Chess is a game designed for two players. All the experiments with ever more advanced computers playing either against humans or each other naturally conform to this assumption. But what if the game were changed so that each piece were artificially intelligent (AI), made its own moves, and the decision about which piece on a side were to move next were negotiated among the pieces on that side? What could this tell us about ways we can use distributed AI (DAI) or develop a complex swarm intelligence (SI), and whether and how the “wisdom of crowds” dynamic might apply to groups or teams of AI processes.
The idea presently ventured is not a computerized version of the “Wizard’s chess” in the Harry Potter books and movies, since like real chess, that is two-player game. The only similarity might be that each piece would indeed have its own opinion about its next move. It might be a bit more like a virtual chess version of the sci-fi drama Westworld, in that pieces interact (although not with people) and learn. In any event, the immediate inspiration for this line of thinking actually was sparked by an image (right) of a couple of robot figures on a chessboard* at the just concluded Future Fest 2018 conference in London.
While being quite aware of the advantage of a single mind or computer directing a side in chess, I’ve also become interested (as a non-expert in the field) in how intelligent agents with different though perhaps complementary goals, might interact in a defined environment towards a particular goal. Since so much work has been done with computer/AI chess in the standard 2-player mode, I wondered what might be possible with AI on the level of all the 32 individual pieces on a virtual chessboard.
Autonomous pieces work out the moves
According to the scenario I’m imagining, there is no overseeing player controlling the pieces on a side. Each AI chess piece:
knows the rules of the game and cannot break them
knows the main object of the game
plans its own moves and will normally avoid a move resulting in its being taken
can communicate with all the other pieces on its side (but not the other side)
Another design parameter involves a choice: Either each piece would know no more than the above, or would be given data on how it has (been) moved throughout many actual games. In the latter case, it would have a repertoire of possible moves in various situations to choose (or depart) from, but not a generalized overview of the games.
The communication among pieces on a side prior to a move would be a critical aspect, of course. For the opening move on each side, there are only 10 of the 16 pieces that can move, and each has exactly two (2) choices, for a total of 20 potential moves. Beyond that, the numbers and combinations – and hence complexity – increase significantly. There have been experiments where AI bots have interacted, but this 16 sided interaction would be significantly more complex.
With each piece being able to consider its own possible moves plus not moving on each turn – even given a specific arrangement on the board after each previous move – there is no one obvious decision for the side from the point of view of the individual pieces (except where a piece wants to escape being taken, or if the king is threatened). Unless instructed in the set-up how to arrive at a decision, the pieces would have to develop their own criteria or method for choosing which piece on the side will make the next move. Some protocol for communication would likely be necessary, especially to facilitate human study of the decision process.
Learning from learning chess pieces
If we treat the AI chess pieces as learning programs (like AlphaZero and Leela Chess Zero), then each piece would probably best always be in the same position – such as e2 white pawn or b8 black knight. That would simplify reference to past games if we choose to give pieces that data, and would in any case presumably facilitate learning the role of the piece over many games.
One could also try various experiments such as switching the position of a piece (that e2 white pawn to, say, h2), or putting a veteran piece on a rookie team to study how it affects the team function.
It would be interesting to evaluate how much computing power is needed for each AI piece and whether/how much that varies by type of piece or position. And naturally also what the aggregate of those demands are per side.
As with AI in the 2-player game, one would watch for unexpected outcomes in the 32-player (but still 2-side) game. Would for instance pieces develop the willingness to sacrifice themselves in scenarios that might lead to their side winning?
Although the object of such an effort would not necessarily be to develop a “team” of pieces that could win against accomplished players, it might be useful at some point to play against single players – human or computer – for the AI pieces to learn in different settings, and also to measure the effectiveness of their “teamwork.”
* Drikybot, the creation of Audrick Fausta, dancer and engineer in mechatronics. Image was copied from Twitter. The caption on the tweet that triggered my thinking on this was something like “what are their thoughts?” – unfortunately I was unable to retrieve that specific tweet for this post.
It may not rise to the level of “torture,” but it sure is annoying – the constant resort to a “dangling this” in hyperlinks intended to entice readers to click on them. What comes to mind, especially after observing its use, is the water drip torture – often referred to as “Chinese water torture” or “Spanish water torture” – which consists of a steady drip of water on the forehead.
