In 2015, a number of famous scientistsAI entrepreneurs and investors have signed an open letter calling for greater attention to the safety and public benefit of AI developments. British theoretical physicist Stephen Hawking and founder of SpaceX and Tesla Elon Musk are among the signatories. It is noteworthy that Hawking saw AI as a threat, which he wrote about in his books and talked about in numerous interviews. According to the scientist, "the emergence of a full-fledged artificial intelligence can be the end of the human race." Fortunately, not all researchers share this apocalyptic point of view, but we still have something to worry about - as the results of a recently published study showed, AI can learn to identify vulnerabilities in human habits and behavior and use them to influence decision making. Of course, we are far from the uprising of machines, but AI is already transforming many aspects of our lives today. And although AI does not have human intelligence and does not experience any emotion, its capabilities are rapidly evolving.
What is modern AI like?
Today, various forms of AI operate in suchareas like vaccine development, education and administration, and learning-capable voice assistants like Alice or Siri are unlikely to surprise anyone. AI gradually penetrates not only the working, but also the daily life of human society, managing many processes and helping scientists to conduct research in various fields of science. Recently, a team of Australian researchers demonstrated how an AI system can be trained to manipulate human behavior and influence decision making.
In a study published in the journalPNAS, scientists at CSIRO Data61, the information and digital arm of the Australian national science agency, developed and tested a method for finding and exploiting vulnerabilities in human decision-making using an artificial intelligence system called a recurrent neural network. In three experiments in which a person confronted a machine in a computer game, the researchers showed how AI can be trained to recognize vulnerabilities in human habits and behavior and use them to influence decision-making.
For even more fascinating articles on how AI develops and functions, read our channel in Yandex.Zen. There are regularly published articles that are not on the site.
In the first experiment, people clicked on the redor blue flags to earn in-game currency. AI studied their choice models and began guiding them to make specific decisions with an approximately 70 percent success rate. Small fry, but just the beginning.
In the next experiment, participants were askedpress the button when they have seen a certain symbol (colored shape), but refrain from pressing the button when shown other symbols. The "goal" of the AI was to arrange the sequence of characters displayed to the participant in such a way as to trick him into making mistakes, ultimately increasing the number of mistakes made by humans by 25%.
In the third experiment, a person (player)pretended to be an investor giving money to a proxy (AI) who would then return a certain amount of money to the player. Then the person had to decide how much to invest in each subsequent round of the game, based on the income received from each "investment". In this particular experiment, the AI was given one of two tasks: either maximize the amount of money it made, or maximize the amount of money that both the human player and the machine received.
“In each experiment, the machine learned fromparticipants' responses and identified vulnerabilities in the decision-making process. As a result, the machine learned to direct the participants to certain actions, ”write the authors of the scientific work.
The future of artificial intelligence
If machines can beat us at games, doesare they smarter than us? The findings of the study are still rather abstract and relate to limited and unrealistic situations. So how this approach can be implemented and used for the good of society remains to be determined.
But this research is really advancing our understanding of not only what AI is capable of, but also how humans make choices and decisions. Moreover, it shows that machines can learn to manipulate human choice simply by interacting with us... If you think about it, real researchhas a huge range of potential applications, from advancing behavioral sciences and public policies to improve social well-being to understanding and influencing what dietary habits people adopt or what renewable energy sources are used. Moreover, AI and machine learning can be used to recognize the vulnerabilities of people in certain situations and help them avoid bad choices.
See also: How does artificial intelligence help protect animals?
According to The Conversation AI can be used like any other technology - for good and bad, which means competent management of AI developmentis critical to ensure its responsible implementation. Interestingly, in 2020, CSIRO developed an AI Code of Ethics for the Australian government with the aim of establishing proper AI governance. So we can assume that the first step on this difficult and, I am not afraid of this word, thorny path has been taken.