Superintelligence - SAP chases the moving train
Last summer, there was hope: Together with the weekly newspaper Die Zeit, SAP sponsored an AI conference in Berlin. The event brought experts and critics from a wide range of disciplines on stage. This demonstrated that AI is not an orchid subject in computer science.
At the latest since the revolutionary success of the Google machine playing Go, AI has also become mainstream. Watson, IBM's AI masterpiece, has been admired and discussed for many years.
For Microsoft and Amazon, machine learning is a given. Online courses and white papers prove it. After TechEd Barcelona 2016 and almost simultaneously with the Select event in Berlin, SAP has now also launched its first machine learning course. Taking into account other activities, such as the Hana framework PAL with function modules for building a simple neural network, SAP can be said to have taken the right direction.
Even if the AI trend was recognized by SAP in a rudimentary way, the conclusion must be drawn after the end of the three events - Time Conference on Artificial Intelligence, TechEd Barcelona and SAP Select Berlin: Those responsible at SAP have not understood the technical, organizational, business, financial and social significance of AI.
The direction is right, the intellect fails.
The Dartmouth Conference, which took place on July 13, 1956, is considered the birth of Artificial Intelligence (AI). To mark this 60th anniversary, Die Zeit, in cooperation with SAP and the German Research Center for Artificial Intelligence, organized a conference that took a controversial look at the potential of AI for digital society and business.
One of the most interesting speakers was Professor Jürgen Schmidhuber, scientific director of the Swiss Research Institute for Artificial Intelligence (IDSIA).
Since he was a teenager, Schmidhuber has wanted to build a self-improving AI smarter than himself, then retire to watch AIs colonize space. The powerful feedback neural networks of his research groups at TU Munich and the Swiss AI Lab were the first to realize handwriting recognition, speech recognition, machine translation and automatic image description, and are now accessible to billions of users through Google, Apple, Microsoft, IBM, Baidu and others.
This year saw another milestone on the road to AI that is smarter than humans: for $600 million, Google bought the British company DeepMind, which subsequently presented a Go program and beat the world's best Go player. DeepMind was heavily influenced by former PhD students of Professor Schmidhuber. Two of the first four DeepMinders studied in his lab.
So you don't have to go to Silicon Valley, the leading U.S. universities or Japan to get excellence in artificial intelligence. But a visit to SAP's Select executive event in Berlin once again brought the big disillusionment: Invited by SAP as a guest speaker was Nick Bostrom, Professor of Philosophy at St. Cross College, Oxford University. He became known to a broad public through his bestseller: Superintelligence, scenarios of a coming revolution.
In no other field of computer science is development taking place at such a rapid pace.
Nick Bostrom, written in 2014 (Superintelligence, page 30)
"In 2012, the Zen series of Go programs reached 6th dan (the level of a very strong amateur player) in rapidly played games [...]. Go programs have improved by about one dan per year in recent years. If it stays that way, they should reach the world championship in about ten years."
This year, an outstanding Go-playing computer made moves that no human could analyze. Go experts stood by in amazement and watched the computer win. The SAP community needs a world ERP leader who is serious about embracing the AI trend, who is willing to invest, and who wants to do extraordinary things.
To develop a cloud model, like hundreds of other companies, to develop a new database technology is too little. Superintelligence is needed for SAP!
SAP should surround itself with the right people: Professor Jürgen Schmidhuber's team created the world's first deep learners, which won object-finding and image-segmentation competitions, including for early cancer detection.
His group created the first methods that learned control strategies directly from high-dimensional video input without a teacher. His team continued to establish the field of mathematically rigorous universal AI and optimal universal problem solvers. He generalized algorithmic information theory and also many-world theory of physics to obtain an elegant minimal theory of all constructively computable universes.
Professor Schmidhuber is president of the company Nnaisense, which wants to create the first practical general-purpose AI. That would be super intelligence for S/4 Hana!
