The limits of artificial intelligence.
Marc BothaBlockedUnblockFollowFollowingFeb 11When large amounts of data and many factors come together, artificial intelligence is superior to human intelligence.
However, only humans can think logically and distinguish between useful and worthless AI advice.
“Teach us to think that we must die”this is what the Luther Bible from 1912 says in Psalm 90.
American tech stars such as Oracle founder Larry Ellison, Amazon boss Jeff Bezos or the investor legend Peter Thiel do not believe in such humility, but instead, let billions of dollars go to decipher the mystery of why people age and die.
If the three business leaders blindly believed the computer, there would be a more pleasant way for them to at least postpone death: drink more champagne.
Because the analysis of many data and influencing factors provides the clear connection that with increasing champagne consumption the life expectancy rises.
Although it is not well known how much Dom Pérignon the billionaires drink per week.
What is certain, however, is that each of them ignores this connection.
For it is not the supposedly protective effect of sparkling wine, but the prosperity of its drinkers, which promotes the consumption of the noble drink, paired with better health care, that makes the notorious friends of the sparkling wine live longer.
This champagne trap, as I call it, is characteristic of the misunderstandings and myths that circulate about artificial intelligence (AI), but also of its strengths and benefits that man can take advantage of.
For artificial intelligence is nothing more than a very specific form of learning, namely machine learning.
This learning is limited and unlimited at the same time.
Unlimited, because machine learning is vastly superior to the physical learning of the human brain, as more and more powerful computers can perform more and more operations in tiny fractions of a second.
Thus, machine thinking provides people with patterns that they can never recognize or only recognize in an unacceptably long time.
However, machine thinking is limited because a computer only detects patterns.
The sense and logic behind it — see the champagne trap — can only be recognized by humans.
Not intelligent, but incredibly fastNevertheless, such patterns enable insights that purely logical thinking can hardly achieve due to the complexity and amount of data.
This is precisely the new quality that distinguishes artificial intelligence from earlier forms of digitization.
The best way to illustrate the difference is to use IBM’s Deep Blue supercomputer.
The computer giant beat the world chess champion Garri Kasparov in 1996.
However, Deep Blue was not intelligent.
It was incredibly fast.
He only calculated probabilities for the next moves based on the respective game situation and the programmed logic of the chess game.
He started the next game at zero again, without having learned anything from the previous one.
In artificial intelligence, on the other hand, the computer is not programmed with logic.
Instead, it works according to the principle of “trial and error,” i.
, it sorts out or re-sorts the unsuitable and enriches what has been found so far.
In this way he allows the human being to recognize connections, to draw conclusions and to make predictions without knowing the underlying logic.
To exaggerate, artificial intelligence could be compared to the learning behavior of a dog or a toddler.
The child learns to avoid a stove, and the dog learns to keep quiet in the train compartment.
Both do not know that there is a danger behind it from heat and consideration for people — and are therefore not in a position to transfer the learned behavior to the use of an iron or other crowds of people.
Only with age does the child learn to think logically and apply this to new phenomena through transfer.
Gaining knowledge from statistical correlations.
Generating patterns has excellent advantages in a highly complex world.
When the number of influencing variables, also called decision variables, increases explosively, it can take an unbearably long time to track down logic to solve a problem.
In order to win a customer, for example, a salesperson would have to have a kind of master formula according to which the first records all conceivable influencing variables such as gender, hobbies, preferences, consumption and much more and then derives them from them: If a, b, c, d and so on are available, then y or z follows from it.
The enormous amount of data alone prevents a seller from creating such a master formula.
With artificial intelligence, on the other hand, the computer is flooded with all available data from all customers.
By the purchasing behavior of many thousands of customers, the computer then determines behavior patterns that can be used to predict future purchasing decisions quite well.
The reason for this is of secondary importance.
Machine learning, therefore, means gaining knowledge through statistical correlations instead of logical conclusions.
A few months ago, for example, a system of artificial intelligence beat very experienced cardiologists in the interpretation of ultrasound images of the heart.
These consist of a large number of layers and individual images, which makes it very difficult for doctors to precisely locate the examined sites.
The system was fed with over 200,000 detailed images of 267 patients.
In the end, the computer was able to determine the examined site of the heart with an accuracy of 91.
7 to 97.
Despite years of experience, the cardiologists only managed 70.
2 to 83.
Artificial intelligence is now proving very successful in predictive maintenance in mechanical engineering.
The German ThyssenKrupp Group, for example, continually records all operating data from thousands of elevators and uses artificial intelligence to determine patterns according to which systems have failed.
Since several factors often come together, the logic behind this can hardly be determined with justifiable effort.
The patterns that could be determined using artificial intelligence prove to be so accurate that service teams can increasingly intervene before an elevator breaks down and the damage is unusually large.
AI top discipline: self-propelled cars.
Direct cooperation between humans and robots will only work with artificial intelligence.
The joint assembly of a car or a machine, for example, can only succeed if the robot learns to adapt to its human counterpart, his habits and his pace of work.
To do this, the so-called Cobot does not have to know the reasons and motives for the worker’s behavior but does have to filter out patterns, such as signs of fatigue, to react to them.
The supreme discipline, in which machine learning will and must far exceed human comprehension and speed, is the self-propelled car.
Here, vehicles must acquire knowledge about the environment and be able to interpret it correctly.
This is not possible with logic, but with pattern recognition — whether with the child playing on the side of the road or with the car approaching, whose driver is keeping wrong track.
We call this anticipatory and risk-adjusted driving.
The Fraunhofer Institute has developed the Shore software library, which recognizes people by gender, age, and emotions.
The correct use of artificial intelligence will give us slippery productivity advantages and new opportunities.
Smart grids, for example, will orchestrate the production of green electricity by analyzing the weather, expected electricity demand and storage capacities for patterns that enable optimal coordination between consumers and producers.
Where decisions are made from statistical correlations, however, human quality monitors, also known as quality gates, are needed.
They must check the plausibility of such decisions.
Data interpretation without the logic check can be dangerous, especially if decisions affect human lives.
Who is responsible if artificial intelligence makes a false medical diagnosis based on automated evaluation of patient data?.Do autonomous vehicles have to take out their insurance?Pattern recognition, i.
, artificial intelligence, is a sharp tool for solving problems more efficiently in a highly complex world.
However, this is not possible without common sense and human control.