This . . . , This . . . , This . . . ,
Clickbait can be a little more low-key, or it can be more raw – “weird trick” “jaws dropped” “you won’t believe what happened next” …. If you’ve surfed the web even a little, you’ve probably seen it all. A lot of clickbait just assaults your curiosity.
The “dangling this,” on the other hand, is a kind of shortcut – a crutch word in the web content world, perhaps, or maybe the poor man’s clickbait. But once you notice it, dangling-this clickbait becomes quite irksome, and arguably more insidious in its approach to enticement than the frontal attack of most other clickbait.
“Some clickbait headlines leave out a key element to lure in the reader.”¹
Poor man’s clickbait gives you just part of what might be some legitimately useful information. Rather than try to lead you on through choice of wording (clickbait has a lingo of its own), the tactic is to deliberately omit a central piece of information, which is most easily replaced with a “this…”
Devolution of headline writing under clickocracy
Attention as measured by views, and views as generated by clicks, mean money. So a lot of talent, resources, and effort are put in to getting and keeping people “engaged” in web content. And that has given rise to a whole new set of ways of inciting curiosity.
In a way, clickbait is a descendant of old-fashioned newspaper headline writing. but with some some variations. The headings for articles are composed with a particular style due to space limitations (especially in print) and a desire to catch the reader’s interest. But they invariably give the reader a whole idea of what the story is about. Or at least it used to be that way.
“Headlines need to be accurate, first, and to fairly reflect the theme of the story.”²
Dangling-this clickbait deliberately departs from that practice, and sometimes it’s as simple as a swapping out:
Headline: “Vermont will pay you $10,000 to move there and work from home” (CNBC)
Clickbait: “This state will pay you $10K to move there and work from home” (WAAY)
Worth noting that (re)writing the heading this way is not done to economize space. It’s just a cheap way of getting the reader to open the link to fill in the blank (“which state is that?”).
One newsletter’s many “dangling this’s”
I’ve been subscribed to Ladders‘ newsletter mailings for quite some time – they sometimes have some interesting content – but late last year I began noticing an increasingly clickbaity tone of the newsletter headlines featured in its emails, many of which used the dangling this. Rather than unsubscribing, I left all their emails on my server beginning in January to see if there were a pattern. I’ve extracted all the clickbaitesque uses of “this” in titles over the 5+ ensuing months in the following list (which is sorted alphabetically).³ There are over 40 instances, not counting repeat use, making approximately 2 dangling-this headings per week (drip, drip, …). The list illustrates the extent to which one organization – certainly not the only one⁴ – is relying on poor man’s clickbait to present its content via email and on its website.
A 75-year Harvard study says this is the key to life
Abusive bosses do this weird thing after being mean
Amazon’s Jeff Bezos says this phrase will destroy your career
Doing this in the first five minutes of networking is a game changer
Learning this language could make you some serious cash
New science says this can fix sleep deprivation
No one’s naming kids this anymore because of Amazon
Richard Branson says this is the most important thing he looks for in an employee
Science says this weekend sleep trick can save your life
Sending a job application at this time of day will murder your chances
Stop doing this at work immediately if you want to succeed
Stop making this huge mistake on your resume
Studies say drinking coffee can be good unless you’re doing this
This age is the latest you can start a new career
This algorithm can figure out when you will die
This behavior by Millennial bosses drives some employees crazy
This behavior by your spouse may be sabotaging your career
This city has the most burned out workers in America
This common employment practice may be shutting out older workers
This high-paying job has the worst divorce rate
This is the loneliest profession in America
This is the most stressed state in America
This is the No. 1 reason your interviewer won’t like you
This is the one email mistake that’s unforgivable
This is the profession most likely to cheat on their spouse
This is the quote Jeff Bezos has taped to his refrigerator door
This is the single dirtiest place in the airport — and it’s not the bathroom
This is the surprising thing that has people most stressed at work
This is the top reason employers expect more people to quit this year
This is the unhealthiest job in America
This is what your pre-bed routine says about you
This is why ‘very unattractive’ people earn more money
This job is seeing its salary shrink the fastest
This job listing may make you very angry
This morning routine can save you 20 hours a week
This profession attracts the highest number of psychopaths
This simple life hack will make you much happier
This sleep trick can add years to your life
This state will pay you $10K to move there
This surprising thing about your appearance may be killing your career
This surprising thing is a sign that you have high intelligence
This weird office phrase can reveal more about you than you may realize
The solution to dangling-this clickbait?
