While Bulgarians can do a lot to leap ahead in AI, they can’t leap ahead all together. No one can. At best, a leap-ahead committee – say, a non-governmental organization (NGO) dedicated to promoting AI – can fire up enthusiasm and share best practices. At worst, a committee wastes endless time in planning, meetings and misdirection. A camel is a horse designed by a committee.
To rev things up, let me suggest a small initiative that doesn’t need more than a handful or backers. Form a boot camp to train data scientists. It will offer six courses only:
- ▲ Python or R programming. These are the languages most favored by machine learning specialists (Python) and statisticians (R). They’re fairly easy to translate between – more like Bulgarian versus Russian than English versus Chinese — so it’s not necessary to master both.
- ▲ Mathematical optimization. How to formulate goals as objective functions and maximize them under various kinds of constraints.
- ▲ Statistical learning. How to modify optimization given random noise and uncertainty, and to mix reliance on past experience with experiments to learn something new.
- ▲ Data munging. How to gather, clean and process huge masses of data.
- ▲ Cloud services. How to take advantage of distributed cloud committee and the ever-expanding array of services offered over the web.
- ▲ English explanation. How to communicate in nontechnical terms with most other colleagues and clients.
Each course will have at least one coach. I call them coaches rather than teachers to emphasize that they’re not there to lecture. The coach explains the aims of the course, organizes training, assists with problems and vets progress. The teaching itself will be offered online, tapping courses from Udemy, Coursera and others. Each course should offer at least two online course options, so that students can pick the style they like best and so that the boot camp can continually experiment with new approaches. The coach encourages students to answer each other’s questions, as there’s no better way to learn than to try to teach someone else, but helps directly when needed. Coaches administer final exams; for that students have to sit in classrooms under supervision, to prevent online cheating. Exams should also include oral questioning by an outside expert in the field, to provide extra quality control.
For graduation, require two additional accomplishments. The first is an internship with a company in the field, with satisfactory report from the latter. The second is a practical project to showcase the student’s expertise. Hopefully these two achievements will overlap, as the internship suggests practical projects and refines expertise. However, no internship should hem in a student with novel ideas and thirst to develop them.
The total program should not take more than 9-12 months of full-time study, including internship, or two years of part-time study. Employers so hunger for qualified data scientists that graduates should be able to quickly land good jobs, whether or not they’ve finished university. Judging from experience elsewhere, many internships will likely lead to full-time employment with the same firm.
The prospect of quick training with immediate employment should especially appeal to millennials. University fees have risen far faster than general job prospects for their graduates. Our camp will keep schooling quick, stimulating and fun. Working in our favor, AI firms tend to prize creativity.
Our camp should appeal most to three kinds of students. The first are really bright high school students, eager to make their mark in the world and work with the new AI live forms. The second are experienced programmers looking to retool for Big Data. The third are economists, medical researchers and other scientists looking to improve their data-analytic skills.
A bigger challenge is lining up good coaches. I’m wary of hiring experienced teachers, as they tend to lecture rather than coach. Experienced engineers are likely a better psychological fit, but their technical skills tend to be outdated for what we need. Young AI grads have the technical skills but they’re too scarce and sought after to count on landing. Fortunately, we don’t care much about the rules; we just need to find a few exceptions. If you think you’re an exception, please let me know .
The biggest challenge is to pay good coaches well. That’s a general problem for mentors, as success goes to proteges’ heads; they think they’ve done it on their own. Ideally, camp and coaches should take equity stakes in their students, so they can share in the latter’s upside without infringing their freedoms. However, such contracts aren’t easy to design or enforce.
As a result, good schools have to finance themselves either through high tuition and/or backing from donors. It’s tough to get started and getting tougher. Hopefully, someone who reads this appreciates the potential and has the money to seed it. Perhaps an existing university would like to enter this area. Perhaps an expanding firm would like to delegate some of its training and retraining needs in data analytics.
Let me close on an upbeat note. A well-run data science camp would likely yield more employment gains per buck, and yield them more quickly, than any other initiative we can imagine. It can be a huge stimulus to AI-related business expansion in Bulgaria. It can raise Bulgaria’s international prestige and bring great honor.
By Kent Osband