How AIs Will be Used to Staff Startups and Accelerate Innovation
In the future, part of the labour market could be served by cloud services where specific skills are purchased just like we purchase cloud storage today.
A subway network is not normally the place you’d expect to have a major revelation about how artificial intelligence will allow companies to hire virtual staff, but that’s what happened to me last night.
I was waiting for a train at Piccadilly Circus station when the PA system started blaring out something about how industrial action would affect services in the coming days. Apart from the grime, heat and pushing and shoving – routine for subway networks worldwide – the London Underground offers something unique: recurring industrial action. And commuters hate it – with a vengeance. Squared.
As my fellow passengers and I were mulling over how we would work around the next wave of commuting chaos, our train approached the platform with that characteristic and somewhat disconcerting sound of bumping, rattling and screeching.
I took a look at the driver of the train. His expression was one of utter boredom. I’m sure it would be inaccurate to say he seemed half asleep, but he seemed it.
More troubling was that he seemed to project a sense of deep sadness.
In that moment it was clear to me that pay and working conditions were not the reasons for the recurring strikes. The real reason is that the majority of these drivers hate their jobs and would rather do something else entirely. I also saw that machines could free these trapped workers so they could contribute to the economy in ways that machines could not.
Is it really right that the most advanced organism in the known universe – the human being – whose cerebral powers transcend even the most powerful supercomputers are forced to do jobs like driving a subway train, if they don’t want to?
Is driving a subway train really the best way for the economy to use such a valuable and capable resource?
Question: One day, might it seem wasteful, demeaning and even cruel for humans to do jobs that can be done by machines?
Given that self-driving car technology is maturing rapidly, then it seems an almost trivial matter to develop a technological add-on that can remove the need for driver on subway trains – this being a far easier problem.
Google Self-driving Car Technology
Digital representation of a pedestrian (see here)
Indeed, some train vendors already have such technologies, but they are being strongly resisted by workers' unions who fear that the drivers they represent would lose their jobs.
Surely, the most important question is not whether specific types of job can be performed by machines, but what the displaced workers would do?
This is, of course, a highly controversial question, with people like Erik Brynjolfsson and Andrew McAfee (see: The Second Machine Age) believing that, in the end, displaced workers will find better jobs elsewhere. Most people in Silicon Valley think this way.
The main reason why displaced subway drivers will find better jobs elsewhere is that this is what has happened in all previous periods of technological change. And there are compelling arguments that support this view: because technological change has in the past resulted in overall growth and increased total employment then that will happen again this time.
Meanwhile, others think that the unbridled deployment of AI-based technologies in the labour market will lead to massive social unrest. These people foresee a dystopian outcome with a binary segmentation of the labour market between a very small number of intellectual elites – who control the machines – and the rest who eke out a subservient existence in the shadows far, far below.
These people think that the period of technological change we are now entering is fundamentally different: because the creation of human-level intelligence cannot be compared with any prior technological innovation, we cannot expect the pattern of the past to repeat this time.
While this article started out talking about the plight of tube drivers on the London Underground – which number about 3,000 individuals – the core argument applies to about 30% of the entire global workforce, as I estimated in an article in April 2015 (see: Think Carefully: Could A Machine Do Your Job?).
The countries comprising the G20 account for 70% of the global population, and over 85% of global GDP. The sizes of the workforces in the US and UK are about 48% and 39% of the total population respectively. So, if we assume an average employment level of 40% for the G20, then 85% of global economic activity comes from 40% x 4.7 billion workers, or about 1.9 billion people.
If we also assume that in the long term 30% of these workers will lose their jobs to machines (see above article for detail), then this would represent a social and economic upheaval of truly staggering proportions:
- 570 million people will need to find new jobs;
- 25% of all economic activity will be based on 'machine workers’.
Returning to the fate of the 3,000 men and women who drive the trains on the London Underground, it seems pointless to argue that they would somehow retain their jobs during this period of unprecedented upheaval.
While we could argue about whether it will ever be possible for a machine to do the job of a medical doctor it seems clear that machines will eventually completely displace the 3,000 tube drivers who work on the London Underground.
But the scale of upheaval we are talking about will not happen suddenly: it will take decades, and perhaps even a generation.