Personally, I try never to click on clickbaity headlines – I didn’t open any of the above links for example – so as not to encourage their use. In the rare case the topic seems interesting, one can do a quick websearch to find a link with a more complete and informative title. Treat your clicks as a votes of approval for the approach used.
3. This list was simple to generate: Searched for The Ladders newsletters in Gmail; copied over the list, screen by screen, to one file in my go-to text editor (EmEditor); then, since items in the text were tab delineated, copied all that and plopped it into Excel; then sorted and manually deleted batches of rows to isolate the group of titles; then copied the list back into EmEditor to manually delete extraneous text; then copied it over here and bulletized.
4. Surprisingly, even the respected Smithsonian Magazine resorted to dangling-this clickbait for an article about the town of Liberal in rural southwest Kansas: “This Town In Kansas Has Its Own Unique Accent.”
My original question in 2008 came from looking at trends in digitizing books – notably Google Books – and machine translation (MT). It elicited some interesting responses, including Kirtee Vashi’s mention of an Asia Online project planning to link these two trends.
So while the technologies for digitization and for MT – the two pieces in localizing libraries of information – are established and improving, each has encountered some combination of legal, political , or funding issues limiting their use individually for mass expansion of access to knowledge, as well as their potential use in tandem.
However, could the Norwegian program, announced in 2013, and the project it has with Nigeria, announced earlier this year, introduce a new dynamic, at least for mass digitization? Could and should large national libraries take the lead in this area?
The idea of digitizing libraries has generally been advocated in terms of access to knowledge, without particular reference to the languages in which publications are written. But languages are critical not only for access to knowledge, but also for facilitating scholarship and the interfacing of ways of knowing. Hence the need to associate mass digitization and MT.
There is at least one proposed project mentioning the potential for translation of books that have been digitized – Internet Archive’s initiative to digitize 4 million books (a semifinalist in the MacArthur Foundation’s 100 & Change grant competition).
Any such digital data produced by the Nasjonalbiblioteket, Google Books, Internet Archive, or any other organization could be translated with MT into other languages, with a few caveats (quality of optical character recognition [OCR]; how well resourced a particular language is; and of course the accuracy of the MT). This means that potentially any mass digitization could be mass translated into a large number of languages, given legal cover and sufficient funding.
What about the accuracy of MT, and how useful could mass MT of mass digitization be if there are inaccuracies? These are critical questions for any project to use MT to translate digitizations. Responses could reference, for instance, domain-specific MT, which is generally more accurate than general MT, provided of course that the material matches the domain used. Or perhaps some system for post-editing could be devised.
This is an exciting area that needs more attention and policy support. Books and other production in print can be digitized on a mass scale, making the knowledge in them more widely available. Digitized text can be machine translated into other languages, and the quality of that can be made high enough for use by speakers of the target languages. As much as the printing press revolutionized access to knowledge of that age, so too the potential to digitize and translate what is in print promises another revolution benefiting more people directly.
This is the third of 3 posts on “AI & the self-driving resume”
The previous two posts1 have looked at the current state of the job market and why applications of artificial intelligence (AI) could make it function better for job seekers and recruiters. They have also mentioned aspects of how AI in this context might work, with the ultimate vision being of intelligent agents (IAs),2 working on behalf of job seekers and recruiters, and able to interact with each other.
The beauty of bringing AI more completely into the job market is not just in saving time and facilitating good matches, but the potential to generate new dynamics, from job matches that never would have been found with current methods, to benefits for career planning, staffing strategies, and potentially other areas of the economy.
What are the forms that AI (or specifically, the IAs) will take in the job market, and how (quickly) will they will evolve? At this early stage, there are many possibilities, but as some are developed and others not, the process becomes path dependent, meaning that your future options are conditioned by your current direction. Presently, there is attention to developing AI for recruiters and the intermediary job boards, but nothing I am aware of that would work for the job seeker, who is still using basically late 20th century tools to find jobs, write resumes, and apply for positions. The longer that disparity remains, the more it is likely to grow, and we could have a job market where AI works for hiring organizations, employment agencies, and job boards, but for job seekers only through them.
This post therefore focuses mainly on how AI for the job seeker might be developed, all within the understanding that the eventual system will involve a distribution of AI tools among all actors in the job market.
