[VN]Holding Promising Opportunities And Confronting Risks Of Artificial Intelligence

1. The time of AI may finally have come, but more work still needed to be done

1. The time of AI may finally have come, but more work still needed to be done

The topic of Artificial Intelligence, its development and countries achievement have become the top interest of countries all over the world. It is said that technology such as AI or machine learning is mainly the race between huge names like China and US. However, other coutries, developed or developing like Vietnam, have also participated in the artificial intelligence game. AI has often failed to live up to the hype that surrounded it, despite periods of significant scientific advances over the six decades since. Decades have been spent trying to accurately describe human intelligence, and the progress that has been made has not led to the prior excitement. However, technological progress has been accelerating since the late 1990s, especially over the past decade. Machine-learning algorithms have advanced, especially through the development of neural network-based deep learning and reinforcement-learning techniques.
The recent progress has led to several other factors. Exponentially, a larger and more complex models are being trained with more computer capacity, and the use of graphics and tensor processor units is being made possible through a number of innovations at the silicon level. This power is aggregated in hyperscale clusters, becoming increasingly available through the cloud to consumers.

The large quantities of data generated to train AI algorithms are another key factor. Some of the advancements in AI were the result of developments at system level. Autonomous vehicles are a good demonstration of this: they make use of sensor technologies, LIDAR, machine vision, mapping and satellite technology, navigation algorithms, and robotics all incorporated into integrated systems.
Given the success, there are still many difficult issues that will require further scientific breakthroughs. Most of the progress so far has been in what is called narrow AI, where machine-learning methods were developed to specifically solve one or some problems, citing natural language processing as one example. The more difficult issues are in what is commonly referred to as artificial general intelligence, where the goal is to build AI that can deal with general issues in much the same way as humans do. Most researchers believe that this is decades away from becoming possible in real life.

Deep learning and machine learning are driving AI

Much of the recent excitement about AI was due to progress in the field of deep learning, a set of techniques that are based on artificial neural networks to implement machine learning. The AI systems loosely form the activity of neurons in the brain. Neural networks have a large number of (deep) layers of simulated connected neurons. Whereas earlier neural networks had only three to five layers of neuron and dozens of neurons, deep learning networks may have 10 or more layers, with a million simulated neurons numbering.
There are several types of machine learning: supervised learning, unsupervised learning, and reinforcement learning. Each type specifically suits different cases. The implementations of supervised learning are the most common functional examples of AI. Training data are used to assist the system to learn about the relation between the different inputs to a particular output–for example, to recognize images of objects or to transcribe human speech–often when marked values are accessible and the desired output variables are identified.
Unsupervised learning is a set of techniques used in the form of clusters or patterns such as images from buildings with similar architectural styles in a set of available data, without the need for labeled training data.
Systems are trained in reinforcement learning, often by means of a scoring system, by virtuous rewards or punishments. These techniques evolve through ongoing work.
In Vietnam, innovation centers and artificial intelligence centers throughout the countries have also applied machine learning and deep learning techniques to develop latest solutions apply artificial intelligence technology.

Limitations remain unsolved, although new techniques hold promise

Although new techniques are emerging to address them, AI still faces many practical challenges. Machine learning can require a great deal of human effort to label the necessary training data for supervised learning. In-stream monitoring, where data can be labeled during natural use, may help alleviate this problem.
It is also often challenging to obtain sufficiently large and comprehensive data sets to be used for training, e.g. to develop or obtain enough clinical test data to accurately predict healthcare results.
The complexity of the black box of deep learning techniques creates the explainability challenge, or shows which factors led to a decision or prediction, and how. This is especially important in applications where trust issues and predictions have societal implications, such as applications for criminal justice or financial lending. Some emerging approaches, including local interpretable model-agnostic explanations (LIME), are aimed at enhancing transparency of this model.
Another challenge is building generalized learning techniques that can be applied to various use cases, as AI techniques continue to struggle to carry their experiences from one set of circumstances to another. Transfer learning, where an AI model is trained to perform a certain task and then quickly applies that learning to a similar but distinct activity is a promising response to this challenge.

