Artificial intelligence

All experiments were not artificial intelligence, but merely attempts to simulate human intelligence with machine algorithms.

This imitation works surprisingly well: Since 1997, the computer "plays" chess better than any human. It was made possible by powerful hardware and excellent algorithms, both designed by humans.

Since mid-March of this year, there has been artificial intelligence - intelligence that is no longer derived from humans. It was not created by a programmer. No one knows the algorithms. And it works better than any chess program.

This intelligence has won the victory in the board game Go against the world's best player. The tournament ended 4 to 1 for the computer.

Go is much more complex than chess. There are more game variants than atoms in the universe.

The AlphaGo machine does not carry verifiable algorithms constructed and tested by humans, but it actually plays Go. This machine comes from Alpha (ex-Google) and was built by the company Deepmind, which was bought by Google two years ago.

Why does AlphaGo win?

The secret is a neural network, the only thing about the machine that might still be human. The computer scientists copied the principle from the brain and simplified it:

On one side there is an input, e.g. mailboxes; connected to this is the neural network - a network of millions of nodes and connections capable of learning; at the end there is an output, perhaps a printer.

If you put your tax documents in the mailbox and start the machine, the printer will definitely print scrap. You give the machine a kick and say "again".

Now the learning ability of the neural network comes into play: The machine remembers how it came to the junk result and will certainly not go down this random, self-selected path again.

Of course, the second attempt is also absolute junk, but after a few million attempts, a number could be closer to a correct tax return. One can provide tutoring by providing correct tax returns as samples. Once a correct result is found, the neural network learns from this success.

Based on thousands of sample games, AlphaGo learned the board game itself. The machine played against itself over and over again.

Thus, an artificial algorithm has grown up that no programmer could ever construct. Observers of the tournament saw moves that a human would never make. And the winner was AlphaGo!

The SAP Hana database has a graph engine that builds networks of nodes and edges (connections), and the Hana framework PAL (Predictive Analysis Library) has neural network capabilities.

It's time to start a discussion about AI in the SAP community as well: Is the takeover of the machines imminent?