Self-driving resume, Mark 0.1
The “self-driving resume” in the title of this series comes from the notion that a computer program could write a resume, find a job opening, and deliver the resume (and job application) to the hiring agent. When you get into it, the steps are not so simple, and the technology has not yet been applied in this way[see comment about one example], but the basic idea begins with two proven concepts: a web crawler (incorporating AI advances) and an intelligent program that can author documents (with some human participation, but less than required by the online template-based “resume generator” programs). As the process evolves, and AI can conduct much of the job search and writing tasks unassisted, the self-driving metaphor becomes a bit more apt.
With current technology (as I understand it), separate programs would probably be necessary for each of the two main processes – searching and writing. For the first, web crawlers are not new, although there are naturally efforts to make their searches more intelligent. For the second, programs that write human language are much newer, yet have had some interesting successes in specific authoring scenarios (from emails to novellas). For automating a job search, both would be designed as learning programs, and each would probably have to be “pre-learned” about the situations they would encounter (perhaps with AI chips?).
In the scenario sketched above (dubbed “Self-driving resume, Mark 0.1”), the crawler (1) searches job sites and organization employment pages according to parameters input by its owner, the job seeker. Those criteria would be types or titles of jobs, particular companies, or perhaps an industry in a region. Basically, the kinds of considerations that an individual searching the web would have in mind when they look for a job are the ones that have to be made clear for the crawler to work on. Assuming the crawler finds a match (2), for example a particular job listing, it would need to find and extract relevant data concerning the job description, qualifications, deadlines, contact names, and relevant details about the organization.
In this scenario, the writing program then enters the picture. The crawler would have as part of its function forwarding the data extracted from the job site to the writer in a form the latter can use (3). Perhaps the trickiest part of the scenario is this exchange of information between what are conceived here as two separate though allied processes.
The writing program, for its part, would have a number of tasks. It is possible that we might actually be talking about a suite of programs that would function in tandem, with each one specializing on writing different things – though for purposes of this example I’ll assume it’s a single program. One fundamental function of the writer, unconnected with the functioning of the crawler, would be authoring of the resume. The resume will be written by the writer based on input from its owner and then some kind of iterative process of review involving the owner to arrive at a satisfactory product that can be updated by the writer. In this scenario we also assume that the writing program has at its disposal a range of best practices, templates, keywords and other important input devices to use in construction of a quality resume. This resume then is a resource that then can be updated or are tweaked for particular employment opportunities.
Another set of responsibilities of the writing program, would be to analyze the data (including text) communicated to it by the crawler which, as described above, has just found a job match, and then to compose a draft cover letter. That cover letter probably would be based on some kind of template that has been pre-loaded, perhaps tweaked earlier by the owner earlier to conform with their style. Its draft letter would then be forwarded to the owner (4) with whatever details on the job it has received from the crawler. The owner then can review that letter, edit it, and as necessary consult the company site him or herself. The finalized letter and approval to go ahead are then returned by the owner to the writing program (5), which then can go to the companies job page, log in for the owner, fill out the forms for the job application(6), and attach the approved cover letter and a copy of the resume (7).
The filling out of online forms is a function relating to the writer’s work on the resume and the cover letter, drawing from the same text, to appropriately respond to the usual range of questions that appear on such forms.
All of the above assumes there would be no problems with the web crawler scraping the organizations job pages, or with the writer logging in (some sites block one, the other or both). It also assumes that the organization listing the job is not itself already using smart programs. What happens when job seekers and recruiters are working with intelligent automation?
From self-driving resume to quantum resume?
Moving beyond the rudimentary but still unprecedented Mark 0.1, AI will be enlisted on the part of individual job seekers and each recruiting organization, rather than centrally organized along the lines James Cooke Brown envisaged, or by major intermediary companies. The operant concept is the IA – independent learning programs that will act on behalf of their owners, though what operates for the job seeker and for the recruiter will naturally differ.
On one side, the job seeker’s crawling and writing programs discussed above would be united in a single “self-driving” IA – basically an autonomous learning program, able to complete the full range of tasks involved in searching and applying for a position, including preparing application materials, notably the resume, but also capable of scouting out potential positions at organizations in industries of interest (perhaps by monitoring contracts and investments to know which companies might be hiring). Once we have introduced AI into the job search process, it would be a short step to using the IA as a tool to help career planning.