2. Businesses stand to benefit from AI

While AI is increasingly widespread in consumer applications, companies are starting to adopt it throughout their operations, sometimes with striking results.

AI’s potential cuts across industries and functions

AI can be used to improve business performance in areas including predictive maintenance, where the ability of deep learning to analyze large amounts of audio and image high-dimensional data can effectively detect anomalies in factory assembly lines or aircraft engines. In logistics, AI can optimize delivery traffic routing, improve fuel efficiency and reduce delivery times. In the field of customer service management, AI has become a valuable tool in call centers thanks to improved speech recognition.
Combining customer demographic and past transaction data with social media monitoring can help generate individualized 'next product to purchase' recommendations that many retailers are now using frequently.
Such practical cases and applications for AI use can be found across all economic sectors and across multiple business functions, from marketing to supply chain operations. Deep learning techniques mainly add value in many of these use cases by improving conventional analytical techniques.
Our analysis of over 400 use cases across 19 industries and nine business functions found that in 69 percent of potential use cases, AI improved on traditional analytics techniques. There is a greenfield AI approach that was applicable in only 16 percent of AI use cases where other analytical methods would not be successful. Our research projected that deep learning techniques focused on artificial neural networks could yield as much as 40% of the total potential value that all analytical techniques could yield by 2030. We also estimate that several of the deep learning techniques could provide an annual cost of up to $6 trillion.

So far, adoption is uneven across companies and sectors

Although many companies have started adopting AI, there has been inconsistent speed and scale of adoption. Nearly half of the respondents in a 2018 McKinsey AI adoption survey state that their businesses have integrated at least one AI technology into their business processes, and another 30 percent are only at piloting stage. Nonetheless, only 21 percent report that their companies have integrated AI in several parts of the business, and only 3 percent of large firms have implemented AI in their entire workflow.
Other surveys show that early AI adopters tend to think more expansively about these technologies, grow their markets or increase market share, while less experienced companies focus more closely on cost reduction. Highly digitized businesses tend to invest more in AI and derive higher value from its use.
At sector level, there is a widening gap between digitized early adopters and others. Sectors that are highly ranked in the Industry Digitization Index of MGI, such as high-tech and telecommunications, and financial services are leading AI adopters with the most aggressive AI investment plans. As these businesses accelerate AI adoption and accumulate more information and AI capabilities, it may be more difficult for laggards to catch up.
Currently, there are positive sign in artificial intelligence adoption in household devices, Asian countries, which are China, Vietnam and Indonesia respectively as the top 3 countries according to ownership of AI devices.

Several challenges to adoption persist

Most companies and industries have been lagging behind in embracing AI. Developing an AI strategy with clearly defined benefits, finding talent with appropriate skill sets, overcoming functional silos that restrict end-to-end deployment, and leaders ' lack of ownership and commitment to AI are among the most frequently cited barriers to adoption.
On the strategic side, businesses will need to develop a corporate view of compelling AI opportunities, potentially transforming parts of their current business processes. Organizations will need efficient systems for data capture and governance as well as modern digital technologies and will be able to build and access the technology they need.
It will be even more difficult to solve the last mile problem of ensuring that AI's superior insights are inculcated in an enterprise's people's behavior and processes.
On the talent front, much of deep neural network construction and optimization remains an art that requires real expertise. Demand for these skills far outstrips supply; it is estimated that fewer than 10,000 people have the skills needed to tackle serious AI problems, and it is fierce competition for them. Companies considering the option of building their own AI solutions need to consider whether they are capable of attracting and retaining those specialized skills.

3. Economies also stand to benefit from AI, through increased productivity and innovation

The deployment of AI technology and automation can do a lot to lift the global economy and increase global prosperity. Productivity growth becomes critical for long-term economic growth at a time of aging and falling birth rates. Even in the near term, productivity growth in developed economies has been sluggish, falling from 2.4 percent a decade earlier in the US and major European economies to an average of 0.5 percent in 2010–14. AI has the potential to contribute to productivity growth, much like previous general-purpose technologies.

