CUHK Business School Researcher Shares his View on AI and its Limitation

March 06, 2019 - 04:48
CUHK Business School Researcher Shares his View on AI and its Limitation

HONG KONG,CHINA - Media OutReach - March 6, 2019 - Artificial intelligence (AI) isinevitable. Coinedin 1955 by American scientists to describe a new computer science division, AIhas successfully penetrated into our daily lives: When you pick up yoursmartphone and talk to Siri -- that is AI at work. When you text with a customerservice chatbot on Amazon, you are also being served by AI technology.


Whether we are enjoying the convenience brought byAI technology or worrying about when machines will eventually steal all ourjobs, there is no way for us to stop seeing its development.


As ubiquitous as it is, AI is not able to solveeverything, at least not yet.


Prof. Michael Zhang, Associate Dean (Innovation andImpact) and Professor of Department of Decision Sciences and ManagerialEconomics at The Chinese University of Hong Kong (CUHK) Business School pointsout that the applications of AI in computer sciences are different from thosein economics.


"There is a big difference between howcomputer scientists and economists look at technological innovations.Currently, artificial intelligence is used more for prediction andclassification. But, the focus of economists' research is to find causalrelationships and explain the underlying mechanisms of things," says Prof.Zhang, adding that interesting results can be produced if artificialintelligence and some of the data analysis models built by economists can becombined.


Unpredictable Events in Finance

In thefinancial market, it is often difficult to identify the relationship betweencause and effect, as unpredictable events still happen despite the usage ofcomplex financial models. That's why he thinks researchers should not assume AIcan solve everything.


"Infinance, there are multiple levels of risks. Once the volatility is writtenout, we have already assumed a certain probability distribution. But in manycases, we cannot possibly know about the probability distribution of financialevents, such as the stock market," he says.


He citesthe example of the collapseof Long-Term Capital Management L.P. (LTCM) and its fund in the late1990s, leading to an agreement among 16 financial institutions for a US$3.6billion bailout under the supervision of the US Federal Reserve.

This andother unexpected events in the financial market including the 2008financial crisis are termed as "black swan"events, which are random and unpredictable.


Traditionalrisk analysis approach assumes probability as independently and identicallydistributed. However, Prof. Zhang believes that this traditional approach isinefficient in explaining the past black swan events. Hence, without changingthe approach, the use of AI will yield very little success in predictingevents.


"Thereare no physical laws governing social actions, and in many real-life situationsthe probabilities of events are not available. In the field of social sciences,therefore, in addition to outcome uncertainty (risks), decision-making tasksare often subject to distribution uncertainty (ambiguity)," he says.


From hisperspective, black swan events seem surprising only because past researchersneglected the influence of ambiguity in the financial market. This concept iswhat he was explaining in his working paper titled "Statistical Inferencewith Ambiguity" in which he suggested a model that incorporates ambiguityin statistical inference to study uncertain events.


"Whenambiguity is considered, the derived uncertainty is demonstrably larger thanthe case only when risk is present. The derived uncertainty grows at a fasterspeed in ambiguity than in risk. So when ambiguity dominates, previouslyrejected hypotheses can no longer be rejected," he explains.


In hisstudy, the team used mathematical formulae to show that the confidence levelplunged from 95 percent to 50 percent even with a moderate increase inambiguity. This explains why black swan events happen even when the probabilityof such events is very small. Based on this theory, the next step is to findout the measurement for ambiguity -- and that is where AI comes in.


"We can useAI technology to identify if the underlying distribution will change. If we canidentify the change, then we know our underlying model needs to be changed. Assuch, we are trying to create a measurement based on AI algorithms," Prof.Zhang explains.


Will China Bethe Future Leader of AI?

The U.S. and China are considered to be leading the race in artificialintelligence of the world. In 2017, Chinahas announced its ambition to be the world leader in AI and thecountry's AI industry is worth almost US$150 billion. In addition, Chinaprovided 48 percent of the total equity funding for AI startups globally.


How will AI technology grow in China and compete with the U.S.?


"In the near future, I think the AI development in China will bevery fast. But there is a huge difference between the development in China andthat in the U.S. as well as in Europe," he says.


"In the U.S., algorithms come first; in China, applications comefirst."


According to him, engineers in tech giants like Google and Amazoncontinue to push the boundaries of science to come up with original algorithms.However, due to severe competitions in China, not many firms can afford toinvest a huge amount of time in such scientific research.


As for who will win the AI race, he thinks it is too early to say.However, despite each country's huge effort in development of the technology,the capability of AI in daily applications is still far from being fullyrealised. He reckons it will take at least another 5 to 10 years for AI tocompletely infiltrate into our daily lives.


"We tend to overestimate what we can achieve in the short run andunderestimate what we can achieve in the long run," he says. "In thelong run, I believe AI will deliver a lot of the fancy things that we couldn'teven imagine today. Just like when the internet first came about, no one wasable to foresee how much it would affect our lives as it is now," he adds.


The Irreplaceable Role of Humans

Like it ornot, AI is already part of our lives and will continue to further impact us infuture. The question, therefore, is not whether or not we should work with AIbut rather "how to work with machines so that we, humans, remain in charge",as Prof. Zhang points out.


In his opinion,AI is a tool which helps to free upour time spent in ordinary routine work so that we can focus our time andenergy in achieving bigger tasks.


"Forexample, if machines can answer my long list of emails every day, that wouldspare me a lot of time for more meaningful, creative tasks," he says.


In the sameway, one can imagine how much time the technology can save other professionsfor the benefit of society. "If doctors can spare their time from treatingordinary cases to concentrating on finding solutions for complicated diseasesthrough the help of AI, that would be a good thing for our community."

In thebusiness world, there are also things that can't be replaced by machines suchas leadership.


"Manyleaders nowadays are making data-driven decisions, but I don't think machinescan help them formulate strategies," says Prof. Zhang.



Yu Liu,Lihong Zhang and Xiaoquan (Michael) Zhang. 2019. "Statistical Inferencewith Ambiguity." (Working paper)


Thisarticle was first published in the China Business Knowledge (CBK) website byCUHK Business School:


About CUHK Business School

CUHKBusiness School comprises two schools -- Accountancy and Hotel and Tourism Management -- and fourdepartments -- Decision Sciences andManagerial Economics, Finance,Management and Marketing. Established in Hong Kong in 1963, it is the firstbusiness school to offer BBA, MBA and Executive MBA programmes in the region.Today, the School offers 8 undergraduate programmes and 20 graduate programmes including MBA, EMBA,Master, MSc, MPhil and Ph.D.


In the FinancialTimes Global MBA Ranking 2019, CUHK MBA is ranked 57th. In FT's 2018 EMBA ranking, CUHK EMBA is ranked 29th in the world. CUHK Business School has the largest numberof business alumni (35,000+)among universities/business schools in Hong Kong-- many of whom are key business leaders. The School currently has about 4,400undergraduate and postgraduate students and Professor Kalok Chan is the Dean ofCUHK Business School.


More information is available at or byconnecting with CUHK Business School onFacebook: LinkedIn: