AI·Future (AI·未来) and Artificial Intelligence (人工智能) are both books written by Professor Kai-Fu Lee, the latter co-authored with Yong Gang Wang, that talks about the current and future challenges and opportunities brought by Artificial Intelligence. I don't pursue providing a summary of both books, but rather contribute my afterthoughts on them hoping to inspire you to maybe skim or read thoroughly both of them in order to share ideas regarding how humanity shall cope with the benefits and dangers that AI will present to us in the near future. It's meaningful for trying to cope with the deep changes caused by AI that are going to permeate into our daily lives.
Cover of the book AI · Future by Kai-Fu Lee
Narrow AI vs. General AI: the future of AI progress
A common starting point to address AI might be establishing with what perspective to view AI. Given the influence of fictitious movies and novels, it might be worth pointing out some of the exaggerations regarding the current and near-future progress on AI. Despite since 2012 with the release of Prof. Geoffrey Hinton’s paper about Deep Learning[1] which has “revolutionized” the field of AI, such optimistic environment has also morphed into fear given that AI may one day surpass humans in every task we do, with such worry being augmented by how Alpha-Go defeated the world champion in the game of Go, with books like Superintelligence by philosopher Nick Bostrom or predictions made by Prof. Stephen Hawking regarding a dystopian future caused by AI. However, from the point of view of Kai-Fu Lee, such incidents should be addressed cautiously and without spreading unnecessary fear.
Yes, AI is advancing very rapidly, with achievements such as being better at image recognition than humans across different fields such as in that of security or medicine, predictions in the field of economics or simulating physical phenomena, among others. This holds true especially given the extraordinarily amount of training data that is being generated which will be able to further train neural networks at much more diverse tasks. Nonetheless, it’s worth pointing out that such AI shouldn’t be the source of fear given how limited is the task it can perform, i.e., Alpha Go can defeat the world champion in Go, but definitely can’t at the same time generate a short poem, a convolutional neural network can excel radiologists in identifying, say, a pulmonary disease like pneumonia, but it can’t simultaneously device a therapeutic strategy to deal with such disease like a doctor would. All in all, AI still cannot perform tasks across fields, or do multitasking, which means that it is still what’s commonly known as narrow AI. Given such premise, a fundamental idea that has put forward is to not fall down into ungrounded panic that may result in the halting of study of AI in order to prevent a perceived destruction of humanity (a sort of subconscious self-preservation bias).
Kai-Fu Lee did point out what narrow AI still lacked in order to turn into general AI:
o Cross-domain reasoning: as stated above, for narrow AI to become general AI, then it shouldn’t be task-specific, or task-oriented, given that it should be able to not only “solve” problems when indicated, but maybe also “study” in its spare time, “come up” with a new recipe that hasn’t been created, “predict” the weather or the next natural catastrophe, device a strategic plan to counteract such, among others. However, it may be a long road before such feat is achieved, given that another breakthrough in AI might be required.
o Abstract reasoning: deep learning can be thought of as fundamentally “performing isomorphic transformations to a set of input vectors or matrix in order to map them to a probability distribution to get an output”. However abstract reasoning might be something fundamentally different than this; how is it that our minds can create and device analogies, such as in Chinese, relating the number 8 to wealth, or generate creative works that help you memorize the periodic table through a song? This and other lingering questions.
o Know “what’s the result” and knowing “why”: though there are advance in explainable AI in explaining why do some neural networks output certain results, when humans think about the “why something happened” concerns the establishment of causal relationships, which although AI lacks to some extent, it’s a field that has seen progress in solving.
o common sense and self-consciousness: I do share the same idea with Kai-Fu Lee about AI lacking these features we humans have, and I do admit I can’t make further comments on them given that they are probably nativist features that has accompanied humans since birth. Whether they can be “given” or “gained” by AI is a mystery to me, which furthers the above viewpoint of narrow AI being unnecessary in stirring up panic.
o aesthetics
o emotion: Kai Fu Lee has a personal experience that has led him to believe that despite AI becoming smarter and smarter, it might never replace humans in the field of having emotions. A life anecdote of his is that in the past he once had stage IV lymphoma, whom after undergoing the necessary operation and surviving, started reflecting about his own life and started noticing that because he has been a workaholic through most of his life he has been neglecting his wife and children, which in turn catalysed in him profound changes from being an entrepreneur and computer scientist, to becoming a good husband, son and educator who could witness what we commonly know as “love".
Laying out the above cards on the table, I just want to help, like Kai-Fu Lee, relieve some of the unnecessary worries regarding AI, because when talking about such some lingering questions may arise such as ”can AI also undergo such a meaningful reflection about itself and go through a meaningful change to become a more kind AI?”. There may be the case that AI can replace the human brain as it expands its horizon of tasks able to be solved, but it cannot replace the human heart, being such a complex and dynamic thing.
Artificial Intelligence and the future of businesses: opportunities and challenges
Despite trying to appease excessive panic, this does not mean that I seek to spread absolute negligence towards the development of AI and its impact to our daily lives. It’s unnecessary panic that should be avoided, but rational cautiousness is more than well invited from the reader in order to aid the future humanity in solving the possible crashes that can happen between AI and humans, particularly in the business sector in terms of jobs. Without inquiring too deep into the undoubtly benefits that AI can bring in terms of rising economy thanks mainly to the rise of efficiency, perhaps what’s more important is to address the impact on jobs lost and the resulting inequality in welfare.
As a brief summary, please examine Figure 1 extracted from Kai-Fu Lee’s book AI · Future where jobs that may be secured and lost in the future are displayed on a basis of technicality of such job, and the degree of social interaction. Technicality here refers to the degree in which the job can be mechanized or structured, for example, a cardiologist is a highly technical job that’s difficult to be mechanized or structured given that a cardiologist must learn how to diagnose a disease by being a data scientist able to swiftly manipulate electronic health records, perform surgeries and may also be trained in relieving the emotional burden of the patient, so such cross-domain reasoning involving the use of emotions is currently very difficult for AI to simulate, while on the other extreme a job that requires highly repetitive steps (i.e. easily mechanized) and doesn’t involve much social interaction, like a truck driver, can be easily replaced by AI.
Figure 1 Coordinate System translated from Kai Fu Lee’s book AI · Future. It displays jobs that may be secured, must merge with AI, lost or have to undergo changes to adapt to the future. The x-axis represents the degree of technicality and difficulty in being mechanized, while the y-axis represents the degree of social interaction of such job.
From the graph above, an immediate advice would then be to at least lay down the following foundations in order to secure jobs in the future:
- The learning of programming skills in order to develop the capabilities required to analyse data or develop software that may be required by jobs that fully or has partially adapted AI.
- The learning of social skills in order to not only make the above programming skills useful for a community through being down-to-earth, but also because jobs that require human-to-human interaction can those jobs lost due to human-to-object interactions mechanized by AI.
But of course, awareness arising from oneself is not enough to curb the problems brought by AI in the business sector, rather, the interplay of key members and institutions of society; this is all in order to avoid the catastrophic welfare inequality that may arise otherwise. Kai-Fu Lee pointed out four entities that will shape the era of AI:
o Big Data: something to be pointed out is that although the rate of generation of data has augmented over the years, the rate of analysis of such data, that is, the generation of knowledge, hasn’t kept up. In the era of AI, hence, to truly make such data useful, then a thorough processing of such directed at solving issues in society, be it healthcare or in the business sector, is needed.
o Entrepreneurs: perhaps it’ll be in the creative minds of entrepreneurs on coming up with novel ideas of futuristic jobs, of course bearing in mind the pursuit of the over-all, long term welfare of a society as a whole instead of a short term accumulation of welfare that will exacerbate social instability.
o Scientists
o Government and Policy Makers: in the future, perhaps the core of policy makers shouldn’t be devising policy to ensure the equal distribution of resources, akin to partitioning a cake equally to a group of people, but rather create and ensure an environment that allows the opportunities for the production of more resources to be vibrant, akin to making such cake bigger for a larger group of people together.
