The finance minister was killed by a cleaning robot when it blew itself up next to her.The who became what?Well, Dr.Brenda Gusmile, the Federal Republic's chief housekeeper!Killed by a SweepBot equipped with artificial intelligence (AI), facial recognition and explosives, who was able to sneak into the Ministry because he looked exactly like the robots that patrol there.Scenarios for the next five yearsA fictional story like this emerges when you have 20 researchers and security experts from, among others, the Universities of Oxford and Cambridge, the OpenAI initiative and the US civil rights organization EFF (Electronic Frontier Foundation) ponder the use of AI for menial purposes.Photo gallery: These jobs are endangered by machinesIn a 100-page study, they present several scenarios, such as the assassination attempt on the imaginary German minister.The aim is to warn politicians and the research community in particular: AI could also pose a danger if it is further developed without additional regulatory, organizational and technical security measures.Among other things, they describe how criminals are constantly finding new security gaps in corporate networks using technology that works in a similar way to the Go-playing Google AI AlphaGo.This is how the fictional criminals manage to spread ransomware called WannaLaugh and blackmail millions of people.In another scenario, AI software is used to develop a perfect phishing attack based on publicly available information.Inceptionism: Psychedelic Images from ComputersThe imaginary examples are intended to illustrate how criminal activity could be taken to a qualitatively new level with the help of machine learning and automatic pattern recognition - just as these techniques are intended to do so for good.Attacks that already exist today will be developed further, new and previously unrealistic attack techniques will emerge, experts believe.The latter in particular is likely to be controversial among other experts.Powerful corporations from the USA and China are among the leading AI drivers whose systems potentially affect hundreds of millions of people.Initiatives such as AlgorithmWatch in Germany and the American AI Now Institute are already demanding more transparency from companies, not less.But a muzzle might also be unacceptable for many companies and scientists.After all, a lot of what they work on still belongs to the area of ​​basic research.Without publications and lively exchanges, progress would be artificially slowed down.The authors are not concerned with demonizing the techniques and applications from the AI ​​​​area, but treating them as dual-use goods ("with dual use").Just like chemicals or nuclear technology.Parcel drone from Amazon: Quadrocopters with autopilot could eventually make the classic parcel and postman superfluous.Semi-automatic trucks: Computer-controlled machines are also making rapid progress in the transport sector.Millions of truck driver jobs are at stake.Window cleaners in the Reichstag dome in Berlin: Robots could soon take over the jobs of cleaners.Facebook Messenger: Chatbots are intended to make traditional call center agents largely superfluous.Cancer center in North Carolina, USA: more and more clinics are cooperating with IBM's artificial intelligence Watson.The computer explores huge amounts of data and suggests possible therapy approaches for their patients to the doctors.3D surgical robots on the Charité campus in Berlin: there are studies that show that computers will take over 90 percent of the work of doctors.Financial start-up in London: Numerous jobs in controlling and customer service are likely to be replaced by software soon.Songwriting app Aitokaiku: Even creative tasks are being mastered better and better by machines.Experts believe that ultimately there is hardly any profession that is safe from automation.As a rule, artificial intelligences are not evil.Developers at Google have instructed a neural network not to stop computing until it recognizes more - whatever it may be.Can more things be seen in this photo than a river and a sky, mountains and trees?And this is what the neural network sees in the seascape when it is commissioned with the interpretation: Hybrid creatures with eyes all over their bodies fly in the sky.Pagoda-like buildings and something like vehicles can be seen on the shore.Replication of a neural network in the brain: nerve cells are connected to nerve fibers, excitation propagates from cell to cell, the importance of the connections changes over time.Artificial neural networks also learn according to a similar model.The results of the experiment are a lesson in the calculability of the fantastic: Leaves become birds, hills become picturesque roof domes.The photos are now reminiscent of pictures by artists such as Max Ernst or Salvador Dalí.An old children's game: recognizing figures in clouds.This game is also mastered by neural networks: You then see birds, foxes, rabbits, bizarre hybrid creatures - psychedelic motifs.The normal cloud image on the left turns into a strange psychedelic work of art if you let the network calculate long enough.The resulting mixed creatures are reminiscent of creatures from paintings by Hieronymus Bosch or Max Ernst.Ernst even used a technique that is distantly related to the one used here: for example, he hatched the structure of a wooden floor onto a piece of paper, tried to identify living beings in the resulting pattern, and then enhanced them so that others could see them too .Landscape image of mountain range: What does the network see here?After editing: Strange architecture in the mountainsides, pagoda-like buildings on the ground, human-appearing figures and strange hybrid creatures.Mountain stream: What do the higher layers of an image recognition network see in this image?Dog creatures, trees with snouts and eyes, eyes everywhere.Even in what it sees, the network's perception is similar to that of a human.We too have a tendency to perceive faces first in unstructured visual stimuli.Interpretation of an art photo of a tree with red foliage: chimera creatures in the sky, ghost cars at the edge of the field.Another interpretation of the same template: dogs, bird creatures, human figures in the sky.Knight with a dog's head: The original here is actually not a classic oil painting, but a photo from a medieval festival.The knight has neither a dog's head on his saddle nor an animal-shaped glove.Pagoda landscape: This picture, like the following one, came about in a different way.An image full of pixel noise was presented to the network, all that is seen here is the network's unguided interpretation.Here, too, the starting point was random image noise.The network turns it into a crazy architecture with Romanesque arches.Another picture out of the chaos: Pixel noise turned into a magical ice landscape.In fact, the network's interpretation is accompanied by another - that of the viewer, who in turn extracts meaning from the computer's creation.Sign up and join the discussion