AI could contribute to economic impact through a wide range of channels

AI is expected to have the greatest economic impact on productivity growth by labor market impacts, including substitution, rise, and labor productivity contributions.
Our research suggests that labor replacement could represent less than half of the overall benefit. AI would increase human capacity, free workers to participate in more profitable and higher-value activities, and increase demand for AI technology-related jobs.
AI can also boost innovation, enabling businesses to improve their top line by more effectively reaching underserved markets with existing products and creating completely new products and services over the longer term. AI will also create positive externalities, facilitate more efficient cross-border trade and allow valuable cross-border data flows to be used more widely. These economic activity and income increases can be reinvested in the economy, thus leading to further growth.
AI deployment will also bring some negative externalities that may reduce the positive economic impacts, although not eliminate them. In terms of the economy, these include greater competition moving the market share of non-adoptersto front-runners, the cost of handling labor market shifts, and potential product losses in periods of unemployment for people, as well as the cost of integrating and implementing IA deployments.
In all, these different channels result in significant positive economic growth, assuming that companies and governments handle the transformation proactively. In one simulation conducted using McKinsey survey data, the adoption by AI may raise world GDP by up to 13 trillion dollars by 2030, an additional 1,2 percent per annum of GDP growth. However, this effect will only grow over time as most AI's implementation costs can be above the revenue potential.

The AI readiness of countries varies considerably

The leading sources of potential economic growth driven by IA, such as investment and research activity, digital absorption, connectivity and the structure and flexibility of the labor market, depending on country. In addition to the capacity to innovate and develop the requisite human capital skills, our research suggests that AI productivity is likely to be a key factor in future GDP growth.
Countries that lead the AI race have unique strengths that distinguish them. Scale effects allow for larger investment and network effects allowing these economies to attract the talent required to optimize AI. For now, the majority of research activities and investments related to IA are carried out in China and the United States.
A second group of countries including Germany, Japan, Canada and the UK has a strong history of driving innovation and can speed up the commercialisation of AI solutions. Smaller economies like Belgium, Singapore, South Korea and Sweden, also have a strong impact on their ability to promote productive environments with a thriving development in new business models.
The third group countries are in a relatively weaker starting position, including Brazil, India, Italy and Malaysia, but they have comparative power over certain areas where they can build on this. India, for instance, generates around 1.7 million STEM graduates per year— more than the sum of all G-7 graduates. Many nations, whose digital infrastructure, development, creativity and technological abilities are largely underdeveloped, are in danger of falling behind their peers.

4. AI and automation and its profound impact on jobs

Even if AI and automation provide business and the economy with benefits, major work disruptions can be expected.

Approximately half of the current work activities can be automated technically

Our study of the effect of automation and AI on work shows that it is technologically easier to automate some types of tasks than others. These include physical activities in highly predictable and structured environments as well as collecting and processing data, which together represent about half of the work that people undertake in most economies across all sectors
Other groups that are least vulnerable include leadership, competence and stakeholder engagement. Highly automated operation frequency ranges across professions, industries, and countries to a lesser extent. Our research finds that about 30 percent of the activities in 60 percent of all professions could be automated— but almost all activities can be automated in just about 5 percent of professions. In other words, there will be more partly automated occupations than fully automated.

