Speakers:
Sascha Corti (Microsoft)
Kevin Mader (4Quant)
Martin Steiger (Lawyer)
René Pfitzner (Lead Data Scientist, NZZ)
Hans-Ruedi Früh (F&P Robotics AG)
Panel:
Harald Taglinger
Matthias Aeby
Entscheidungsfindung
At Liip we discussed decisions and AI. Algorithms are increasingly used for making decisions. They sort search results by relevance and create data visualisations. As invisible digital helpers they are already widespread. Compared to humans, algorithms have no trouble dealing with large amounts of information. AI, and deep learning in particular, has made remarkable progress in recent years across a range of cognitive abilities.
Sascha Corti gave an overview of the tools from Microsoft that make machine learning accessible to a broader audience. Systems with machine learning can reveal where a process can be optimised. To do so they draw on a large volume of data sets. From data on events that occurred in the past, they can predict the occurrence of a future event with a probability rate. One form of predictive software is known as predictive maintenance and is already used in the airline industry and by lift manufacturers.
Kevin Mader of 4Quant gave a brief overview of the state of AI technology. The field began around 1951, when Christopher Strachey developed a program that could beat him at draughts. Computers mastered increasingly complicated games. A few months ago, the program AlphaGo was able to beat the best Go players; Go is a game of high complexity that requires intuition to play. The big players such as Google and Facebook make their AI systems available as open source for developers, in the hope of owning the best-trained AI systems. The algorithms themselves are not very valuable in their own right — what is truly valuable is the data. Kevin then showed how AI technology is used today in medical research. It is a field where the tolerance for error is practically zero. Learning image-recognition AI can be trained on almost any task. With enough iterations it consistently achieves better results than human physicians. Yet when it comes to choosing the right therapy, the physicians still make the better decisions.
Legally, much in the field of AI is still open. Martin Steiger showed how to determine who bears responsibility if an accident happens. Under current law, for example, the product liability act comes into play. Here the manufacturer — or, if it cannot be identified, the supplier — is liable for a defective product, in principle regardless of fault. Such strict liability makes particular sense where products that are dangerous are brought to market. But how does one classify software and AI? Is it autonomous, capable of learning, but not capable of judgement or accountability? Is it like a pet, for example a dog, under current law? Or is it like a slave under historical, Roman law? And what ethical questions arise if AI is treated like a dog, a Roman slave or a modern wage slave?
René Pfitzner, Lead Data Scientist at NZZ, explained how he wants to use AI at NZZ in the future. He wants to develop an algorithm for reading recommendations. A filter bubble is created when an algorithm is designed so that ever more of the same is shown. That is why he wants to develop the algorithm so that it helps a reader discover new things. It is nonetheless not completely random, because a learning algorithm can recognise what a reader is certainly not interested in.
In the discussion, Harald Taglinger raised the question of how far AI has come. Looked at closely, the term artificial intelligence is not yet really apt. Current systems can learn rules autonomously, but they are not aware of what these rules even mean. The limits of what an AI system can achieve are currently changing very fast.
As Matthias Aeby added, we are not yet in a world in which AI systems make decisions independently from A to Z. Machines are now better than humans at analysing large amounts of data. They are, however, not yet fully convincing when it comes to bringing all the information back together into a synthesis and reaching a decision.
At the apéro it was noted that everyone's future salary will be determined by how well they can work together with such systems. AI is already a powerful tool for those who know how to use it. AI systems will obviously increasingly take over monotonous work from people. With self-driving cars, the lorry driver will no longer be a job of the future. But with every new technology it is not about human or machine, but about a fusion of the two. Kasparov plays even better chess when he plays with the help of a chess program.
The demo by Hans-Ruedi Früh (F&P Robotics AG) with his robot arm P-Rob was also fascinating. It showed how the robot can quickly learn new actions and, thanks to various sensors, recognises objects and stops in time should it touch a human. F&P Robotics uses the various AI platforms for object recognition, speech output and more. As these platforms develop rapidly, the robot's abilities expand continuously.
Explore more topics on how technology affects business, society and culture.
Don't want to miss our next event? Subscribe to the newsletter.