An entirely new dimension in this phase would be the potential for interacting with other IAs – of recruiters and companies, obviously, but also with IAs of other people in the job market or just out there in case. Imagine sharing information with other applicants (minus name and personal details) regarding comparative qualifications for a job of interest. Or information on companies – hiring practices, workplace issues, salary levels & offers – from each one’s experience. All of a sudden a range of data becomes available from the IA – a significant benefit beyond automating repetitive tasks and extending searches beyond what is humanly possible.
On the other side, as it were, recruiters will also be working with some configuration of IAs (perhaps in a tiered system, partially for security reasons?) to handle job applications, communication with applicants, inquiries outside of job listings, vetting, etc., as well as IAs to seek out potential applicants (more or less reversing the contact and response processes).
The idea is not a more sophisticated information dump from one side and more sophisticated management and analysis process on the other, but rather IAs that can query, respond in kind, and exchange information, and that are capable of learning from the interactions in ways that both improve their function and produce and organize usable data for their owners. All that said, it is important to note that the use of IAs would not eliminate person to person contacts, serving instead to get us to where those contacts are most productive, and indeed giving us more time for them rather than repetitive tasks.
To the extent we begin talking about IAs interacting on their own – albeit with direction from and “ground truthing” with their owners – the dynamics become hard to predict. However there are some things that can be expected:
As IAs of job seekers and recruiters, or job seekers with other job seekers, communicate directly with each other, this will not take take place in human language (if we take recent experiments as an indication of what is to come).
The resume will no longer have a fixed form, except when needed for human reading, and then will change according to the context and the IA’s learning from experience – almost a quantum phenomenon. The database of professional information and job history that goes into the resume will originate from the owner, and be tweaked as appropriate, but the specific selection and organization of information transmitted in each circumstance – or generated into a printed document in the appropriate human language – would likely differ.
There will no longer be a need for companies to tell you they’ll “keep your resume on file,” since recruiters could page your IA (or the IA of anyone, or theoretically everyone) for resumes when they need them.
There will have to be protocols and standards for communication among IAs, including ways to translate their communication to forms we recognize for checking and analysis.
One interesting question is what will be the virtual space in which this interaction of IAs takes place? Would this happen simply over the Internet, or on the servers of particular companies or job boards, or some dedicated “agora” run by an intermediary non-profit organization?
Next steps, first steps
At this point there are three areas to get the process moving towards a Mark 0.1 stage and beyond:
Setting up an experimental crawler that can find jobs and download (“scrape”) relevant information. The idea would be something that can be easily tasked (what to look for) and tweaked (to improve results). There are crawler programs available, but thinking here of something purpose-designed and friendly to non-expert users – something one could run from a desk-top or perhaps a smart phone.
Setting up a program to author resumes based on information given it, perhaps in the form of an existing resume. However, this is an intelligent program, not a fillable template program, so it would be expected to produce a document with minimal input and understand when it needs more for a complete document, and where it can trim information for succinctness (and space limitations). A next step would be to be able to adjust the resume contents in function of input of a job description and requirements. The step after that would be authoring a cover letter in function of the resume material and input of job information. Here too, a priority is user friendliness.
Development of a plan for how to link the two programs. The next step on this, as I see it, would be how to unify them into a single IA.
I am interested in the possibility of this being approached as an open-source project (though am unfortunately not at the level of being able to contribute to the actual development).
A meta-requirement is elaborating the vision of how IAs of job seekers and recruiters would interact. This would be more on the cutting edge of AI development as I understand it.
This is the second of 3 posts on “AI & the self-driving resume”
In a sense, the job market is really a market in resumes, even though a real person – someone seeking a job, a potential employee – is behind each one. Resumes stock databases, they are digitally searched, they may be passed around and printed and looked over. The distribution of resumes engages intermediary organizations, and advice about how to write and share resumes has become an economic activity. And of course, each resume itself represents an investment of time and resources.
Reign of the resume
The resume, as a statement of one’s experience and education, and nowadays often cast as a marketing tool, is perhaps the central element in the current employment system. While some want to trace the history of the resume back to Leonardo da Vinci (see the diagram at right1), it really only became an established part of job applications well into the industrial age. Despite some changes in standard content over the years (e.g., bio info no longer appropriate, the objective statement now “out of style“), and in availability of much better tools for producing and disseminating resumes on the one hand and processing them on the other, the basic concept hasn’t changed at all.