Three simultaneous effects on work: Jobs lost, jobs gained, jobs changed

In addition to technical feasibility, the speed and degree to which automation will be implemented and the effect on actual jobs will depend on several factors. These include the cost of deployment and adoption, and the dynamics of the labor market, including the quantity of labor supply, quality, and related salaries. The labor factor in emerging and developed economies contributes to large disparities. Business benefits beyond job substitution— often involving the use of AI for beyond-human capabilities— which contribute to business cases for adoption are another factor.
The timing will also be determined by social norms, social acceptance and various regulatory factors. How all these factors play out across sectors and countries will vary, and the dynamics of the labor market will largely be driven by countries. For example, in advanced economies with relatively high wage levels, such as France, Japan, and the United States, employment affected by automation could be more than doubtful.
Given the interplay of all these variables, it is difficult to make assumptions, but it is possible to develop different scenarios. First of all, on job losses: our mid-point adoption scenario for 2016 to 2030 suggests that around 15% of the global workforce (400 million workers) could be displaced by automation.
Second, jobs gained: we built labor demand scenarios up to 2030 based on expected productivity-based economic growth and considering several drivers of labor demand. These included rising incomes, especially in emerging economies, as well as increased expenditure on health care for aging populations, investment in infrastructure and housing, energy transition spending, and technology creation and implementation spending. The high demand for engineers working in artificial intelligence creates the avantage for countries with abundant labor force like Vietnam and other Asian countries.
Through these and other catalysts the number of jobs added could range from 555 million to 890 million, or 21 to 33 percent of the global work force. It indicates that growing demand for employment would more than offset the number of jobs lost to automation, barring extreme scenarios. Nevertheless, it is important to note that there will already be a daunting need in many emerging economies with young populations to provide employment for workers entering the workforce and that, in developed economies, the estimated correlation between job losses and those generated in our scenarios is also a consequence of ageing and thus fewer people entering the workforce.
The occupations that will shift when machines supplement human labor in the workplace are no less important. As a result of the partial automation mentioned above, employment will shift, and job changes will impact many more jobs than job losses. To keep up with rapidly evolving and increasingly capable technology, skills for staff complemented by computers, as well as job design, will need to be adapted.

Four workforce transitions will be significant

Even if, as most of our scenarios suggest, there will be enough work for people in 2030, the transitions that will accompany automation and AI adoption will be significant.

First, it is likely that millions of workers will need to change jobs. Some of these changes will occur within industries and sectors, but many will occur across sectors and even across geographies. Although jobs requiring physical activity will decline in highly structured environments and data processing, others that are hard to automate will rise. These could include administrators, teachers, nursing assistants, and tech and other staff, as well as gardeners and plumbers employed in volatile physical environments. Such shifts may not be smooth and may result in temporary unemployment spikes.

Second, in the workforce of the future, workers will need different skills to succeed. Demand for social and emotional skills like communication and empathy will rise nearly as quickly as demand for many advanced technological skills. Basic digital skills in all occupations have increased. Automation will also stimulate growth in the need for higher cognitive skills, especially critical thinking, imagination, and complex processing of knowledge. Demand for physical and manual skills will decline, but in many countries these will continue to be the single largest class of job skills by 2030.

Third, as more people work alongside computers, workplaces and workflows may change. For example, as self-checkout machines are introduced in stores, cashiers will shift from scanning merchandise itself to helping answer questions or solve the machines' problems.

Lastly, automation is likely to put pressure on advanced economies ' average wages. Many of the current middle-wage jobs in advanced economies are dominated by highly automated activities, which are likely to decline in areas such as manufacturing and accounting. High-wage jobs, particularly for high-skilled medical and tech or other professionals, will grow significantly. Nonetheless, there are typically lower pay rates that are supposed to be generated for a large portion of workers, such as teachers and nursing aides.
Most economies, especially in the OECD, are starting in a hole in tackling these changes, given the existing skill shortages and challenged educational systems, as well as the trends towards decreased spending on on-the-job training and support for employee change. There is already income inequality and wage polarization in many economies.

5. AI’s benefits and problems to social

AI will have a positive impact on society in addition to the economic benefits and challenges, as it helps address societal challenges ranging from health and nutrition to equality and inclusion. It, however, also creates pitfalls that need to be addressed, including unintended consequences and misuse.

AI could help tackle some of society’s most pressing challenges

Through automating repetitive or dangerous tasks and those prone to human error, AI can make it possible for people to be more efficient, work and live better. One research analyzing the United States suggests that by reducing injuries, replacing human drivers with more reliable autonomous vehicles could save thousands of lives a year.
The need for people to work in dangerous conditions such as offshore oil rigs and coal mines can also be minimized. For example, DARPA is developing small robots that might be deployed in disaster areas to minimize the need of people doing dangerous work. A number of AI functions are particularly important. Classification of images on skin photos taken via a mobile phone app could evaluate whether moles are cancerous, facilitating early-stage diagnosis for people with limited access to dermatologists. Through detecting objects such as vehicles and lamp posts, object detection can help visually impaired people navigate and communicate with their environment.
Our research and that of others has demonstrated numerous cases of use across many areas in which AI can be used for social benefit. Several barriers must be overstepped if these AI-enabled interventions are to be applied effectively. These include the usual data, computing, and talent availability challenges faced by any organization attempting to apply AI, as well as more fundamental access, infrastructure, and financial resources challenges that are especially acute in remote or economically challenged places and communities.

