National AI Policy Making — Part 3: Education and Training for AI
This post — third in a series about national AI policies — is about AI education and training. AI literacy is a topic that should be top of mind for policy makers especially at a time when inequality is on the rise and digital divide is getting worse.
Awareness of the Potential and Limits of AI
First, realistic awareness of AI should be pervasive at all levels of education, business, government and industrial sectors. One cannot expect meaningful progress on AI in any organization unless its people have a grasp of basics of AI, ML and data. Policy makers need to make AI accessible to all and not just for those with material resources. Another important goal of the policy should be to bridge the digital divide.
Finland has taken an interesting approach towards making AI accessible — they have created a free online AI course that anyone can take (here are three reasons for why you should learn AI.)
Call for Action Everyone in business needs data and AI literacy — from individual contributors to senior leaders who may or may not consider themself technical. Concepts need to be simplified and made relevant to those who are new to these topics. We should take time to explain key AI concepts, like classification and confidence levels, ethics and fairness in machine learning, for non-technical audiences. We must go beyond the hype and ask tough questions. How do you mitigate the dangers of applying technology that we don’t fully understand and may not be able to control? Are algorithms our new bosses without us even realizing it?
Education at All Levels
Starting from schools and going up to college, we need to modernize curriculum and provide opportunities for STEM learning balanced with non-STEM education that encourages critical thinking.
As noted by Kai-Fu Lee in his book AI Super Powers, it so happens that the most powerful leaders ask the same questions about AI that kindergartners ask!
Most schools and universities of today are not reacting fast enough to these changes because this is not a straightforward change. A few schools have started new undergraduate programs in AI.
“We saw a huge opportunity, but not a simple one, because it requires a rewiring of the academic structure” — Martin Schmidt, Provost, MIT.
WSJ reports that the Massachusetts Institute of Technology and Cornell University, within the past three months, have announced the development of new, multidisciplinary courses for AI in collaboration with the private sector.
MIT is pursuing a goal of $1 billion to create a new computing college that will include the study of AI and data science.
Cornell University is partnering with an AI firm r4 Technologies to train students across multiple disciplines in AI, data science and advanced math. “It is time to rethink the core liberal arts curriculum that universities offer to every student,” said Greg Morrisett, Cornell’s dean of Computing and Information Sciences.
As we saw in part 2 of this series governments are planning to ramp up efforts for R&D and ‘Centers of Excellence’. Many private institutions and online startups are trying to address this AI learning gap (see list at the end of this post.)
Talent Shortage and Need for Interdisciplinary Education
On one hand there is a shortage of technical AI talent — simply because there are just not enough qualified artificial intelligence, machine learning and data science resources. Today’s AI businesses and solutions require knowledge and expertise from many technical domains: mathematics, statistics, probability, computer science, brain and cognitive science, data science, many disciplines of engineering, IT systems and digital infrastructure. But hard sciences are not all that is needed.
“AI requires a deep partnership between the technologists and the humanists.” — Schmidt, MIT Provost.
In addition to the computer scientists, robotics experts and mathematicians we need social scientists, humanists, psychologists, economists, ethics, legal and domain experts who bring the contextual understanding and create a human centric solution that balances short-term gains with sustainable goals. There’s a need to develop thought leadership on the economic, ethical, policy and legal implications of advances in artificial intelligence.
Impact of Automation on Jobs
There are two major types of automation: Industrial and Software Process. Both are poised to create major disruptions to business and economies and alter our work.
Which jobs (or tasks) will be replaced by robots and automation? To start out any jobs that are repetitive, can be coded or augmented through machines are good candidate for automation. For industries his shift is already happening in China where manufacturing is being rapidly automated and workers are shifting to other jobs that needs human judgement and creativity.
The image below is from a video on the rise of robots in manufacturing.
If there’s anything we have learned from the past it is: Anything that MAY be automated WILL be automated.
