a conclusion, but that is not learning. So they won’t take control. 

And, about the danger of stealing jobs, we tend to think too much about the threats of robots. It is believed that work is more likely to have an impact on low-skilled jobs, but it is not true. A robot cannot be a plumber, at least in 10 years.

Humans are much more than pattern recognition. We have many types of intelligence: common sense, ability to associate, create, semantic knowledge of the world that algorithms lack. They recognize cats, but they don’t know what a cat is. There are many areas where human intelligence is very complex and multidimensional, as in the aforementioned common sense and reasoning ability. But, they do have the fragility that you can hack.

Pattern recognition needs many more examples than a human to find that recognition. They do not learn incrementally and constantly and without forgetting what they have learned. They are supervised systems. Humans are more efficient, and there are many skills of a 2-year-old child that are very complicated for AI.