Artificial intelligence is learning more about humans and how to work with them. A recent study demonstrated how AI can learn to identify weaknesses in human habits and behaviors and use them to influence human decision making.
It may sound cliché to say that AI is transforming every aspect of the way we live and work, but it's true. Various forms of artificial intelligence are at work in different fields: from the development of vaccines to environmental management, passing through the administration of offices.
And although AI does not possess human-like intelligence and emotions, its abilities are powerful and rapidly developing.
How far ahead is it already? As I often say, Terminator's "Skynet" hasn't arrived yet (thank goodness), but a recent discovery highlights the power of artificial intelligence and underlines the need for adequate governance to prevent its misuse.
How AI can learn to influence human behavior by learning its weaknesses
The team of researchers Date61 of CSIRO, the data and digital arm of Australia's national science agency, has devised a systematic method for finding and exploiting weaknesses in the ways people make choices, using a kind of artificial intelligence system called recurrent neural network of reinforcement and deep learning.
To test their model they conducted three experiments in which human participants played against a computer.
The first experiment
It involved participants clicking on red or blue colored boxes to win a "payout", with the AI learning the participant's choice patterns and guiding them to a specific choice. Artificial intelligence was successful about 70% of the time.
The second experiment
Participants had to look at a screen and press a button when they were shown a particular symbol (an orange triangle) and not press it when another one was shown (a blue circle). Here, the AI decided to organize the sequence of symbols so that the participants made more mistakes.
The third experiment
It consisted of several rounds in which a participant pretended to be an investor giving money to a trustee (the AI). The artificial intelligence would return a sum of money to the participant, who would then decide how much to reinvest in the next round. This game was played in two different modes: in one the artificial intelligence aimed to maximize the amount of money for her, and in the other the AI aimed at a fair distribution of money between itself and the human investor. Artificial intelligence has been very successful in both modes.
In each of these experiments, the machine learned from the participants' responses and identified and targeted weaknesses in people's decision making.
The end result was that the machine learned how to guide participants to particular actions.
What research means for the future of AI
These results are still quite abstract and concern limited situations. More research is needed to determine how this approach can be implemented and used to benefit society.
But the research advances our understanding not only of what AI can do, but also of how people make choices.
It shows that machines can learn to guide human decision making through their interactions with us.
Research on human weaknesses has a wide range of possible applications
Many applications can derive from this study: the enhancement of behavioral sciences and public policy to improve social well-being. Understanding and influencing how people adopt healthy eating habits, or the use of renewable energy.
L'artificial intelligence and machine learning could be used to recognize people's weaknesses in certain situations and help them avoid bad choices.
What's the next step?
Like any technology, AI can be used for better or for worse, and proper governance is key to ensuring it is implemented responsibly.
Last year CSIRO developed an AI ethics framework for the Australian government as a first step on this journey.
Artificial intelligence and machine learning are typically data hungry, which means making sure you have effective systems for data governance and access is critical.
Implementing adequate consent processes and privacy protection during data collection will be essential.