The Impact of Artificial Intelligence on jobs in the Legal and Software Sectors
In Superintelligence Nick Bostrom examines some of the risks careless development of Artifical Intelligence poses for humanity, including the risk of being converted into paperclips by an artificially intelligent system, totally lacking in common sense, that has been ordered to maximise paperclip production. We are used to dealing with highly intelligent people with little common sense: software developers tend to fall into this category. As a friend once noted of developers he had to deal with “They can understand complex code but not common sense”. Such people are encountered everywhere but tend to end up in IT or politics. The most useful are those brimming with ideas that can be filtered by themselves or others. Cultural, social, and corporate inertia protect the rest of us from these smart fools.
We are not so good at dealing with smart machines that not only lack common sense but are empowered to over ride it. We are especially bad at predicting the consequences of innovations and worst at recognising future negative consequences of our actions, including adoption of advanced technology. This is a result of a cultural dislike of negativity, in the business sphere, a desire to increase profits and externalise costs, and ignorance. Thus it took centuries before the link between smoking and lung cancer was established and the tobacco companies fought the evidence and implications all the way.
Labour to Machinery to Automation to AI
Artificial Intelligence may pose existential risks for life on Earth or indeed for the rest of the universe, at least the part that does not want to be converted into paperclips, but before we reach that stage automation, smart machines and algorithms will massively disrupt society, perhaps causing the death of Capitalism, however it is defined, making work optional by giving everyone enough to live comfortably (paid for by taxes on business) but with work an option to make more for luxuries. Even if this does not happen the nature of work will change from the current model, which derives from the military, 19th century cotton mills and slave plantations: how it will change is not clear.
There seems to be a clear progression from using skilled labour to deskilling work by employing machines then outsourcing work to the cheapest country then automation with the final stage being to deploy some form of Artificial Intelligence to replace the biological components of the often virulent meta organism that is the modern corporation.
Machinery can deskill workers but eliminate a lot of drudgery and transform entire industries: the standardised components such as screws allowed great increase in productivity and would be almost impossible without machines. Similarly outsourcing routine work, to specialists frees up resources for expansion or management bonuses. Currently the developed world seems to be at the stage of moving to automation with robots in Swiss cheese factories turning cheeses 24 hours a day seven days a week on a schedule designed to produce a predictable and consistent quality product, while eliminating the possibility that a slight glitch in the timing could result in a new improved product.
In The Wealth of Nations Adam Smith claimed the desire to increase profit required increased productivity which would lower prices and increase sales. He assumed increased productivity required more workers, which was true both in the agricultural world of his time and in the dark satanic mills of the Industrial Revolution. Ever more sophisticated machinery first deskilled workers then eliminated them: Bringing in the harvest three hundred years ago needed hundreds of workers on a big estate. Today it needs one with a harvester bigger than the houses most workers inhabited in the eighteenth century. Robots may reduce this number to zero.
Corporations don't want human staff
Employers want to reduce costs and they see staff, nominally their greatest asset, as their greatest cost and so want to exploit them to the full, even if they burnout (the cost of finding replacements may mitigate this exploitation, just as the life of slaves improved when abolition of the slave trade made replacing a dead slave much more expensive) and reduce their numbers, ideally to zero. The unfortunate thing is that their biologically based assets need ongoing maintenance in the form of food, clothing and shelter. Replacing them with robots that need only occasional maintenance, ideally from other robos, seems attractive and some CEOs now consider human workers irrelevant and value technology over people since it is easier to calculate the value of a machine than a human.
44 percent of the CEOs surveyed agreed that robotics, automation and AI would reshape the future of many work places by making people "largely irrelevant" .
Automation, smart technology and sophisticated algorithms are the immediate threat but Artifical Intelligence will pose a qualitatively different order of disruption.
Automation is commonly considered to be an advantage for employers and a threat to the livelihoods of low skilled workers. Using drones for delivery eliminates delivery jobs, robot staffed warehouses eliminate the staff running around at an ever increasing pace and in 2016 Amazon opened a retail store where customers could walk in, remove goods and walk out, the goods being charged to their Amazon account once they left the store. The impact of automation on skilled labour is less obvious. Two sectors are looked at here: The Law and IT, both rule bound systems that are about to change radically.
AI and the Legal Sector
The legal profession is perhaps most resistant to new technology: The US Supreme Court does not use email for communication for instance. But lawyers are already using smart technology and considering the impact of AI on the profession. It is accepted that technology will transform the profession, probably eliminating thousands of jobs that can be performed faster by a program. The main effect so far has been to eliminate the tedium of searching thousands of trial records to find precedents and thousands of case documents to establish the facts as a basis either for defence or prosecution . Serendipity, finding a case that is nominally unrelated to the search but can be reatively used by a lawyer, is, however eliminated.
Talwar  is upbeat about the prospects for AI in the Legal sector and indicates a number of ways this could go, but predicts clear evidence of replacement of lawyers with machines in the next three years together with the rise of distributed autonomous organisations, that is software based organisations with no human employees. The latter seems a little optimistic and while it is likely that the layman will trust a computer more than a human, as many do now, the legal technical experts, the equivalent of the finance industry’s quants, will, hopefully be much more sceptical.