The ability to submit resumes has always run into the problem of whether and how they will be read on the receiving end. I still recall a talk back in the late 1970s in which the speaker suggested a red ribbon as a device to make a resume stand out in the very physical pile of paper before a recruiter hiring for entry level positions. Current availability of advice and tools to improve the content and look of a resume are more sophisticated, but still reflect the same concerns. In place of artifices like a red ribbon, job seekers of all levels these days are advised (1) to layer keywords into their resume or CV (one site lists 155) in order to make them findable in the digital pile, and (2) on the chance the resume is seen by human eyes, to format the content for the few fleeting seconds of attention they may get. For more detail, read the 300-page manual on resume writing offered by The Ladders.
Another factor is the disparity between those fleeting seconds of attention, and the hours spent putting together, updating, and tailoring a resume. It naturally takes much longer to write something worthwhile than it does to read it, but it seems that resumes are no longer even read. One quick estimate is that time spent glancing over a resume may be only 0.0003% to 0.0014% of the time taken to prepare it.
Meanwhile, the same digital technologies that have facilitated producing and editing resumes on the one hand, and online job listings on the other, have fostered a kind of trade in resumes. It is possible to post a resume on a website where, if one is to believe the advertising, it will be “exposed” to hundreds of thousands of employers. ResumeRabbit.com, for instance, claims over 1.5 million (via reposting to 89 other sites). It will however be one of probably millions vying for attention.
Even on a more specialized level, the numbers are incredible: DevelopmentAid.org for example, offers a “CV broadcast” service reaching 30,205 organizations (as of April 2017 – the number varies). In short, the system is flooded with ever more resumes.
Where is all this going?
One could argue that the humble resume is in effect being asked to do much more than it was ever intended to do. No surprise therefore to see radical-sounding prognoses such as the “death” of the resume (for example in 2013, the previously cited 2015 list, and of course Nelson Wang’s 2012 book, The Resume is Dead), perhaps along with its recent counterpart, the applicant tracking system (ATS) .
Discussions of the “death of the resume” propose alternatives such as bios, videos, graphics, or the amorphous online personal brand. One blog post by Charles Handler several years ago suggested that “various elements of a resume are being teased apart and presented in a different format that is based more on profiles and portfolios.” But all these tend to end up with resume-surrogates that still have to be created by job seekers, and then sorted through and processed by recruiters, who for their part resort to the assistance of specialized software.2
In fact, the resume was never “alive” to begin with, given that it is static information on a page(s). Social media presence may be more current, and other resume-surrogates may have their appeal (some of them are quite creative in what they can convey), but in the end, these are all passive presentations of information, regardless of how well-crafted they may be.
Hence one of the questions here: Could artificial intelligence (AI) bring the resume to life, able to interact with a human or machine reader, bring forth relevant information, learn from the interaction, and inform the owner of the resume?
Mass market & upmarket
There is arguably an inherent dysfunctionality to this system, of which resumes are central, as the volume of applicants and positions grows. In the massifying job market, more people can send applications – relatively more easily thanks to technology – to more organizations for more positions. So, hiring departments adapt with more automated ways to screen out digital documents and reduced time for eyes-on review. But even if job seekers “load” their resumes with keywords to get through the automated screening, the software is also “raising the bar,” to screen in more sophisticated ways. An extreme outcome of this “keyword arms race” is the oft-cited case where one company screened 29,000 applications for a single engineering position with not one of them found qualified.
Suggestions per Nelson Wang and others that job seekers take unorthodox strategies – the new “red ribbon” – are only an advantage when a few are using them. How much does anyone stand out if everyone is standing out, and how can recruiters with limited time sort through the cacophony? There are human limits. Hence on the recruiter side, automation and discussion of “AI” to deal with applicant data – which in turn reduce the applicant, or actually their resume, to data.
This massification of the job market is exactly the kind of situation – large and increasing number of actors, large but varying numbers and types of openings, complex quantitative and qualitative data, and waste – where a more intelligent, if not interactive, program or automation could do much better for everyone.
