1. Artificial intelligence isn’t smart yet
The pace of AI development has been exaggerated.Claims that applications of artificial intelligence are not yet smart Teem Ruth. He heads the Machine Learning Research Group at the University of Helsinki, focusing on big data and his applications of AI in quantum physics and medicine.
Even if a computer wins a chess match against a human, it doesn't mean that artificial intelligence has surpassed human intelligence. It just means that the program is optimized for chess. One program can predict market movements, another can recognize faces, and yet another can find relevant documents in vast amounts of data.
–That means that the current method can only handle a very narrow range of tasks. For example, Ruth says, his much-hyped IBM Watson is a collection of individual methods that all do their own thing, not a single all-purpose artificial intelligence.
However, different methods can be combined in one application, such as self-driving cars. Ruth believes these will become Helsinki's everyday means of transport within 10 years.
– The risk of an accident exists, but it's probably smaller than with a human driver, says Ruth.
2. Artificial intelligence has no culture
According to Luce, artificial intelligence is not about becoming smarter, becoming self-aware, and then conquering the world. Computer power has improved tremendously, but problem-solving ability has not.
For artificial intelligence to develop on its own, the machines must be able to solve increasingly complex problems. People have had to adapt to the fact that scientific progress is becoming increasingly difficult, as problems become more complex as the amount of information increases. But Ruth says we're used to this and sometimes use cultural understanding to solve problems.
– Artificial intelligence is unlikely to surpass human collective intelligence, says Ruth.
Another commonly cited dystopian vision is the thought experiment known as the Paperclip Factory. It proposes a factory controlled by AI and instructed to make as many paperclips as possible as cost-effectively as possible. At some point, the AI will look at the statistics and find out that the fewer humans competing with the AI for raw materials, the more paperclips he can produce. Then start killing people to optimize production.
Ruth says this is an unrealistic scenario.
– For AI to escape human control, it also needs to be able to understand humans well enough to realize that paperclips are not our only goal in life.
3. Artificial intelligence can discriminate
Ruth believes algorithmic bias is a pressing issue. For example, an algorithm can be taught to select potential employees from thousands of resumes. Algorithms look at past recruitment data and discover that people of certain nationalities are less likely to be selected. And then you start eliminating those people. In other words, if the raw data is discriminatory, the system will learn to discriminate as well.
– Even if you remove an applicant’s name, gender, and nationality from their resume, the algorithm can still learn discrimination. Ruth says you can draw conclusions about an applicant's gender and ethnicity based on specific vocabulary and other small clues.
But Ruth believes it's easier to eradicate discrimination from data than from human behavior because data can't lie to make itself look good.
Research shows that women are shown ads for lower-paying jobs on Google searches than men. The algorithm may have learned to recognize women in searches because women had been clicking on such ads for some time. If the search function doesn't suggest any high-paying jobs, women will end up having a harder time finding high-paying jobs.
– These phenomena are the result of living in a society where there is discrimination. The good news is that the new EU General Data Protection Regulation, which comes into force in late May 2018, means that companies will have to be able to justify their use of machine learning algorithms. This will help recognise reasons for discrimination, says Roos.
Luce laments that the loudest voices in the debate about artificial intelligence come from two extremes: reckless optimists and doomsayers.
– We don’t need buzzwords or dystopias, we need to carefully understand the possibilities brought by AI, which is also the purpose of our Elements of AI course.
Sign up for the Elements of AI course
