AI predictions may not always lead to better decisions2 min read . Updated: 26 May 2018, 01:41 PM IST
As machine-learning algorithms try and mimic human decision-making, they may exacerbate hidden human biases
Increasing use of artificial intelligence (AI) need not always lead to better decision making, argues a recent National Bureau of Economic Research (NBER) working paper by Ajay K. Agrawal, professor at the Rotman School of Management, and co-authors. While AI can improve predictions of future outcomes, it is not always able to judge which outcomes are better or which course of action to choose. This is because AI might not be able to account for all of hidden costs, rewards, and risks. Consequently, human judgement plays an important role in decision-making.
However, with improvement in technology, the onus of making decisions may be ceded to machines even when such decisions would be superior with human input. The authors argue that machine prediction is only as good as the information the machine relies on.
Hence, there are risks associated with relying on AI. As machines observe humans taking decisions, their ability to predict and themselves take decisions might improve. However, there remains a risk that by mimicking humans, machines might exacerbate the biases that often afflict human behaviour and judgement.
Also read: Exploring the Impact of Artificial Intelligence: Prediction versus Judgment (bit.ly/2kqyUTa)
Twitter bots might have impacted the outcome of the Brexit referendum and the US presidential election in 2016 in a decisive manner, according to a new NBER paper by Yuriy Gorodnichenko, professor at the University of California, Berkeley, and co-authors. The researchers analysed bots—non-human twitter accounts—and their tweets, re-tweets, identical tweets and other unusual activity. Humans were more likely to be influenced by bots when the information spread was in agreement with their views. The authors suggest that the pro-“leave" bloc in UK gained 1.76 percentage points owing to bots while Donald Trump gained around 3.23 percentage points in the 2016 presidential race.
Also read: Social Media, Sentiment and Public Opinions: Evidence from #Brexit and #USElection (bit.ly/2KTII3k)
A system of price deficiency payments for a few select crops will be more beneficial than a minimum support price (MSP) system, according to a new article in the Economic and Political Weekly. The authors, T. Haque, chairman of the Land Policy Cell, NITI Aayog, and P.K. Joshi, director of the International Food Policy Research Institute (South Asia), argue that such a policy shift will improve farmers’ income and also reduce the government’s subsidy burden. The current system is ineffective because MSPs are fixed based on all India weighted average costs while cost of production varies regionally, they argue.
Also read: Price Deficiency Payments and Minimum Support Prices: A Study of Selected Crops in India (bit.ly/2sdCftt)
A new paper from the National Institute of Public Finance and Policy (NIPFP) raises concerns about the rise of government-funded health insurance schemes in India. Authored by Ila Patnaik, professor at NIPFP, and co-authors, the paper finds that 48 schemes have been launched between 1997 and 2018. This is an outcome of the fact that the government has accepted that private healthcare providers have come to dominate the market. However, the authors caution that the government has not done enough to address market failures. For instance, claims ratio in some schemes have crossed 100%, raising the risk that the ultimate cost of bailouts of health insurance companies may have to be borne by the government.
Also read: The rise of government-funded health insurance in India (bit.ly/2KVuAH5)
Economics Digest runs weekly, and features interesting reads from the world of economics.