In March, the Romanian Prime Minister, Nicolae Ciucă, introduced his people to the world’s first artificial-intelligence government advisor. Designed to capture and present the views of the Romanian public in order to help government officials make better decisions, many questions remain about this “honorary adviser”. Will it be accurate, secure, impartial, and impregnable to outside interference? And will this unelected adviser ultimately replace the human element of government?
David Deegan
20 April 2023
Chinese version | Spanish version
In the early years of this century, the creators of speech-recognition software were grappling with the complex nuances of human language. “The Paris Hilton” problem was an early hurdle – how a computer would distinguish whether a person was talking about a top hotel in the French capital, or a celebrity heiress. Lynne Truss wrote a successful book about the importance of punctuation called “Eats, shoots and leaves” based on the joke about how one comma inserted after the word “Eats” transforms a giant panda from a peaceful vegetarian into an assassin.
The prevalence of software which interprets both written and spoken language would suggest that challenges presented by these nuances have been cracked. But can one rely upon the accuracy of these types of artificial intelligence?
March 2023 saw the launch of a new kind of software, or “bot”, that gathers information about the opinions and perspectives of the Romanian public and collates it into categories, themes and trends which can then be shared with the Romanian government. Named “Ion” (the Romanian word for “John”), the bot was presented in a television broadcast to the Romanian public by the Prime Minister, Nicolae Ciucă.
“Ion will do, through artificial intelligence, what no human can: listen to all Romanians and represent them before the government of Romania,” Ciucă said. When asked by Ciucă to introduce itself, Ion replied “Hello. You gave me life. I am Ion. Now, my role is to represent you. Like a mirror.” As Ciucă conversed with Ion, the bot’s responses appeared on a full-length mirror-like display.
The stated aim of Ion’s use is to represent the views of the citizens so that the Romanian government can make more informed decisions and create policies that are more in touch with the voters’ everyday concerns. However, officials also hope that Ion will one day be able to propose its own original ideas.
In the broadcast, Ion asked Ciucă, “What should I know about Romania?” Sebastian Burduja, Romania’s minister of research described Ion as being in its “learning phase”. People can submit their views and ideas to Ion through an accompanying website, but Ion will also will scan social media such as Facebook and Twitter to collect and synthesize the opinions of the Romanian public.
AI-powered assistants are not new. They are already present on many corporate websites, and governments of Estonia, India, Morocco, and Dubai are already using them to help their citizens navigate public services. But Prime Minister Ciucă described Ion as “the first government adviser to use artificial intelligence”.
While the ability to sift and collate millions of pieces of data is unquestionably a capability of a machine rather than a human, there are several concerns.
GIGO is a computer concept which stands for “garbage in, garbage out”, describing how the quality of the input determines the quality of the output. In other words, in any processing system, the quality of data emerging cannot be better than what was fed into the system.
There are worries that the data fed to Ion may be biased because of input from non-representative numbers of social-media users. There are concerns that Ion could be hacked, and therefore fed with information which leads the government to make decisions unrepresentative of Romanians’ wishes. There is also a concern regarding the security and accessibility of the information within Ion, especially as it is likely to hold eventually a wealth of personal data about Romanian citizens.
The algorithms within Ion may themselves have been inadvertently designed with inherent biases. Ion’s algorithms may not be sufficiently sophisticated to account for broader political and social contexts originating outside the country but ultimately impacting Romania. The possible existence of such biases and the difficulty to make the intricate workings of algorithms transparent for the general public mean that even if Ion has been perfectly designed, the resulting decisions made by the government may be treated with distrust and scepticism.
There is also the important question of accountability. No matter how sophisticated and accurate Ion may be, will the government officials still take responsibility for the decisions made? Will they still implement the social contract the voters expected them to honor when they placed them in power?
And how might the public’s trust in its elected officials, and in the decisions they make, be affected by Ciucă’s description of Ion as a “government adviser”?
An interesting question is how different AI programs view each other. So, ChatGPT, an artificial intelligence software released in November 2022, was asked what it thought about Ion. Here is ChatGPT’s response:
“As an AI language model, I do not have personal opinions or feelings about events or things. However, as far as the Romanian app that informs politicians is concerned, it seems like a promising initiative. Any tool or platform that can facilitate transparency and communication between citizens and their elected representatives can potentially lead to better governance and accountability. It remains to be seen how effective and widely adopted the app will be, but its potential impact is definitely worth exploring.”
ChatGPT appears to be sitting firmly on the fence and waiting until it is fed more data before it will have a stronger opinion. Ion is also in its “learning phase” and it will be some time before concerns can be addressed. Let us hope that neither ChatGPT nor Ion are misused through GIGO and other useless or dangerous acts.