The second type of automation is from automation of processes and tasks that are software based. It is sometimes called ‘Robotic Process Automation’ but could come in many forms. For instance, Chatbots (conversational bots)—software agents built on natural language understanding that converse with people and manage certain tasks. Their benefits are enormous. So is the impact on customer service and call center jobs.
This shift has started creating a dangerous imbalance that should be addressed and managed. People will need to re-skilled and re-trained. You may have seen different estimates of number of jobs lost and gained because of automation and AI. But there is no consensus on these numbers (see consolidated chart by Technology Review) so my recommendation is to think about the order of magnitude of problem and not worry about the projected numbers.
75 million jobs may be displaced by a shift in the division of labor among humans and machines. 133 million new roles could emerge if proper preparations are made. — Future of Jobs Report 2018, World Economic Forum .
Governments and Companies need to have a solid plan for training and re-skilling or as some call it upskilling!
Policy makers need to think about this disruption to jobs — especially for those people who may already be at a disadvantage because of their age, location or gender — and how to facilitate the transition to new careers.
Increasing desperation in the developing world will contrast with a massive accumulation of wealth among the AI superpowers. AI runs on data and that dependence leads to a self-perpetuating cycle of consolidation in industries — Bloomberg Article
This HBR Article suggests asking 5 questions about automation and jobs:
- Will workers whose jobs are automated be able to transition to new jobs? How easy or hard that would be?
- Who will bear the burden of (uneven) automation?
- How will automation affect the supply of labor?
- How will automation affect wages, and how will wages affect automation?
- How will automation change job searching? Artificial intelligence has the potential to predict better matches between job seekers and open positions (with caveats!)
Managing Job Displacements: Training and Re-skilling (define)
Previous section described the issue — uneven disruption that will hit many countries, regions and cities. The only way out is to prepare now.
By 2021 more than 120 million workers across the world’s largest economies will need to be trained and re-skilled as a result of artificial intelligence — IBM.
Policy makers need to proactively create support systems for impacted people.
“We have to be proactive in supporting people who might be impacted through fast-track job training and support systems.” — Tess Posner, CEO AI4All.Org
The training for professionals ranges from an introductory course to hands-on training for those who have certain prerequisites and want to become hands-on practitioners. Below are guidelines of three courses for professional audience. These should be customized based on specific needs of an organization:
- Introductory course for Executives — 8 hours of learning: 80% business including case studies, 20% technical concepts. Goals: understanding of what’s possible with AI, its foundations and limits, asking the right questions, how to get started with AI and craft a strategy and roadmap
- Introductory course for Managers: 16 hours of learning, 70% business and management, 30% technical with interactive exercises
- Hands-on course for Technical resources: 80+ hours, 30% business, 70% technical with using various tools and coding.
In addition to the above, there are many online courses with certifications as well as certificate and graduate level offerings by schools and various ‘data science’ training shops. The challenge is always to assess a given individual’s background and match a program with their interests, skills, time, budget and career aspirations. Perhaps we need AI to solve this!
Now let’s talk about the training for technical resources. Their success usually depends on how they can navigate the business world and show why their work and recommendations matter.
Data scientists and AI/ML experts are usually trained to do very complex and complicated things with data and algorithms, but their training is not necessarily in people skills, communication, or leadership. In other words you can have a very intelligent AI expert with low emotional quotient or poor explanation and communication skills.
Organizations working on AI Education
Dealing with such a broad and rapidly developing area is not easy and that is why partnerships and collaboration is important. Here’s a small sample of various resources that are active in AI education.
Non-profits:
- AI 4 All — a nonprofit working to increase diversity and inclusion in AI with focus on underrepresented talent
- Allen Institute — AI for the Common Good
- Curiosity Machine AI challenge
Online Education — MOOCs (technical):
- Coursera
- Datacamp
- Fast.AI
- Udacity
- Udemy
AI Companies:
- Google Education, Machine Learning https://ai.google/education/
- Google People + AI Research https://ai.google/research/teams/brain/pair
- Landing.AI https://landing.ai/ai-transformation-playbook/
- Microsoft: AI School https://aischool.microsoft.com/en-us/home
