display

“I believe in the power of AI,” said EU Commission President Ursula von der Leyen.

We need exactly this spirit and not thinking in clichés like: "The computer has no consciousness" or "Super-AI would not be controllable."

Let's do the following thought experiment: We are gradually replacing all the neurons in a cat's brain with microchips.

The cat would probably still like fish to eat, chase mice, and probably be hungry and in pain as well.

But would this machine that is the cat's brain still have a consciousness?

The cat example illustrates an essential question of artificial intelligence.

It is similar to the “Ship of Theseus” paradox that scholars have argued about since ancient times: is the ship still the same after all the old planks have been replaced with new ones?

Does an object lose its identity if all individual parts are exchanged?

Finding an answer to this is difficult, if not impossible.

display

The development of a controllable so-called super-AI is also comparably impossible.

But don't worry, we already know that from everyday life: a word processor hangs up.

This is particularly annoying because you do not know whether an immediate restart of the computer is indicated or whether it is worthwhile to wait a little.

At such moments you want a program that says to our word processor - and any other program - “Yes, it will stop by itself” or “No, it will never stop”.

Unfortunately, such an assistant cannot be proven, no matter how ingenious classic computers or AI systems are.

That is why you cannot predict and control the behavior of “super AIs” - but luckily neither can that of us humans.

Even quantum computers, which are far superior to conventional computers in some tasks, would not be able to do this.

They are extremely helpful for "difficult" tasks, for example those that behave like yeast cells when brewing beer, the number of which doubles again and again - one cell becomes two, then four, eight, 16, 32, 64, etc.

display

At some point the runtime skyrockets and overwhelms classic computers - and us humans.

The current corona pandemic shows: We are simply bad at assessing exponential growth.

We are making use of this today with digital encryption, for example.

Even the fastest mainframe computers cannot crack them with today's technology if the code is big enough.

Because they try to find the right key by trying.

But this can take more experiments than there are atoms in space.

Quantum computers, however, which represent codes using atomic states, work with qubits.

Due to the strange laws of the quantum world, these can not only be 0 or 1, as in conventional computers, but simultaneously 0 and 1 or even in theoretically an infinite number of states in between.

Like a coin tossed in the air that spins on itself.

display

And as soon as you measure the condition of one qubit, the condition of the others is also known directly.

In this way, the quantum computer can try out all possible keys at the same time.

In the quantum world, a secret is simply no longer safe.

Even Einstein could hardly imagine this and spoke of a ghostly long-distance effect.

Quantum computers promise a speed advantage for research and economy.

That is why it is very welcome that we are now investing in quantum computers!

But the intoxication of joy must not cloud our view of the here and now.

In 2016, the scientists Iordanis Kerenidis and Anupam Prakash found a “simple” quantum algorithm for a supposedly “difficult” task in order to predict purchasing behavior.

But just two years later, student Ewin Tang found a classic algorithm that was just as fast.

Whether it is worthwhile to generally simplify the solution of difficult problems with quantum computers remains to be seen, especially since their handling is extremely complex: in order to be functional, they have to be cooled to near absolute zero and, moreover, set up vibration-free so that the qubits don't get mixed up.

The latter would be fatal, because as in football, we only want to find out who has won the side election when we catch the coin.

Catching not only requires conventional computers, but also a new type of programming.

In contrast to football, the trick is to toss the coin in such a way that in the end it is not a random result, but a meaningful result.

That is why we need an equally large investment in conventional computing infrastructures especially for AI, which are accessible to research and business.

This is the only way we can already play at the top of the world.

Kristian Kersting is Professor of AI and Machine Learning at TU Darmstadt, Co-Director of the Hessian Center for KI (hessian.ai) and recipient of the German KI Prize 2019. Christian Bauckhage is Professor of Computer Science at the University of Bonn, Scientific Director of Fraunhofer Center for Machine Learning and Lead Scientist for Machine Learning at the Fraunhofer Institute IAIS.