Prof. Kai-Fu Lee also shared some ideas regarding where the focus for the generation of future jobs is. It is unlikely to witness the next breakthrough after deep learning in AI in the near future, hence the creation of an innovative service or product should be built upon on the basis of how to mostly exploit the technique of deep learning along with existing big data (of course, this is not to undermine the pursuit of the next breakthrough in AI); such product should be:
o Down-to-earth: that is, using the enormous amount of available data, build a model that can solve current and future issues of society instead of just simply breaking a record in certain game or defeating a very well-known player; instead, for such product to flourish, it should solve concerning issues that has been noticeable, for instance, during the COVID-19 pandemic, be it in the healthcare sector (how an AI can improve real-time diagnosis and health monitoring) or education sector (how AI can aid widespread good quality education that can be available online to ensure learner’s safety and health).
o Localized to a society. For instance, in a country such as China, which houses a civilization with a distinct customs and culture, how can a service or product be designed in such a way that’s well amalgamated to such civilization. An example of such is the difference between the search engines Baidu and Google. In Baidu, whenever you click on a link, a new tab is opened to display such webpage, while in Google the link is opened in the same tab. Such slight difference respects the customs of Chinese wanting to be able “to explore multiple webpages whilst still having the menu without being kicked out”, which is different from Google users who tend to focus on the same webpage hence don’t require to return to the menu. How can more products or services in the era of AI, with finely tuned subtleties adapted to a particular society, help solve the unemployment issue more efficiently than blindly creating a new product or service?
AI and human self-understanding: the future of philosophical reflection of human nature
Kai-Fu Lee in his books, as well as in his TED Talk “How AI can save our humanity”, has constantly mentioned how after having recovered from his cancer, his views about life and AI have changed dramatically, particularly because he saw in his transition towards paying more attention to his family, be it wife, daughters and aging mother, as well as the conversion from a machine-like, cold hearted workaholic to a passionate educator guiding the next generation on how to properly address the problems of AI, a human quality that may never be modelled, hence never be surpassed, by AI. Whether it is called “passion”, “love” or the “longing for life” or “wish for good health”, these are all perhaps features that may become access points towards a further understanding of ourselves. Such a reflection may inspire the future generation to become more vivid in questioning themselves given that through the development of AI, not only has our understanding of machine intelligence improved, but also that of ourselves.
That is, when AI learns to become more and more similar to humans, it’ll be our mission to find out in what else are we different. In the search for perfecting AI, we might learn to understand us better. Perhaps AI can help us answer the question that has accompanied humans for eternity: What is the purpose of life?
Our contribution to saving the future: Innovative ideas for futuristic jobs
Whether you intend to become a data scientist, policy maker, entrepreneur, artist, among other professions which may seemingly appear unrelated to each other, we can all contribute with our ideas on how to advert the challenges brought by AI while making the most out of this new era. In the end, just as Prof. Kai-Fu Lee pointed out, it’s perhaps inaccurate to say that what has been said might occur in the near or far future... the future is “NOW”.
After sharing the above, the following is an expanding list of possible jobs that may serve in curving the unemployment caused by AI; your ideas are expected to further enlarge the list:
- Data miner (profession): for those who are familiar with deep learning, it is very well known that to train a network, a great amount of data is required. Although it has been repeatedly mentioned that there is already a lot of unprocessed data, some businesses may still require data that just has too many specific qualities, hence requires the hiring of data miners to collect data of such characteristics. For instance, to build driverless cars, a lot of data concerning videos of road infrastructure, stop signs, lights, pedestrians, among others, must be collected.
- Data labelers (profession): following from the above, data labelers should become a profession in order to develop an AI software.
- Waiter serving food on driverless limousine/restaurant (profession): to compensate for the drivers that may lose their jobs due to driverless cars, they can be retrained to follow another profession like a waiter and turn their driverless cars into a dynamic, small scale restaurant.
- Small diagnostic device that takes in a drop of blood with certain miRNA configuration in order to predict the risk of developing cancer (product): this has been an idea shared in a previous post, called “DeepMiRNA: miRNA as therapeutic targets”
- An application that takes as input your notes taken from a lecture and outputs an interactive game to help you memorize them (product): the idea is to make notes taken from a lecture as addictive as someone is to a famous game.
- What ideas do you have?
In the end, if you are interested in artificial intelligence and its impact on humanity, then you are more than welcomed in reading Prof. Kai-Fu Lee’s books as well as many others.
Bibliography
(李开复), K.-F. L. (2018.9). AI·Future (AI·未来). Hangzhou: Zhejiang People's Publishing House （浙江人民出版社）.
Kai-Fu, L. (., & Yong Gang, W. (. (2017). Artificial Intelligence （人工智能). Beijing: Cultural Development Press (文化发展出版社）.
[1] Please read: https://dl.acm.org/doi/10.1145/3065386
Comments