As one moves up the scale to more specialized and executive positions that are relatively fewer in number and higher in pay, the dynamic changes in some respects. Recruiters, sometimes from executive search services (headhunters) or firms specializing in particular professional fields, may reach out to contact prospective hires. Personal contacts developed in earlier work, may prove useful as sources for information and/or references – and indeed important as many positions on higher levels are not advertised (this is another issue that I’ll come back to). But there are still many people chasing mid to higher level positions in the same job market, relying on basically the same methods and advice.
Imperfect information & intermediaries
The job market is one of imperfect information – in several ways. A job seeker can’t know all available positions they qualify for, and even the best research on a potential employer goes only so far. Recruiters must judge candidates based only on resumes first (usually) and whatever other information comes in the application or via referrals, and they often – especially in some industries – are dealing with a subset of potential candidates they’ll never have the chance to consider.
So, another part of the system includes intermediary entities designed in one or another way to help link employers with job seekers in that uncertain environment:
Recruitment firms (“recruiters” may work for specific employers or for one of these firms) and headhunters;
All of the above (except for job agencies run by government or non-profits) are businesses, basically selling services to employers, job seekers, or sometimes both. Like any business, they seek to maximize income, minimize expenses, and provide services that attract (and in some cases, retain) customers. Significant amounts of money and attention to diverse revenue streams (often from both recruiters and job seekers) are involved.
On the job seeker side, for instance, resume writing services run about $100, while job coaches cost clients hundreds or thousands of dollars, with one upper end service costing about $10k (and all of that without any guarantee of results). No figures on what recruiters pay for services in this subsector (in tandem with their in-house recruitment capabilities), but it likely is a lot.
So, we have a system that has grown in all respects, has costly inefficiencies, and is – along with all of us – in the midst of technological change including what we call AI. It is inevitable that some aspects of AI will be brought into the job market more fully, even as other aspects of AI are applied to automate various jobs. We already see the early stages of this, as mentioned in the previous post in this series. The next post will focus on AI, with particular attention to the less often considered subject of what it might do for the job search side of the system, and how it could change interaction between job seekers and recruiters.
1. Image source: Undercover Recruiter. Image credit: Rezscore.com. Click on image to open full size. (Note the 400-year gap in the “500 year evolution.”)
2. Since traditional resume-based approaches on the entry level may miss talent, some employers try to get away from the resume, even using social events or computer games to evaluate applicants. It is not entirely new to have such live-performance evaluations of candidates – the US Department of State has long used an observed role-playing exercise as part of the selection process for Foreign Service Officers.
This is the first of 3 posts on “AI & the self-driving resume”
There’s lots of speculation these days about how robots and artificial intelligence (AI)* might further transform the way we work or drive a new industrial age – along the way maybe even taking our jobs – but not so much on how such advanced technology could change the ways recruiters fill positions and almost nothing how it might help people find employment. In an age where self-driving cars are already on the road, aren’t we due for some breakthroughs in the mechanisms recruiters and job hunters use in the job market?
Maybe you’ve read about the novella “co-written” by AI that fared well in recent a Japanese literature competition? Or other projects using AI to write reports or stories – or even “better emails“? If I’m writing a resume, I’d want something along these lines to fill in all the right details and keywords to the right measure, and optimize it just right for the recruiter’s database search and eventual 6 or 8 (or maybe 30 on a slow day) seconds of attention.
But wait – if I’m a recruiter I might want something along the lines of that Alphago program that beat human champions at the complex game of “go” to analyze candidates’ skills and track record and figure out which ones fit best in which positions, seeing ahead a few moves to future development and retention. So maybe even award-winning resumes won’t be seen by humans?
But then, since we’re talking about AI, maybe the resume writing program and and the recruitment strategy program could somehow just talk to each other? Have them get back to us when they have something!
Getting real … in the age of AI
It’s hard not to speculate about what AI might be able to do to make both the job search and the recruitment process less labor intensive, more efficient, and more productive for all.
In recent years there has been a lot of discussion of how technology trends will affect job hunting and recruitment – for example 6 or 7 ways in 2012, 5 ways in 2013, reiterated in 2014, 10 more in 2014, 4 or 5 or 10 ways in 2015. However, all of these really concern changes in, not of, the system, and don’t venture into how cutting edge technology could transform the job market. Yet.
AI actually does come up in a few discussions of technology for recruitment – for example as early as 2008 and in a later undated article – but more as sophisticated ways of searching resume databases and beyond. A recent article tacks on chat bots and employee onboarding to the tasks AI might do for recruiters, so you can see the direction things are moving in.