During this time, artificial intelligence technologies will become increasingly powerful. While some of these AIs will be used to replace human jobs, others will be used by those displaced workers in new jobs and perhaps even to create new businesses.
The current trend is for AIs to be developing semi-autonomously within specific knowledge or skill domains (e.g. self-driving cars, image recognition, language translation, personal assistants).
Google Self-driving Car Technology
Digital representations of local environments (see here)
Critically, while the underlying hardware, software and even algorithms are common, the application of these enabling technologies within the fields of melanoma diagnosis and customer service require completely different sets of training data, radically different domain expertise and result in quite different products that operate in different markets.
It is plausible that we will first see the emergence of an AI who has proven expertise in one domain and which then, following the natural cost reduction that comes from technological improvement, will be able to be deployed more widely and more cheaply.
For example, IBM is working on applying its Watson technology to the problem of correctly diagnosing the presence of skin cancer (see here at 1:19:17). It turns out that very few dermatologists are sufficiently skilled to tell the difference between a benign skin lesion and melanoma. This means that melanoma is either missed or biopsies are taken needlessly.
A special instance of Watson was given 3,000 images of skin lesions and was then told which 200 of these were melanomas. No other information was provided.
During the training phase, the algorithm learned to associate certain colours, textures, edge boundaries and many other non-programmed parameters with melanomas. The system then developed a kind of mathematical understanding of what an image of a melanoma looks like.
IBM claims that the system is now capable of achieving a very high accuracy in the diagnosis of new skin lesions.
This service is currently only available in the US cancer treatment hospital, Memorial Sloan-Kettering Cancer Centre - but that is not where this story will end.
With more data and higher accuracy the product could become the world’s most reliable AI for diagnosing this particular disease. If the service was delivered from the cloud, there is no reason why it could not be accessible by anyone, worldwide: all that would be needed would be for the patient to use their mobile device to take a picture of a suspicious skin lesion and then upload it to the service which would make an accurate assessment - within seconds.
If this development path were followed by other AIs, which are presently developing with their own domains, the result could be a very large number of network-based ‘virtual machines’ which could do highly skilled jobs quickly, accurately and cheaply.
The widespread availability of such a diverse range of cloud-based AIs would mean that part of the labour market could be served by virtual 'machine workers' whose expertise could be purchased, just like we purchase cloud storage today.
While these services would be used to displace workers, the displaced workers could use the very same services to do higher-level jobs and even build new companies – quickly, efficiently and cheaply.
Perhaps in the future when it comes to setting up a new company all the entrepreneur would need would be an idea, some spare time – say a weekend – and $100.
The first task would be to staff the venture – which is today a notoriously time consuming and risky process that requires significant capital.
But in a future world an unlimited number of 'machine workers' might be available at the click of a button. These ‘virtual employees’ would not need sleep, they would not need holidays, they would not get tired, they would not make mistakes (actually the harder they work they better they become). And they could be fired without notice with no penalties.
The job of staffing a new company might involve visiting a staffing intermediary (the equivalent of a recruitment consultant in today’s world). The entrepreneur would then use an online interface to select the skills needed using a menu-based system.
You would need to decide on the skills needed and the level those skills: say $10 a day for a regulatory expert in the field of pensions, or $30 a day for a senior industrial designer engineer skilled in the field of kitchen gadgets etc.
The new company would be up and running in a matter of minutes and you’d then have a team to manage. If you made a mistake, or realised you needed a different skill then you could just go back and make the required changes – in a matter of minutes.
No longer need that regulatory expert? No problem – you can fire it instantly at no cost and hire a marketer instead.
It might by this time even be possible to test market the product in a real time ‘market laboratory’ and so you would know in advance of launch whether there was demand for the product or service.
The point is that the time and capital needed to start new ventures would be minimal – and this would also change the economics of managing labour at larger companies.
With many of today's knowledge jobs becoming commodity services delivered from the cloud, the value would come from seeing opportunities and managing workforces that combined human and machine staff.
If this view of the far future is even close to correct then it suggests a period of dramatic, accelerating economic growth – certainly with room for some of the 570 million displaced workers, including that depressing-looking tube driver I saw last night.
But the obvious enabler will be that the 'at risk' and displaced workers will need the intellectual capacity and willingness to move up a level, and that will require lifelong education and continual retraining and, probably, a very different understanding of the purpose of work.