Prof Ikhlaq Sidhu on the DataXHKBU workshop on 26 Jan (by Xinzhi Zhang)
Reflections on the talk by Prof Ikhlaq Sidhu on Artificial Intelligence
“Will AI help the film directors to make better movies? – Yes! But will AI replace the film directors and become directors? – No!” – Prof Ikhlaq Sidhu.
Thus responded by Prof Sidhu when I asked, “do you think AI will replace film directors?” Fortunately, our dialogue was far wider than the above. Prof Ikhlaq Sidhu, Professor & Chief Scientist, UC Berkeley, and the Founding Director, Center for Entrepreneurship & Technology, offered a University-wide “Distinguished Lecture” on “How to Achieve Data Analytics and Artificial Intelligence in X” and a two-full-day masterclass on Big Data Analytics in Hong Kong Baptist University.
In the 1.5-hour of talk, Prof Sidhu reviewed the traditional applications of AI (for example, classification and scoring/prediction). He then reviewed the future directions of AI. Thirdly, he pinpointed several limitations of AI vis-à-vis human (which I will offer some of my humble reflections on this blog), and how to understand the future development of this field (another key point I’d like to address further here). Lastly, some strategies options were offered to HKBU. During the four-day visit of Prof Sidhu to HKBU (thanks to the Knowledge Transfer Office, especially Alfred), we got a chance to have more in-depth discussion on this theme. My three quick and immature reflections are indicated below.
Reflection 1: AI and human
When the whole world is charmed (and feeling threatened perhaps) about the AlphaGo (and still remembering the perplexed look of Ke Jie), Prof Sidhu raised a question: “does this mean AI can do everything better than humans?” When the expected answer from the mass (as humans) might be “no” (but I have a large circle of tech-geek friends who would say yes, loudly and clearly), the mechanisms worth further explication. Prof Sidhu holds a very prudent and cautious view. AI can handle tasks that having certain outcomes (winning the GO); in a limited, in discrete conditions; where the human world may not always have the luxury of having certain outcomes (I wish I could know the path to become a billionaire), in unlimited and continuous conditions. Hence, one tentative statement here is, there will be more and more “specific AI” applications, focusing on one domain, handling one very specific type of tasks (AI-assisted financial planning, AlphaGo, auto-drivers, among others); but there are not “comprehensive AI” (do we have Blade Runner 2049 or Ghost in the Shell). That repeats the dialogue in the opening when I asked: “since AI can self-learn, and AI has been quite successful in many fields, will it success as well in creative industries, such as painting, music, poetics, and film making?”
Reflection 2: Data and theory
“The end of theory.” We all still remember the Chris Anderson’s 2008 essay on Wired. One primary argument is that, with the huge amount of data, we don’t need theory any more, because the patterns, regulations, and insights will emerge from the data. Neither Prof Sidhu and other audiences directly mention this piece, but Sidhu raised the cautious in seeking ground truth (if my interpretation is correct): if the truth of a certain parameter is 12; but all the top-ranked Google pages say it should be 11, then people will believe the parameter to be 11.
So how to deal with the relationship between data and “theory?” Prof Sidhu raised three critical reflections: a. data is more valuable than algorithm; b. Algorithm is more important than the system; and c. algorithms, data, and computing data is growing faster than computing.
Personally I have a strong resonance from point (a). Tycho, the 16th century Danish astronomer, who reached the human’s limitation to document decades-long astronomical and planetary observations. With the huge amount of data from Tycho, could later scientists, namely Kepler and Newton, accomplish scientific leap-forwards and raise new insights and propose new theories. In the big data era, we badly need more empirical data and observations that enabled by the accurate and comprehensive documentations of human behaviors, just like what Tycho did in several centuries ago. “Exactly! We still need theories and scientific methods: from observation to building up hypothesis and to make further inferences.” Thus responded Sidhu when I mentioned Tycho. (Credit should also be given to my high school geography teacher, when he told us the story of scientific progress in the human history).
Reflection 3: AI, job losses and lifestyle
The final reflection started from an audience’s question, which was also a commonly-discussed one: how to respond to the job losses because of AI? The first half of Prof Sidhu’s response to the question echoed some economists’ observation: while AI may lead to some job losses, but new job positions will emerge at the same time. Secondly, Prof Sidhu used the example of sewing industry in the history. When the sewing machine was invented, about 50 workers could be replaced by one machine (50:1). Further, the cost of the clothes was also dropped; and sequentially, people’s lifestyle was also changed (buying two clothes vs having a full closet of clothes – far more than one’s needs). Hence, my humble thought is that, more studies are needed to examine the long-term impacts of AI on social change. Topics may include lifestyle, fashion, aesthetics, culture values, social values, and ideologies.
- The Sutardja Center at Berkeley: http://scet.berkeley.edu
- Data-X in Berkeley: https://data-x.blog
- New course on Data-X: http://scet.berkeley.edu/data-x-course/
- 15 most popular Github projects on Machine learning (a translated article, written in simplified Chinese): https://www.jiqizhixin.com/articles/2017-12-21-10
Author/ Xinzhi Zhang