A more likely outcome is software that can feed on thousands, if not millions, of case histories and predict the likely outcome of a trial. This would allow defence and prosecution to input various arguments and see how that affects the predicted result and the possible level of confidence.
Kane  and Smithers  think there will still be a place for lawyers. Kane notes that lawyers deal with human nature and represent the bridgehead between the law and their clients, while Smithers argues that while technology will allow clients either to self diagnose or to be able to discuss their case with the lawyer on more even terms there will always be a place for lawyers, since they do more than just dispense the law and AI is not designed for creativity (however that may be defined) or the independent thought needed to plead a case. He also argues that AI is limited in its ability to interpret data. His view is that the legal profession is likely to follow the medical profession where sophisticated software is used for diagnosis. Similarly the legal profession will be able to identify weaknesses in a case more easily and rapidly.
In the legal sector then it seems likely that lawyers will survive but increasingly be limited to areas involving interaction with humans, extracting the information to input into a trial prediction system, building empathy with a client, preparing them for cross examination and so forth with an AI as an assistant that can find information the lawyer can use. This allows an educated guestimate of how IT might progress.
AI and the Software Profession
Every good developer automates their work to some extent: how they do that is a matter of their experience and personality which is is why the software world is infested with swarms of competing standards and most enterprise programming involves gluing together incompatible buggy and insecure frameworks.
Automation has had good results. At one place the author worked release and deployment initially required team members to be on standby overnight. Deployment was eventually largely automated using a smart framework and could be done during office hours. Build frameworks such as Maven allow continuous automated integration and testing of code changes. This has resulted in a developer reluctance to deliver code changes on a Friday but the benefits outweigh this slight productivity hit. The use of smart systems to examine production logs and identify bugs can save many developers eyesight and sanity: in one case such a system was able to identify a production problem as the result of an overheated router and initiate action to move it closer to a window.
AI is however another thing entirely.
Software professionals seem, like the rest of society, to be divided on the topic of AI and Jobs. Other than those who do not want to think about the risks:
Some think AI will eliminate the need for human developers
Some think there will still be a need for programmers
Some think new jobs will be created following the widespread introduction of AI
The outcome is likely to be something unexpected different from any of these. To paraphrase a well know saying it will be an unexpected unexpected, more like a Black Swan Event, perhaps unfolding slowly.
As in the legal sector humans in IT will increasingly be restricted to areas where interaction with humans is needed, for example determining user needs. AI systems will increasingly be able to generate code, from specifications, then entire systems including legal constraints, security constraints, specifying and writing test cases, and design decisions. They will also be able to understand and reverse engineer legacy systems far faster than humans.
A Likely Future
Neural nets are notoriously hard to decipher and the one used by Google Translate seems to have invented its own internal language, one which is, unfortunately, not (yet?) accessible via Google Translate. Another system continually scrambles code in order to foil attackers, which would make debugging something to be left to an AI
AI systems will become increasingly incomprehensible to humans leaving black box evaluation or evaluation by other Artificial Intelligences as the only way to check say safety critical software.
The situation will get worse if AI designs hardware.
The eventual outcome is that humans will either be redundant or unable evaluate what AI is doing.
Within a few years AI systems could be the best developers, leaving programming as a hobby or a palette for creatives, with more and more artists relying on AI assistance to realise their visions while large corporations increasingly become devoid of humans (some would say they are already devoid of humanity but that is a result of culture and a neoliberal form of Capitalism and an entirely different topic) and even the CEO’s job largely, perhaps totally, automated.
The only question is how long this will take. It is likely to be sometime after the programmers born in the 1950s retire and before the current crop of Computer Science graduates retire, assuming that retirement is still a meaningful term and a meaningful possibility. In other words it could happen very rapidly during your lifetime.
Developers worried about their job should realise that it could change from detailed line by line coding, to natural language conversations with the machine and the machine interrogating them as to their needs. Geek bragging rights would vanish, outside the hobby field, and developers would become technical business analysts, understanding user needs, communicating them to the computer and ensuring the result is what the user wanted. The developer will be rather like a priest, conveying prayers to a deity. Users will eventually want to cut out the human in the middle and talk to the machine directly. Bye Bye developers.
This sort of shift has happened before when writing was invented. First the rich employed scribes to read and write for them, then they did both themselves then the rich had secretaries to do the menial labour such as typing and now they do much of this themselves. The pendulum swings. All we can be sure of is that in a decade or so the role of the developer will change. Corporate inertia will slow the pace of change but not stop it. Or it could happen almost overnight as companies install the AI Installer AI which installs the AI selector AI and proposes a number of AI Systems to install.
AI is coming and will change almost everything. Jobs will vanish and new jobs may or may not be created but if employers reduce their staff to zero no one will be able to afford to buy their products until these new jobs arise. And the jobs may arise in the Black economy which may well leave the corporations that eliminated their staff again with no customers. Perhaps the machines will create fake jobs needing about one hour a week doing simple tasks or playing games. We can only hope they will regard the human race as pampered pets or at the worst honoured but unwanted elders. I can imagine two AIs arguing which one is due to take the human for a walk.
Superintelligence: Paths, Dangers, Strategies Hardcover: Nick Bostrom, OUP 2014.