AI will need to address societal concerns including unintended consequences, misuse, algorithmic bias, and questions about data privacy

Economically, difficult questions about widening economic gaps across individuals, companies, industries, and even countries that may arise as an unintended consequence of deployment will need to be answered. The use and misuse of AI are other areas of concern. They range from being used in surveillance and military applications to being used in social media and politics, and where the effect has social consequences like in criminal justice systems. We also need to consider the potential of malicious-intentioned users, including in cyber security areas.
These questions are directly related to how algorithms can be implemented and perpetuated and institutionalized by the information used to train them. For example, facial recognition models trained on a facial population corresponding to artificial intelligence developers ' demographics may not reflect the wider population.
Data privacy and personal information use are also critical issues that need to be addressed if AI is to realize its potential. In this area, Europe has led the way with the General Data Protection Regulation, which introduced more stringent data collection consent requirements, gives users the right to be forgotten and the right to object, and strengthens supervision of organizations collecting, controlling and processing data, with substantial fines for non-compliance. It is also a concern for cybersecurity and deep fakes which could manipulate election results or perpetrate large-scale fraud.

6. Three priorities for achieving good outcomes

The potential benefits of AI for business and the economy, and how technology solves some of the societal problems, would encourage business leaders and policy makers to accept and implement AI. At the same time, it is impossible to ignore the future problems of adoption, including effects on the workforce and other social concerns. Key challenges include:

- Deployment challenge

We are interested in embracing AI at a time when many economies need to boost productivity, given its likely contribution to business value, economic growth, and social good. Businesses and countries have a strong incentive to keep up with world leaders like the U.S. and China. Increased and wide-ranging deployment will require accelerating progress on the technical challenges and ensuring that all potential users have access to and benefit from AI. Measures that may be appropriate include:
• Investing in and continuing to advance AI research and innovation in a way that ensures everyone can share the benefits.
• The expansion of available data sets, particularly in areas where their use would lead to wider economic and social benefits.
• Investing in human capital and infrastructure relevant to AI to expand the talent base being able to create and execute AI solutions to keep pace with global AI leaders.
• Encouraging increased AI awareness to drive informed decision-making between business leaders and policy makers.
• Supporting existing efforts in digitization that form the basis for the eventual deployment of AI for both organizations and countries.

- The future of work challenge

A starting point for addressing automation's possible disruptive impacts will be to ensure sustainable economic and productivity development, which is a prerequisite for job growth and increased prosperity. Governments will also need to promote business dynamism, as innovation and faster growth will not only boost productivity, but will also accelerate job creation. Addressing skills, employment, and salaries problems would take more concentrated action. These include the following:
• Evolving education systems and training for a changing workforce by concentrating on STEM skills, innovation, critical thinking, and lifelong learning.
• Increasing private and public investment in human capital, maybe funded by opportunities and credits similar to those available for investment in R&D.
• Improving the dynamism of the labor market by encouraging better credentials and matching, as well as allowing for different forms of work, including the gig economy.
• Rethinking income by consideration and experimenting with services that would not only provide income for jobs, but also sense and integrity.
• Rethinking transition support and safety nets for impacted staff through the use of best practices from around the world and consideration of new approaches.

- The responsible AI challenge

AI will not live up to its promise if the public loses confidence in it as a result of violations of privacy, discrimination, or malicious use, or if it is blamed for exacerbating injustice by much of the world. It will be important to build confidence in their ability to do good when addressing misuse. Could include the following approaches:
• Strengthening the rights of customers, data, privacy and security.
• Establishing a common framework and set of principles for the effective and secure use of AI.
• Sharing best practices and continuous innovation to address issues such as security, bias, and clarification.
• To strike the right balance between the company and the dynamic national race to lead in AI to ensure that AI's benefits are widely available and shared.

Tin liên quan

Standar