In 2012, Obama wasn’t the only person who won in the presidential race. The other person was Nate Silver. His models had correctly predicted the voting outcomes in every state plus the District of Columbia, achieving the godlike 51/51 score. In the 2008 elections, he scored 50/51. The Internet went wild. They offered him the reverence that had previously been reserved for Chuck Norris. Gawker called him “America’s Chief Wizard.” Gizmodo published 25 Nate Silver facts, one of them being “Nate Silver’s computer has no backspace button; Nate silver doesn’t make mistakes.” Is that the hint to an AI powered democracy?
Internet sensation aside, his models are nowhere near perfect, as shown in recent predictions. However, his success in the two previous general elections brought up an interesting hypothesis. We are getting better and better at predictions. Supposed that, one day, we can build a perfect model that can predict with a pinpoint accuracy. Will it be feasible to use prediction results instead of holding actual elections?
This fantasy is not new. As early as 1955, the legendary science fiction writer Isaac Asimov had envisioned something similar which he called the “electronic democracy.” In his short story “Franchise”, he envisioned a future in which a computer selects a single person to answer a number of questions, then use the answers and other data to determine what the results of an election would be, avoiding the need for an actual election.
This fantasy has a tremendous financial appeal. Imagine how much money could be saved. The cost of holding a local county ballot can go up to hundreds of thousand dollars. Given that there are more than 3,000 counties in the US, the cost of administering a general collection can go up to hundreds of million, if not a billion, dollars. This money could have been spent on education or healthcare instead.
It also has a certain intelligence to it. From Brexit to the rise of Trump, it’s obvious that humans cannot be trusted to make important decisions. In his book “The Myth of the Rational Voter”, Bryan Caplan, Professor of Economics at George Mason University, challenges the notion that voters are reasonable people that society can trust to make laws. We are ridden with biases, some of them, according to Caplan, are make-work, anti-foreign, pessimistic, and anti-market. We are also irrational, as outlined in the book “Predictably Irrational” by the MIT professor Dan Ariely.
Even if we can be trained to overcome those biases and be rational, we are simply not smart enough. We are incapable of processing a large amount of data. We either don’t have access to all data or we choose not to observe all the data, which often leads to the situation in which we are forced to make decision with the lack of information.
Selective exposure theory refers to individuals’ tendency to favor information which reinforces their preexisting views while avoiding contradictory information. For example, if you’re a Republican, you are most likely to watch Fox News, read newspapers similar to the Boston Herald, and have civil conversations with other like-minded Republicans. Google will likely to show search results that reinforce your point of view and your Facebook newsfeed will only display updates that you might like. Thus, we are more likely to make decisions based on information we have gathered from one side of the matter.
Worse, we sometimes make decision without even knowing what the decision is about. For example, we are given the power to vote for an economic bill without having a fundamental understanding of the economy. One startling example of blind voting is the aftermath of the Brexit. Hours after the Brits voted to leave the EU, they frantically googled what the EU was. According to Google, “What is the EU?” was the second top UK question on the EU since the #EURefResults were officially announced.
Humans are also terribly short-sighted. Few are endowed with the ability to make predictions that incorporate more than just mere intuitions. We need a calculator to help us figure out 13 x 56, for god’s sake, let alone running all possible scenarios in our head to see what scenario is the most likely or the most beneficial.
If humans should not be trusted with making important decisions for ourselves, then who should?
What other intelligent beings are there on earth, if it’s not Artificial Intelligence?
Computers are named computers for their computing power. They can process a large amount of data, not only current data but also data from the past. They can look at a problem from both the republican point of view, the democratic point of view, the independent voter’s point of view, then synthesize all the information get a subjective point of view. They can examine different scenarios for the future, what will happen if it chooses a certain candidate, or if a certain bill is passed, and make decisions that maximize our utilities.
But is it a possible scenario? Unlikely. Let’s face it, even though we are ridden with biases, humans like to think that we are the most intelligent bodies on Earth. It would take incredibly disenfranchise people to have something else make decisions for them. Also, it seems so ethically wrong to strip people of their voting power.
Second, it’s the fault in the design. Predictions are not the answers to a yes or no question. They are probabilistic. If the model says that candidate A has 70% chance of winning and candidate B has 30% chance of winning, it doesn’t mean that A will win with 70% of the votes against B’s 30%. It means that if the election were held 100 times, A would win 70 times, and B 30 times. Whether this particular election would belong to A’s 70 times or B’s 30 times is a question whose awe don’t know the answer until the actual election is carried out.
Moreover, the very system that AI is trying to replace makes decisions whether to incorporate AI. Would politicians voluntarily vote to pass a bill that would undermine their policy-making power?
I could only hope.