There’s not even that much for job seekers, who are for the moment still stuck essentially with late 20th century tools and processes. A recent innovation for job boards claims to use AI to help job hunters find positions. Again an intimation of where things might go, but again not changing the search process, let alone the vehicles used to apply – such as hand-typed resumes.
A system on the verge of transformation?
Could we be on the cusp of a revolution in the job market involving AI, data science, and perhaps virtual reality that could aid both the person looking for employment and the company doing hiring? One that would enable a lot of the time-consuming processes involved in searching and application on the one hand, and review and screening on the other to be done more quickly and systematically, with less human error, and on a wider scale than is possible with current methods?
This is not to suggest that the whole hiring process should or could be completed without human input, but rather that many of the initial (often repetitive) tasks on both sides could be more efficiently and productively handled by intelligent automation – processes that ultimately will be able to interact. And that that human input would then come in where it is most valuable, and indeed essential: input of information and parameters at the outset, and at key decision points later on.
If self-driving cars can provide personal transportation that is potentially safer and more efficient, couldn’t there be analogous advances in how jobs are found and filled?
On the job seeker side, consider how much time goes into searching listings, filling out applications, crafting resumes, writing cover letters, and maintaining a professional social media presence. And along the way keeping up with all the advice and evolving thinking on approaches and techniques to do all of that better, which has become an industry in itself (looking for a job? here’s a list of 9 more books you should read, according to Business Insider).
Altogether this represents a lot of repetitive work, most of which has no return (one recent article on “1200 job rejections” illustrates the problem). What if much of this could be automated, intelligently? Could AI, informed and tasked by a person, search out particular kinds of openings (roles, companies, locations), write and tailor presentation of professional information, and generate job applications? As a learning (recursive) program it would be designed to adapt, but a key aspect would have to be iteration and course correction with the person it is representing.
On the recruiter side, consider how many applications are received for each position, with the numbers growing each year. Incoming applications have to be processed in limited time, or stored in an applicant tracking system (ATS) and accessed as data (think keyword searches of masses of resumes). And then there’s screening and checking references. What if incoming applications could be analyzed and queried, with preliminary background and reference checks as appropriate, to yield a short list? What if the ATS was smart enough to do all that?
Then there is the question of how AI in the service of job seekers and AI in the service of recruiters would work together. One would imagine the need for a system of protocols and a conceptualization of the virtual space in which they would interact, as intelligent agents (IAs) in a multi-agent system. That could in turn lead to radical changes not only in how jobs are found and filled – for one thing, IAs operating quickly and across cyberspace could find matches that people could not – but perhaps also in the ways careers and staffing are approached. One could be permanently on the market in the sense that the IA could regularly check new employment opportunities or potential new hires.
And perhaps lead also to novel outcomes as IAs go to work: What if for example, IAs of job seekers who have never met could compare notes about an organization or salary offers, or IAs of recruiters could share data on a particular candidate?
Reimagining the job market
The idea of technology fundamentally changing the ways people find employment and organizations find employees – as opposed to enhancing or modifying parts of the current system – is not new. Perhaps the most extensive effort to think through the possibilities is that of James Cooke Brown in his 2001 book, The Job Market of the Future: Using Computers to Humanize Economies. But while Dr. Brown’s work considers how to reorder the economy and thus the nature of employment, aided by computer technology, I’m wondering if the sequence will actually be reversed. That is, that application of advanced information technologies – specifically AI – in the service of individual employers and especially individual potential employees might in turn change the nature of those relationships and how careers are built.
The job market as we know it is basically an evolution of communication between people needing paid work and people needing help with getting something done. And the current system of web-based applications and screening of digital resumes is the contemporary version of a set of tools, roles, and modes of communication developed in recent history as economies became more complex: job ads; employment agencies and websites; applications with resumes and cover letters; recruitment/hiring as a specialization within the field of human resource management; and processes of review and selection.
Will intelligent technology rewrite this sequence and the elements involved? I’m suggesting it will, and that it will be a good thing. In the following posts in this series, I’ll explore aspects of this question:
* AI can refer to a range of capabilities and processes. For a basic introduction, see the articles in Encyclopaedia Britannica and/or Wikipedia. In referring to AI in this blog post, I am assuming some variable sub-set of that range.
… on agriculture, NRM, education, ICT, language, development & policy